[{"id": "a1-001", "domain": 1, "task_id": "1.3", "objective": "Iterate prompts to improve output quality", "situation": "A marketing coordinator's first attempt at a campaign brief comes back disappointing. She wants to improve the result on her next attempt.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Rewrite her request at much greater length, since more description generally produces a more considered answer.", "correct": false, "explanation": "Length is not the variable. A longer version of an unclear request is still unclear, and padding can bury the one instruction that mattered."}, {"letter": "B", "text": "Send the same request again, on the basis that a second attempt often produces a noticeably better result.", "correct": false, "explanation": "Resending an unchanged request repeats the conditions that produced the weak answer, so any improvement is chance rather than a fix."}, {"letter": "C", "text": "Work out what is actually wrong with the answer, since a vague ask and missing material need different fixes.", "correct": true, "explanation": "The causes call for opposite responses: a vague ask needs a tighter specification, missing source material needs the material supplied, and a wrong task needs restating. Changing something before knowing which it is means guessing."}, {"letter": "D", "text": "Break the brief into four smaller requests, so that each part of the campaign is handled on its own terms.", "correct": false, "explanation": "Splitting up helps when a request is genuinely doing too much, but applied blindly it produces four requests carrying the same original flaw."}], "correct": "C", "select": 1}, {"id": "a1-002", "domain": 1, "task_id": "1.3", "objective": "Iterate prompts to improve output quality", "situation": "An events officer changed four things about her request at once and the result improved. She now wants to keep improving it over the coming weeks.", "question": "What is the drawback of working this way?", "options": [{"letter": "A", "text": "She cannot tell which of the four changes helped, so she has learned nothing she can rely on next time.", "correct": true, "explanation": "Bundled changes produce a result but no knowledge. Two of the four may have helped, one may have made no difference and one may have hurt, and she carries all four forward without knowing which is which."}, {"letter": "B", "text": "The four changes together make the request considerably longer than the version she originally started from.", "correct": false, "explanation": "A longer request is a side effect rather than the problem, and length is acceptable when every part of it is earning its place."}, {"letter": "C", "text": "Changing four things at once takes more of her time than changing one thing and reviewing the result.", "correct": false, "explanation": "Changing several things at once is quicker in the moment, so effort is not what she loses by doing it."}, {"letter": "D", "text": "The improvement may not last, because results vary from one attempt to the next even with identical wording.", "correct": false, "explanation": "Some variation between attempts is normal, but that is a reason to check a change more than once rather than an argument against improving the request."}], "correct": "A", "select": 1}, {"id": "a1-003", "domain": 1, "task_id": "1.3", "objective": "Iterate prompts to improve output quality", "situation": "A project administrator's summary of a status report keeps omitting the risks section. Her third attempt reads: 'IMPORTANT: you MUST include the risks. This is CRITICAL.' The risks are still thin.", "question": "What is a better next change?", "options": [{"letter": "A", "text": "Ask for the summary and then ask, in a follow-up, whether anything important was left out of the first answer.", "correct": false, "explanation": "A follow-up question can surface omissions, but it leaves the original request unimproved and repeats the work every week."}, {"letter": "B", "text": "Move the emphasised instruction to the very top of the request, so it is the first thing that gets read.", "correct": false, "explanation": "Position helps a little with genuinely long requests, but an unspecific instruction at the top is still unspecific."}, {"letter": "C", "text": "State what the risks section must contain, such as each open risk, its owner and the date it was last reviewed.", "correct": true, "explanation": "Emphasis conveys that something matters, not what it should contain. Naming the fields the section must carry turns an insistence into a specification that can actually be followed and checked."}, {"letter": "D", "text": "Add the same emphasised instruction a second time at the very end, so the requirement appears twice in the request.", "correct": false, "explanation": "Repeating an instruction verbatim adds volume rather than information, and the second copy is as vague as the first."}], "correct": "C", "select": 1}, {"id": "a1-004", "domain": 1, "task_id": "1.3", "objective": "Iterate prompts to improve output quality", "situation": "After five rounds of corrections a policy draft keeps regaining problems that were already fixed, and the chat has become hard to follow. Which two steps should the author take?", "question": "Select two.", "options": [{"letter": "A", "text": "State plainly which parts of the current draft are settled and must be left untouched.", "correct": true, "explanation": "Protecting the settled sections is a distinct move from restating requirements: it constrains what may change, which is what stops regressions."}, {"letter": "B", "text": "Ask Claude to look back through the chat and remember all corrections made so far.", "correct": false, "explanation": "Asking for recall does not resolve corrections that conflict with each other, and it leaves the requirements spread across many turns rather than stated once."}, {"letter": "C", "text": "Start again with one prompt gathering every agreed requirement in one place.", "correct": true, "explanation": "Consolidating scattered corrections into one clean specification removes the contradictions that accumulate across a long thread of piecemeal edits."}, {"letter": "D", "text": "Carry on with more rounds, since each round has been fixing at least one real issue.", "correct": false, "explanation": "Fixing one thing per round while breaking another is not progress. Continuing the same loop simply repeats the pattern that produced the problem."}, {"letter": "E", "text": "Change to a different topic briefly and then return, to give the draft a fresh start.", "correct": false, "explanation": "A change of subject does not reset anything about the draft. It adds more material to an already cluttered conversation without addressing the cause."}], "correct": ["A", "C"], "select": 2}, {"id": "a1-005", "domain": 1, "task_id": "1.2", "objective": "Apply task decomposition techniques to structure complex requests", "situation": "A support team wants to build a comprehensive FAQ from a large set of historical customer tickets.", "question": "Which decomposition technique is most suitable?", "options": [{"letter": "A", "text": "Employ a third-party FAQ generator instead of using Claude.", "correct": false, "explanation": "Employing a third-party generator avoids the decomposition challenge rather than addressing it, and does not use Claude. The question requires a technique that structures the task for Claude."}, {"letter": "B", "text": "Cluster tickets by topic, generate FAQs per cluster, then merge.", "correct": true, "explanation": "Clustering tickets by topic breaks the large set into manageable sub-problems, enabling focused FAQ generation per cluster. Merging these ensures comprehensive coverage and a well-organized final document."}, {"letter": "C", "text": "Ask Claude to summarize each ticket, then combine into an FAQ.", "correct": false, "explanation": "While summarizing each ticket is a useful preprocessing step, it does not group related issues, so combining summaries still lacks the topic-based organization needed for a coherent FAQ."}, {"letter": "D", "text": "Concatenate all tickets and ask Claude to produce the FAQ.", "correct": false, "explanation": "Concatenating all tickets and asking Claude to produce the FAQ risks exceeding context limits and lacks structured input, making it difficult to identify recurring themes and organize answers coherently."}], "correct": "B", "select": 1}, {"id": "a1-006", "domain": 1, "task_id": "1.4", "objective": "Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)", "situation": "A business strategist needs to compare three competing products on features, pricing, and market positioning.", "question": "Which prompt will yield the most useful structured comparison?", "options": [{"letter": "A", "text": "Prompt Claude to compare the three given products, letting the model decide which attributes to cover and how to structure the output, without indicating a preferred format such as a table, bullet points, or paragraph text.", "correct": false, "explanation": "Leaving the attributes and output format entirely up to the model often results in an uneven or incomplete comparison. Without specifying a preferred structure, the output may lack consistency and be difficult to parse for actionable insights."}, {"letter": "B", "text": "Instruct Claude to describe each product in detail covering available features, benefits, and market positioning, then compile a detailed list of similarities and differences drawn from those descriptions.", "correct": false, "explanation": "This prompt asks for detailed descriptions and a list of similarities and differences, but it omits pricing, a key comparison attribute, and does not enforce a structured table. The resulting list may be less precise and harder to interpret quickly than a side-by-side comparison."}, {"letter": "C", "text": "Guide Claude to gather details on features, pricing, and market positioning for each product, then create a comparison table with rows for each product and columns for features, price, positioning, and USPs.", "correct": true, "explanation": "This prompt explicitly defines the key attributes (features, pricing, market positioning) and requests a comparison table with rows for products and columns for features, price, positioning, and USPs. This ensures a structured, consistent, and actionable output that facilitates quick comparison."}, {"letter": "D", "text": "Ask Claude to write an essay comparing the three products, with sections on their development history, market approach, and competitive evolution, rather than providing a point-by-point feature comparison.", "correct": false, "explanation": "An essay format emphasizes narrative over clear, attribute-by-attribute analysis and focuses on development history, market approach, and competitive evolution rather than the requested features, pricing, and positioning. This makes it less suitable for deriving structured business insights."}], "correct": "C", "select": 1}, {"id": "a1-007", "domain": 1, "task_id": "1.3", "objective": "Iterate prompts to improve output quality", "situation": "A finance assistant's variance commentary reads well, but several of the figures quoted in it are wrong. She is about to start rewording her request.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Check the figures she supplied, since a commentary can only be as accurate as the numbers it was given.", "correct": true, "explanation": "When the reasoning and the writing are sound but the numbers are wrong, the input is the first suspect. Rewording the request cannot repair figures that were incorrect before they were pasted in."}, {"letter": "B", "text": "Ask for the commentary to be produced again, this time with each figure checked as it is written out.", "correct": false, "explanation": "A self-check re-reads the same numbers she supplied, so wrong inputs are confirmed rather than corrected."}, {"letter": "C", "text": "Ask for every figure in the commentary to be listed separately, so she can check them all in one place.", "correct": false, "explanation": "A separate list makes checking easier, but it is a step she should take after establishing whether the input data was right."}, {"letter": "D", "text": "Add an instruction that accuracy of the figures matters more than the readability of the commentary.", "correct": false, "explanation": "Instructions about priorities do not correct an underlying number, and they invite more careful presentation of the same wrong values."}], "correct": "A", "select": 1}, {"id": "a1-008", "domain": 1, "task_id": "1.2", "objective": "Apply task decomposition techniques to structure complex requests", "situation": "A training manager has split a course redesign into four pieces: rewriting the learning objectives, updating the slide deck, refreshing the exercises, and drafting the end-of-course quiz.", "question": "Which observation should shape the order?", "options": [{"letter": "A", "text": "The quiz is what learners remember most clearly, so drafting it first would set the standard for the rest of the course.", "correct": false, "explanation": "Assessment written before the objectives ends up defining them by accident, so the course is shaped by whatever the quiz happened to test."}, {"letter": "B", "text": "The slide deck is the largest piece, so starting there gets the greatest share of the work out of the way early.", "correct": false, "explanation": "Size is a workload observation. Building slides before the objectives are agreed means rebuilding them when the objectives change."}, {"letter": "C", "text": "The objectives govern the other three pieces, so they are settled first while those three can then run in any order.", "correct": true, "explanation": "Only one real dependency exists here: slides, exercises and quiz all serve the objectives. Once the objectives are fixed, the remaining three are independent of each other and their order is a matter of convenience."}, {"letter": "D", "text": "The exercises take the longest to review by others, so they should be finished ahead of the remaining three pieces.", "correct": false, "explanation": "Review time is a scheduling consideration, and exercises built against objectives that later change will need reworking anyway."}], "correct": "C", "select": 1}, {"id": "a1-009", "domain": 1, "task_id": "1.1", "objective": "Create effective prompts for business and technical tasks", "situation": "A marketing analyst pastes six months of customer survey comments and asks Claude to summarise the main complaints. She intends to present the result to her director as the complete picture.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The summary will be organised by month rather than by theme, making the leading complaints harder to compare.", "correct": false, "explanation": "Ordering is a presentation issue that a follow-up request fixes in seconds, and it does not mislead anyone about coverage."}, {"letter": "B", "text": "The comments will all be rewritten in fresh wording, so her director never sees any customer's verbatim phrasing.", "correct": false, "explanation": "Losing verbatim quotes weakens the presentation, but she can ask for examples. It does not change what the director believes about the shape of the complaint data."}, {"letter": "C", "text": "A few prominent themes may be reported as though they were everything, hiding complaints that recur less often.", "correct": true, "explanation": "She asked for the main complaints, not for complete coverage, and she plans to present the answer as complete. A low-frequency but serious complaint can be omitted entirely while the summary still reads as authoritative, so the request should state that coverage must be exhaustive."}, {"letter": "D", "text": "Claude will ask which reporting period to cover instead of summarising, delaying the work until she answers.", "correct": false, "explanation": "A clarifying question costs a moment and is easily answered, so it is an inconvenience rather than the risk that matters here."}], "correct": "C", "select": 1}, {"id": "a1-010", "domain": 1, "task_id": "1.2", "objective": "Apply task decomposition techniques to structure complex requests", "situation": "A grants manager must choose between eleven internal funding applications and plans to work through them in stages with Claude.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Sort the applications by the amount requested, so the largest commitments receive the closest attention first.", "correct": false, "explanation": "Amount requested is one relevant factor among several, and using it to allocate attention first can push a small, strong application aside."}, {"letter": "B", "text": "Ask which of the eleven applications look strongest, and use that shortlist to focus the more detailed stages.", "correct": false, "explanation": "Asking which look strongest outsources the judgement she owns, and it does so before any shared standard exists to justify the ranking."}, {"letter": "C", "text": "Agree what a fundable application must demonstrate, since every later stage compares applications on that basis.", "correct": true, "explanation": "The funding standard defines what every later stage looks for, so summaries, comparisons and the final decision all inherit it. Set afterwards, it simply ratifies whatever the earlier stages happened to surface."}, {"letter": "D", "text": "Ask for a summary of each of the eleven applications, so she has a manageable overview before deciding anything.", "correct": false, "explanation": "Summaries produced before the standard is set capture whatever each application emphasised about itself, which is not a comparable basis for choosing."}], "correct": "C", "select": 1}, {"id": "a1-011", "domain": 1, "task_id": "1.4", "objective": "Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)", "situation": "A team wants to generate innovative safety features for autonomous vehicles by first brainstorming ways the vehicle could fail or cause accidents, then inverting those to create safety solutions.", "question": "Which prompt best implements this 'reverse brainstorming' technique?", "options": [{"letter": "A", "text": "Instruct Claude to create a comprehensive list of safety features for autonomous vehicles, drawing from common industry practices, without first systematically identifying specific failure modes or accident scenarios that could be inverted into solutions.", "correct": false, "explanation": "This prompt skips the essential first step of the reverse brainstorming technique, which is to systematically identify failure modes or accident scenarios that could be inverted into solutions. Instead, it directly asks for safety features based on common industry practices, missing the opportunity to generate novel innovations from identified risks."}, {"letter": "B", "text": "Ask Claude to think of potential ways autonomous vehicles could fail and then suggest general safety fixes, without mapping each failure to a specific inverted feature, thereby generating broad advice rather than traceable innovations.", "correct": false, "explanation": "Although it considers potential failures, it only requests general safety fixes without mapping each failure to a specific inverted feature. This approach yields broad advice rather than the precise, traceable innovations that reverse brainstorming requires."}, {"letter": "C", "text": "Direct Claude to examine available accident data for autonomous vehicles and suggest safety features that target the most common accident types, without first brainstorming potential failure modes that could be inverted into solutions.", "correct": false, "explanation": "This prompt relies on existing accident data and targets only the most common types, which is an analytical approach rather than the creative process of reverse brainstorming. It skips the crucial first step of brainstorming potential failure modes that can be inverted into solutions, missing the technique's core."}, {"letter": "D", "text": "Guide Claude to first identify potential failure modes and accident scenarios for autonomous vehicles, then for each, design a safety feature that directly addresses that risk, ensuring each solution is traceable to a specific problem.", "correct": true, "explanation": "This prompt follows the reverse brainstorming technique by first identifying failure modes and accident scenarios, then for each, designing a safety feature that directly addresses that risk. It ensures each solution is traceable to a specific problem, fostering targeted innovation."}], "correct": "D", "select": 1}, {"id": "a1-012", "domain": 1, "task_id": "1.4", "objective": "Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)", "situation": "A researcher asks Claude to answer a factual question that could be verified from several independent sources before being treated as reliable.", "question": "Which instruction best supports getting a trustworthy research answer?", "options": [{"letter": "A", "text": "Ask Claude to leave out which sources it consulted so the answer stays concise", "correct": false, "explanation": "Hiding which sources were consulted removes transparency and makes it harder to judge how well the claim was actually verified."}, {"letter": "B", "text": "Ask Claude to answer only from the first source it finds so the reply comes back fastest", "correct": false, "explanation": "Relying on the first source found skips the verification step needed to establish that a research finding is reliable."}, {"letter": "C", "text": "Ask Claude to cross-check the claim against several sources before calling it settled", "correct": true, "explanation": "Encouraging source verification, checking a claim across multiple sources, is a recommended practice for getting reliable results on research tasks."}, {"letter": "D", "text": "Ask Claude to state the answer confidently no matter how many sources were checked", "correct": false, "explanation": "Confident phrasing unrelated to actual verification can misrepresent how well-supported the claim really is."}], "correct": "C", "select": 1}, {"id": "a1-013", "domain": 1, "task_id": "1.3", "objective": "Iterate prompts to improve output quality", "situation": "A press officer is unhappy with a draft announcement and types back: 'This isn't working, make it better.' The next draft is different but no closer to what she wants.", "question": "What would make her feedback more effective?", "options": [{"letter": "A", "text": "Describing the reaction she wants readers to have, and letting the wording follow from that intended reaction.", "correct": false, "explanation": "Describing an intended reaction is a step up from 'make it better', but it remains an outcome rather than a defect she can point to."}, {"letter": "B", "text": "Pointing at the specific parts that miss, such as an opening that buries the change readers must act on.", "correct": true, "explanation": "Useful feedback identifies what is wrong with this draft and where. Naming the passage and the defect gives a target, whereas a general verdict leaves the next attempt to guess in a different direction."}, {"letter": "C", "text": "Asking for several fresh drafts in different styles, so she can identify which of them comes closest to it.", "correct": false, "explanation": "Generating variety substitutes volume for direction, and she still has to articulate what is wrong once she has picked one."}, {"letter": "D", "text": "Explaining her general preferences in announcements, such as a warm tone and short, uncomplicated sentences.", "correct": false, "explanation": "Stated preferences help across many drafts, but they do not say what this draft got wrong, so the same mistake can recur inside a warm, short-sentenced version."}], "correct": "B", "select": 1}, {"id": "a1-014", "domain": 1, "task_id": "1.2", "objective": "Apply task decomposition techniques to structure complex requests", "situation": "A market research team has Claude summarise findings from six regional reports and now wants a final recommendation. Which two things should they do before asking for that recommendation?", "question": "Select two.", "options": [{"letter": "A", "text": "Agree the criteria the recommendation will be judged on, such as cost, risk and timing.", "correct": true, "explanation": "This is an independent point about specification rather than verification: without stated criteria, the recommendation cannot be argued for or assessed."}, {"letter": "B", "text": "Check the summarised findings against the reports so the next step builds on solid ground.", "correct": true, "explanation": "The recommendation inherits every error in the summary stage. Verifying the intermediate output is what makes decomposition safer than one long prompt."}, {"letter": "C", "text": "Paste all six original reports back in again so nothing at all can have been left out.", "correct": false, "explanation": "Re-pasting the full reports discards the work already done and pushes the team back to the single overloaded request that decomposition was meant to avoid."}, {"letter": "D", "text": "Ask for three alternative recommendations at once and pick whichever one reads strongest.", "correct": false, "explanation": "Choosing on how a draft reads substitutes fluency for judgment, and none of the three options has been measured against anything the team actually cares about."}, {"letter": "E", "text": "Move to a faster response setting, since the remaining step is short and mostly writing.", "correct": false, "explanation": "The final step is the judgment-heavy one, so speed is the wrong thing to optimise for here, and it does nothing about unverified inputs."}], "correct": ["A", "B"], "select": 2}, {"id": "a1-015", "domain": 1, "task_id": "1.1", "objective": "Create effective prompts for business and technical tasks", "situation": "A prompt for an AI physician's assistant includes patient symptom notes, medical records, and appointment history, and Claude's diagnostic summaries sometimes reference irrelevant details buried in the documents.", "question": "What technique reduces this noise?", "options": [{"letter": "A", "text": "Ask Claude to first quote the relevant passages in a dedicated tag before writing its summary", "correct": true, "explanation": "Asking Claude to quote relevant parts of long documents before carrying out the task grounds the response in the source material and helps it cut through noise from irrelevant sections."}, {"letter": "B", "text": "Tell Claude to answer the question first and read the documents afterward to check it", "correct": false, "explanation": "Reading the documents only after answering means Claude's initial answer isn't grounded in the source material at all, which is the opposite of the intended fix."}, {"letter": "C", "text": "Instruct Claude to skip the patient records entirely and rely on general medical knowledge", "correct": false, "explanation": "Ignoring the patient records removes the exact source information needed for an accurate, grounded diagnosis and would make the task worse, not better."}, {"letter": "D", "text": "Request that Claude summarize each document in one word before generating a diagnosis", "correct": false, "explanation": "A one-word summary would discard nearly all of the clinically relevant detail rather than helping Claude focus only on what matters."}], "correct": "A", "select": 1}, {"id": "a1-016", "domain": 1, "task_id": "1.1", "objective": "Create effective prompts for business and technical tasks", "situation": "A technical writer's system prompt says \"NEVER use ellipses\", but Claude occasionally still uses them in unusual contexts. The real reason for the rule is that Claude's responses are read aloud by a text-to-speech engine, and the writer has not told Claude that. The writer wants Claude to generalize the rule reliably.", "question": "What should they add to the instruction?", "options": [{"letter": "A", "text": "Explain that the text-to-speech engine reading the responses aloud cannot pronounce ellipses", "correct": true, "explanation": "Anthropic's prompting guidance says that providing the context or motivation behind an instruction helps Claude understand the goal, and its own example pairs a bare ban on ellipses with the reason that a text-to-speech engine cannot pronounce them. Knowing the reason lets Claude generalize the rule to cases the writer did not anticipate."}, {"letter": "B", "text": "Add several more exclamation marks to the instruction to emphasize how important the rule is", "correct": false, "explanation": "Adding emphasis through extra punctuation or capitalization does not tell Claude why the rule exists, so it does not help Claude generalize to edge cases the writer never listed."}, {"letter": "C", "text": "Replace the instruction with a much longer list of every punctuation mark Claude should avoid", "correct": false, "explanation": "Listing more banned punctuation marks addresses symptoms rather than the underlying reason, and still would not cover every unanticipated case."}, {"letter": "D", "text": "Ask Claude to run a careful spell-check pass after the response to catch any stray ellipses", "correct": false, "explanation": "A separate post-generation check adds an extra step but does not address why Claude keeps producing ellipses in the first place, because Claude still does not know the purpose of the rule."}], "correct": "A", "select": 1}, {"id": "a1-017", "domain": 1, "task_id": "1.2", "objective": "Apply task decomposition techniques to structure complex requests", "situation": "A communications manager needs a twelve-page annual review written. She wants to catch problems early rather than after a full draft exists.", "question": "What should she ask for as the first stage?", "options": [{"letter": "A", "text": "A section-by-section outline naming what each section covers, which she can correct before any prose is written.", "correct": true, "explanation": "An outline is the cheapest artefact that still exposes the decisions that matter: coverage, emphasis and order. Correcting it takes minutes, whereas the same corrections applied to finished prose mean rewriting most of it."}, {"letter": "B", "text": "A list of every source document that the review will need, so that she can gather all of them before writing begins.", "correct": false, "explanation": "A source list is a useful piece of preparation, but it says nothing about what the review will argue or how it will be organised."}, {"letter": "C", "text": "The opening two pages written in full, so she can judge the tone and standard the rest will be written to.", "correct": false, "explanation": "Finished opening pages test tone but not structure, and the review's real risk is that the wrong things are covered in the wrong proportion."}, {"letter": "D", "text": "A full first draft written quickly, on the understanding that she will mark it up heavily on her first read.", "correct": false, "explanation": "Marking up a complete draft is precisely the expensive rework she is trying to avoid, since structural changes at that point mean rewriting whole sections."}], "correct": "A", "select": 1}, {"id": "a1-018", "domain": 1, "task_id": "1.1", "objective": "Create effective prompts for business and technical tasks", "situation": "A teacher has written end-of-year report comments for every pupil in her class. The school's rule is that no comment may be pasted into Claude until the school office has confirmed that it is cleared for sharing with parents. She wants Claude to turn the cleared comments into short parent-friendly summaries, one pupil at a time, and the office wants every summary written at the same reading level and length.", "question": "Which sequence should she follow when putting the request together?", "options": [{"letter": "A", "text": "Confirm which comments are cleared for sharing; paste the cleared comments, by pupil; ask for one summary per pupil, leaving the reading level and length for Claude to choose.", "correct": false, "explanation": "The clearance check comes first and only cleared comments are pasted, but the request leaves the reading level and length for Claude to choose. The office wants every summary at the same reading level and length, and Claude's prompting guidance is to be specific about the desired output format and constraints, so those requirements have to be stated in the request."}, {"letter": "B", "text": "Confirm which comments are cleared for sharing; paste the cleared comments in one block, without pupil names; ask for one combined summary of the whole class, stating the reading level and length.", "correct": false, "explanation": "The clearance check comes first, but pasting the comments in one block without pupil names throws away the pupil-by-pupil structure, and a single combined summary of the whole class is not the one summary per pupil the teacher wants. Stating the reading level and length does not repair a request for the wrong deliverable."}, {"letter": "C", "text": "Paste every comment, by pupil; confirm which comments are cleared for sharing; ask for one summary per pupil of the cleared comments only, stating the reading level and length.", "correct": false, "explanation": "Every comment, including any the office has not cleared, is pasted before the clearance check, so the school's rule is already broken. Asking Claude to use only the cleared comments afterwards does not undo that, because the rule governs what may be pasted at all."}, {"letter": "D", "text": "Confirm which comments are cleared for sharing; paste only the cleared comments, by pupil; ask for one summary per pupil, stating the reading level and length you want.", "correct": true, "explanation": "The school's rule says nothing is pasted until the office has confirmed it is cleared, so the check comes first and only the cleared comments go in. They are organized by pupil so each summary can be tied to one pupil, and the request asks for one summary per pupil while stating the reading level and length, which gives Claude the specific output constraints the office wants. Each step follows from the rule or the task in the scenario."}], "correct": "D", "select": 1}, {"id": "a1-019", "domain": 1, "task_id": "1.3", "objective": "Iterate prompts to improve output quality", "situation": "A policy analyst asked for an assessment of a proposed change and received four short paragraphs where she expected several pages of reasoning.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Ask for the same assessment again and request that this time it should be considerably more detailed.", "correct": false, "explanation": "More detailed is the same instruction the first attempt already implied, so it usually returns the same shape with a few extra sentences."}, {"letter": "B", "text": "Ask what else could be said about this proposal, and add whatever comes back to the assessment she has.", "correct": false, "explanation": "Asking what else could be said produces additional material of unknown relevance, bolted onto an assessment that is still not aimed at her readers."}, {"letter": "C", "text": "Ask for each of the four short paragraphs to be expanded in turn, then join those expansions into one assessment.", "correct": false, "explanation": "Expanding each paragraph inflates the existing structure, which cannot introduce the reasoning the original assessment never attempted."}, {"letter": "D", "text": "State the depth she needs: who reads it, roughly how long it should run, and which questions it must settle.", "correct": true, "explanation": "Shallow output is usually an unstated expectation about depth. Naming the reader, the approximate length and the questions that must be settled converts a vague wish for more into a specification that can be met."}], "correct": "D", "select": 1}, {"id": "a1-020", "domain": 1, "task_id": "1.2", "objective": "Apply task decomposition techniques to structure complex requests", "situation": "A policy officer must work through a 180-page regulatory guidance document that is too long to handle usefully in one request.", "question": "How should she divide it up?", "options": [{"letter": "A", "text": "Into the first fifty pages and the last fifty pages, on the basis that the middle is likely to be mostly detail.", "correct": false, "explanation": "Deciding in advance that the middle is unimportant discards material she has not read, and guidance documents commonly place the operative detail there."}, {"letter": "B", "text": "Into as many small pieces as possible, so that each request stays short and every piece is handled thoroughly.", "correct": false, "explanation": "Very small pieces multiply the number of merges and make it more likely that a rule is read without the qualification that governs it."}, {"letter": "C", "text": "Into equal blocks of thirty pages each, so that every part of the work is the same size and easy to plan around.", "correct": false, "explanation": "Fixed page blocks cut through the middle of arguments and definitions, so a part often contains a rule whose scope was set on the previous page."}, {"letter": "D", "text": "Into the document's own chapters, so each part covers one complete topic and can be understood on its own terms.", "correct": true, "explanation": "The document's own structure marks where topics begin and end, so each part arrives complete, carries its own definitions, and can be summarised without depending on material that was cut away."}], "correct": "D", "select": 1}, {"id": "a1-021", "domain": 1, "task_id": "1.3", "objective": "Iterate prompts to improve output quality", "situation": "An HR officer has spent forty minutes rewording a request for a summary of her organisation's absence policy. Each version comes back generic. She has never pasted the policy itself into the chat.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The successive versions will drift in tone, so the summary reads differently from her earlier internal notes.", "correct": false, "explanation": "Tone drift across versions is visible and easily corrected, and it is not what is keeping the summary generic."}, {"letter": "B", "text": "She will settle on a version whose phrasing suits this policy but works poorly for other documents later on.", "correct": false, "explanation": "Wording tuned narrowly to one document is a genuine limitation, and it presumes the current wording works, which here it does not."}, {"letter": "C", "text": "She keeps refining wording for a problem caused by absent material, so no rewording can produce the answer.", "correct": true, "explanation": "No phrasing can summarise a document that was never supplied. The forty minutes are spent on the wrong variable, and the fix is to paste the policy rather than to rewrite the request again."}, {"letter": "D", "text": "The wording will grow so long over successive attempts that the central instruction is lost among the detail.", "correct": false, "explanation": "Accumulated wording is a real hazard in long iteration, but a shorter request without the policy attached returns exactly the same generic answer."}], "correct": "C", "select": 1}, {"id": "a1-022", "domain": 1, "task_id": "1.1", "objective": "Create effective prompts for business and technical tasks", "situation": "A project manager pastes her internal meeting notes into the chat and asks for the weekly client update. The notes include unreleased pricing and a blunt remark about the client's own team.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The update may come back shorter than usual, since the notes cover only what was raised in that one meeting.", "correct": false, "explanation": "A shorter update is a quality concern she would spot immediately, not a consequence that reaches the client unnoticed."}, {"letter": "B", "text": "Material she supplied only as background may surface in the client-facing draft and be sent before anyone notices.", "correct": true, "explanation": "Everything pasted is treated as material available for the task unless the request says otherwise. Unreleased pricing and a candid internal remark are exactly the content that must be marked as background only, or removed before pasting."}, {"letter": "C", "text": "The notes may be too long for one message, so part of the week's discussion is left out of the client update.", "correct": false, "explanation": "A set of meeting notes sits comfortably within a single message, and omissions from length are not what makes this situation hazardous."}, {"letter": "D", "text": "The update may adopt a tone that is far too casual for the client, because her meeting notes were written casually.", "correct": false, "explanation": "Tone drift is real but visible on a first read, and it is corrected with one instruction about register."}], "correct": "B", "select": 1}, {"id": "a1-023", "domain": 1, "task_id": "1.1", "objective": "Create effective prompts for business and technical tasks", "situation": "An operations coordinator will paste Claude's output straight into a tracking spreadsheet her team already maintains.", "question": "What should her request specify about the output?", "options": [{"letter": "A", "text": "That the result should be as detailed as possible, so she can delete whatever her tracker has no room to hold.", "correct": false, "explanation": "Maximum detail moves the work rather than removing it, and deleting surplus content by hand is the step she was trying to avoid."}, {"letter": "B", "text": "That the result should be well organised and easy to read, leaving the exact presentation to Claude's judgement.", "correct": false, "explanation": "Easy to read is judged by a human reader, not by a spreadsheet. A well-organised answer in the wrong shape still has to be retyped column by column."}, {"letter": "C", "text": "The exact columns she needs, in the order her tracker uses them, with one row for each item being tracked.", "correct": true, "explanation": "The destination dictates the format, so naming the columns, their order and the row unit produces output that transfers directly instead of output she has to rearrange."}, {"letter": "D", "text": "That the result should be written as short paragraphs, which are simpler to copy across than a formatted table.", "correct": false, "explanation": "Prose is the hardest shape to move into a spreadsheet, because each paragraph has to be split into fields manually."}], "correct": "C", "select": 1}, {"id": "a1-024", "domain": 1, "task_id": "1.1", "objective": "Create effective prompts for business and technical tasks", "situation": "A grants officer asks Claude to draft the \"community impact\" section of a funding application, using a one-line prompt that names only the section. The funder publishes its assessment criteria for that section, and the application form strictly enforces a 500-word limit on it. Which two additions to that prompt would most improve the draft?", "question": "Select two.", "options": [{"letter": "A", "text": "Mention that the submission deadline falls on Friday so the request is urgent.", "correct": false, "explanation": "Deadlines and urgency are information about the officer's week, not about the task. Nothing in the brief becomes clearer, so the draft that comes back is the same draft."}, {"letter": "B", "text": "State the 500-word cap and note that the application form enforces it strictly.", "correct": true, "explanation": "and for a separate reason from the criteria: it supplies a checkable constraint plus the reason behind it. A stated hard cap lets Claude budget the section, and knowing the form enforces it stops it from treating 500 words as a rough target."}, {"letter": "C", "text": "Ask Claude to produce four separate versions and then pick the strongest itself.", "correct": false, "explanation": "Generating four versions from one under-specified brief produces four equally generic sections, and asking Claude to choose does not introduce the missing criteria or word limit. Volume is not a substitute for specification."}, {"letter": "D", "text": "Add that the section should read as compelling and memorable for the reviewers.", "correct": false, "explanation": "\"Compelling and memorable\" are style adjectives that supply no facts and no measurable target. The draft changes register slightly and stays just as thin on the substance the funder asked for."}, {"letter": "E", "text": "Paste the funder's published assessment criteria for that section.", "correct": true, "explanation": "The funder's criteria are the source material the draft has to satisfy, and pasting the real text beats describing it. Without them Claude writes generically about impact and can only guess what will be scored."}], "correct": ["B", "E"], "select": 2}, {"id": "a1-025", "domain": 1, "task_id": "1.4", "objective": "Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)", "situation": "A writer wants Claude to draft technical explanations as flowing prose paragraphs rather than bullet-heavy markdown.", "question": "Which instruction best achieves this for a drafting task?", "options": [{"letter": "A", "text": "Ask Claude to bullet every key point so the explanation is easy to scan", "correct": false, "explanation": "Requesting bullets for every point produces the opposite of the flowing prose paragraphs the writer is asking for."}, {"letter": "B", "text": "Tell Claude to write flowing prose, reserving markdown for code and simple headings", "correct": true, "explanation": "Telling Claude what to do instead of what not to do, describing the desired prose style directly, is the more effective way to steer output formatting for long-form drafting."}, {"letter": "C", "text": "Tell Claude not to use any markdown, without saying what the response should look like", "correct": false, "explanation": "A bare prohibition without describing the desired prose format is a less effective way to steer formatting than telling Claude what to produce instead."}, {"letter": "D", "text": "Leave the formatting instructions out and hope the default style matches", "correct": false, "explanation": "Omitting formatting guidance leaves the output style to chance rather than steering it toward the flowing prose the writer wants."}], "correct": "B", "select": 1}, {"id": "a1-026", "domain": 1, "task_id": "1.2", "objective": "Apply task decomposition techniques to structure complex requests", "situation": "An operations manager is asked to improve her team's onboarding process and wants to work through it with Claude across several stages.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Ask Claude to propose a plan for improving onboarding, and treat that plan as the set of stages to work through.", "correct": false, "explanation": "A proposed plan is a reasonable input, but accepting it as the structure hands over the framing decision before she has said what the work is for."}, {"letter": "B", "text": "Name what the finished work must produce, since that determines how the work divides.", "correct": true, "explanation": "Settling the deliverable is what makes a split meaningful: a revised checklist divides differently from a business case for more headcount. Until the finished product is named, any breakdown is arbitrary."}, {"letter": "C", "text": "Ask for onboarding practices used by similar teams, so the stages can be modelled on what already works well.", "correct": false, "explanation": "Comparative practice is useful later. Gathered first, it shapes the whole exercise around other teams' problems rather than her own."}, {"letter": "D", "text": "Break the work into equal-sized pieces so that each stage can be completed in roughly the same amount of time.", "correct": false, "explanation": "Equal effort is a scheduling convenience, not a structure. It cuts across the natural dependencies in the work and leaves stages that cannot start until another finishes."}], "correct": "B", "select": 1}, {"id": "a1-027", "domain": 1, "task_id": "1.3", "objective": "Iterate prompts to improve output quality", "situation": "A team has revised a customer-facing explanation nine times, each round aimed at making it read better, and they are pleased with the current version.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "Nine rounds of rewriting have changed what the text asserts, and nobody has re-checked those claims since.", "correct": true, "explanation": "Rewriting for readability quietly alters meaning: a qualifier disappears, a figure is rounded, a conditional becomes a promise. Because every round was judged on how it reads, nothing in the process re-tested accuracy."}, {"letter": "B", "text": "The nine rounds may have taken longer than writing the explanation from scratch would have taken the team.", "correct": false, "explanation": "Effort spent is a cost the team accepted, and it does not affect what customers are being told."}, {"letter": "C", "text": "The team may have grown so used to the text that they no longer notice awkward phrasing that remains in it.", "correct": false, "explanation": "Familiarity blunting a team's ear is a genuine effect, and it concerns style rather than whether the content is still true."}, {"letter": "D", "text": "The explanation may now be shorter than customers need, since each round tended to remove material from it.", "correct": false, "explanation": "Successive edits do tend to trim, but a shorter explanation is visible to the team and easily restored."}], "correct": "A", "select": 1}, {"id": "a1-028", "domain": 1, "task_id": "1.2", "objective": "Apply task decomposition techniques to structure complex requests", "situation": "A business analyst is deciding whether to send one large request or break it into stages. Which two features of a request are the strongest signals that it should be decomposed?", "question": "Select two.(Select 2)", "options": [{"letter": "A", "text": "It bundles several distinct deliverables that are meant for different audiences and uses.", "correct": true, "explanation": "Multiple deliverables with different readers pull the output in different directions, and one response has to compromise. Splitting lets each part be specified properly."}, {"letter": "B", "text": "It concerns a subject area the analyst has not personally worked in for several years.", "correct": false, "explanation": "Unfamiliarity affects how carefully the analyst must review the result, but it says nothing about whether the request itself is structurally overloaded."}, {"letter": "C", "text": "It runs to a long prompt because a good deal of background has been pasted in with it.", "correct": false, "explanation": "A long prompt is not the same as an overloaded one. Pasting rich source material is good practice and often makes a single-step request work better."}, {"letter": "D", "text": "It asks for the final answer to be presented as a table rather than as flowing paragraphs.", "correct": false, "explanation": "Output format is a presentation choice. A table can be produced in one step just as easily as prose, so it is not a reason to break the work up."}, {"letter": "E", "text": "Later parts are only worth producing if the earlier factual work is correct.", "correct": true, "explanation": "Dependency is a separate signal from bundling: when step two builds on step one, a single pass propagates any early error into everything downstream unchecked."}], "correct": ["A", "E"], "select": 2}, {"id": "a1-029", "domain": 1, "task_id": "1.4", "objective": "Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)", "situation": "A data team asks Claude to analyze a dataset and wants extended reasoning reserved for questions that truly need multi-step thought, while simple lookups get a direct answer.", "question": "How should this guidance best be phrased?", "options": [{"letter": "A", "text": "Reserve deeper reasoning for cases needing multi-step thought, answering directly otherwise", "correct": true, "explanation": "Guidance that reasoning adds latency and should be reserved for problems needing multi-step thought, defaulting to a direct response otherwise, matches the recommended way to calibrate reasoning use to task complexity."}, {"letter": "B", "text": "Tell Claude to always reason at the absolute maximum depth no matter how simple the lookup is", "correct": false, "explanation": "Always reasoning at maximum depth adds unnecessary latency for simple lookups that don't need multi-step analysis."}, {"letter": "C", "text": "Tell Claude to randomly alternate between reasoning and direct answers each time", "correct": false, "explanation": "Randomly alternating disconnects the use of reasoning from actual question complexity, which is the opposite of calibrated guidance."}, {"letter": "D", "text": "Tell Claude to never reason through any question so all replies return quickly", "correct": false, "explanation": "Never reasoning would undermine the multi-step analysis the data team wants for harder questions in the dataset."}], "correct": "A", "select": 1}, {"id": "a1-030", "domain": 1, "task_id": "1.4", "objective": "Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)", "situation": "A learning and development lead needs Claude to gather evidence on training approaches used across the sector, which is a research task rather than a drafting task. Which two prompt elements matter most here?", "question": "Select two.", "options": [{"letter": "A", "text": "The register and tone the finished piece should adopt for the board reading it later.", "correct": false, "explanation": "Tone is a drafting concern. Deciding the register now shapes how findings are worded before the lead knows what the findings actually are."}, {"letter": "B", "text": "A filled-in template that the returned material should be poured into section by section.", "correct": false, "explanation": "A rigid template presumes the shape of the answer in advance and can push the research to fill boxes rather than report what the evidence supports."}, {"letter": "C", "text": "An exact word count for the output so that it fits the slot in the meeting pack neatly.", "correct": false, "explanation": "A word count constrains presentation, and enforcing one during evidence gathering risks findings being trimmed to fit rather than reported in full."}, {"letter": "D", "text": "The boundaries of the search, such as which countries and which years should be covered.", "correct": true, "explanation": "Research needs a defined scope, otherwise there is no way to say whether the result is complete or merely whatever surfaced first."}, {"letter": "E", "text": "A request that each claim be traceable to a source the lead can open herself.", "correct": true, "explanation": "Traceability is separate from scope and is what makes a research output checkable, since the lead can confirm each claim against its origin."}], "correct": ["D", "E"], "select": 2}, {"id": "a1-031", "domain": 1, "task_id": "1.4", "objective": "Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)", "situation": "A developer wants Claude to analyze a dataset and provide a reliable root-cause diagnosis.", "question": "Which prompt structure is most effective for guiding Claude to systematically analyze the data and avoid premature conclusions?", "options": [{"letter": "A", "text": "Instruct Claude to examine the provided data and give an immediate answer by selecting the most frequent pattern and assuming it causes the observed outcome, without verifying correlation.", "correct": false, "explanation": "This approach prompts Claude to jump to a conclusion based solely on frequency, without establishing a causal link. Anthropic's prompt engineering best practices emphasize giving Claude time to think and verifying its work. Simply assuming the most frequent pattern is the cause ignores the need to correlate patterns with outcomes and test hypotheses."}, {"letter": "B", "text": "Ask Claude to filter the data for entries labeled 'critical' and report the most common entry as the root cause, without verifying its connection to the observed outcome.", "correct": false, "explanation": "This reduces analysis to a keyword filter and frequency count, ignoring the need to correlate patterns with outcomes and test hypotheses. Anthropic best practices warn against shallow heuristics and emphasize thorough, verified reasoning. Simply listing the most common 'critical' entry does not confirm causality and may miss subtle root causes."}, {"letter": "C", "text": "Guide Claude to systematically analyze the data: isolate pattern frequency per category, correlate patterns with observed outcomes, formulate hypotheses, and recommend specific tests to confirm.", "correct": true, "explanation": "This prompt structure aligns with Anthropic's recommended practices for effective prompting. According to Anthropic documentation, Claude should be given clear, step-by-step instructions and a way to verify its work (e.g., recommending tests). This systematic approach—isolating frequencies, correlating with outcomes, formulating hypotheses, and validating—encourages Claude to think carefully and avoid premature conclusions, leading to more reliable diagnoses."}, {"letter": "D", "text": "Prompt Claude to identify the first occurrence of each pattern and then conclude that the pattern that appeared earliest is the likely root cause.", "correct": false, "explanation": "This relies on a temporal heuristic—earliest occurrence—which does not establish causality. An early pattern may be unrelated to the outcome. This method skips systematic correlation and hypothesis testing, which are essential for reliable analysis."}], "correct": "C", "select": 1}, {"id": "a1-032", "domain": 1, "task_id": "1.1", "objective": "Create effective prompts for business and technical tasks", "situation": "A product manager asks Claude to \"Create an analytics dashboard\" and receives a bare-bones result with only the most basic charts. The manager wants a fully-featured result on the first try.", "question": "What should they change about the prompt?", "options": [{"letter": "A", "text": "Lower the temperature setting so that Claude systematically generates more creative dashboard layouts.", "correct": false, "explanation": "Lowering the temperature setting makes outputs more deterministic and less random, not more creative or detailed. This parameter does not compensate for a lack of explicit feature requests—Claude will still lack direction on what constitutes a fully-featured dashboard."}, {"letter": "B", "text": "Repeat the identical short prompt several times; this repetition may signal Claude to add detail on its own.", "correct": false, "explanation": "Repeating the same short prompt adds no new information about the required features or scope, so Claude has no additional guidance to produce a more detailed result. Mere repetition does not signal that it should add detail on its own."}, {"letter": "C", "text": "Ask Claude to first list ten dashboards it admires, then design its own from scratch based on those.", "correct": false, "explanation": "Asking Claude to list admired dashboards introduces an unrelated preliminary step without specifying the desired features or completeness. Even if it then designs based on those, the result may miss the specific elements the manager wants, making it an inefficient approach."}, {"letter": "D", "text": "Explicitly request many relevant features and go beyond basic charts for a fully-featured dashboard.", "correct": true, "explanation": "Explicitly requesting many relevant features and going beyond basic charts directly communicates the desired scope and completeness to Claude. This replaces the vague prompt with clear, detailed instructions, ensuring a fully-featured dashboard on the first attempt."}], "correct": "D", "select": 1}, {"id": "a1-033", "domain": 1, "task_id": "1.2", "objective": "Apply task decomposition techniques to structure complex requests", "situation": "A legal team needs to compare two lengthy contracts to highlight differences.", "question": "What is the safest decomposition approach with Claude?", "options": [{"letter": "A", "text": "Ask Claude to output a comparison table with both contracts in a single request.", "correct": false, "explanation": "Asking Claude to output a comparison table with both contracts in a single request risks exceeding the context window limit, leading to truncated input or incomplete comparisons. This approach also makes it harder to manage nuanced differences safely."}, {"letter": "B", "text": "Summarize each contract separately, then compare the two summaries for key differences.", "correct": false, "explanation": "Summarizing each contract separately may omit important legal details and precise wording differences necessary for accurate comparison. Comparing only summaries is less reliable than directly comparing extracted clauses."}, {"letter": "C", "text": "Extract the key clauses from each contract first, then compare them point by point.", "correct": true, "explanation": "Extracting key clauses first reduces complexity, focuses on relevant parts, and minimizes context overflow. This step-by-step decomposition improves accuracy by breaking the task into manageable, point-by-point comparisons."}, {"letter": "D", "text": "Load both contracts into a single context window and ask for a detailed line-by-line diff.", "correct": false, "explanation": "Loading both lengthy contracts into a single context window to generate a detailed line-by-line diff risks exceeding the token limit, resulting in truncated input or an incomplete diff. This approach is less safe than decomposing the task for thorough comparison."}], "correct": "C", "select": 1}, {"id": "a1-034", "domain": 1, "task_id": "1.3", "objective": "Iterate prompts to improve output quality", "situation": "A training lead asked for a course overview and got four short paragraphs where a detailed document was needed. Which two prompt revisions best address a response that is too brief?", "question": "Select two.", "options": [{"letter": "A", "text": "Request twice as many overviews so there is more material to choose between later.", "correct": false, "explanation": "Several shallow overviews are no substitute for one adequate document, and comparing them costs time without raising the depth of any of them."}, {"letter": "B", "text": "Ask for the answer as bullet points rather than the paragraphs that were returned.", "correct": false, "explanation": "Bullets change the presentation only. Reformatting thin content tends to make it look even thinner rather than adding what is missing."}, {"letter": "C", "text": "Set out the sections required and what each of them has to cover for this audience.", "correct": true, "explanation": "A required section list converts \"more detail\" into a concrete specification, and it makes any remaining gap visible when the next draft arrives."}, {"letter": "D", "text": "Ask for the same overview again and add the note \"please give me much more detail\".", "correct": false, "explanation": "\"More detail\" without a definition of detail is guesswork. It is the standard weak retry and usually returns a slightly longer version of the same thing."}, {"letter": "E", "text": "Say who will read it and why they need the reasoning as well as the conclusions.", "correct": true, "explanation": "Audience and purpose set the depth from a different direction: they tell Claude which explanations matter, not merely how many headings to fill."}], "correct": ["C", "E"], "select": 2}, {"id": "a1-035", "domain": 1, "task_id": "1.4", "objective": "Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)", "situation": "A team lead wants Claude to brainstorm mitigation options for a project risk and then, separately, critically evaluate the feasibility of each option before a final recommendation.", "question": "Which structure fits this two-part need?", "options": [{"letter": "A", "text": "Ask Claude to evaluate feasibility before any mitigation ideas even exist yet", "correct": false, "explanation": "Evaluating feasibility before any ideas exist reverses the necessary order of brainstorming first and then assessing the generated options."}, {"letter": "B", "text": "Skip generating multiple ideas and produce only the single most feasible option", "correct": false, "explanation": "Producing only a single mitigation option skips the broad brainstorming step meant to surface a range of ideas before evaluation."}, {"letter": "C", "text": "Ask for the brainstormed ideas and the evaluation together in one combined response", "correct": false, "explanation": "Combining both steps into one response does not give the separately inspectable brainstorm and evaluation stages the team lead is asking for."}, {"letter": "D", "text": "Chain two calls: generate mitigation ideas, then evaluate each one's feasibility", "correct": true, "explanation": "Chaining separate calls, generating ideas first and then reviewing them against criteria in a distinct step, fits the self-correction chaining pattern and supports separately inspecting the brainstormed ideas and the feasibility evaluation."}], "correct": "D", "select": 1}, {"id": "a1-036", "domain": 1, "task_id": "1.1", "objective": "Create effective prompts for business and technical tasks", "situation": "A marketing coordinator asked Claude to \"write an email about our new feature\" and received a generic announcement that could describe almost any product. Which two changes to the prompt would most improve the next draft?", "question": "Select two.", "options": [{"letter": "A", "text": "Send the same request a second time in capitals so the requirements are harder to miss.", "correct": false, "explanation": "Repeating an identical prompt with added emphasis is the classic non-fix. Nothing new has been communicated, so there is no reason for the output to improve."}, {"letter": "B", "text": "Say who will receive the email and what they should do after reading it.", "correct": true, "explanation": "Telling Claude who will receive the email and what they should do afterwards defines what the email is for, without assuming anything about who those readers are. This is a separate fix from supplying facts: it constrains framing, emphasis and the call to action."}, {"letter": "C", "text": "Paste in the real release notes rather than describing the feature in a single line.", "correct": true, "explanation": "A generic draft is usually a symptom of missing source material. Pasting the real release notes gives Claude the specifics it otherwise has to invent or gloss over."}, {"letter": "D", "text": "Ask for a much longer draft, on the view that a longer email carries more useful detail.", "correct": false, "explanation": "Length is not the problem. Asking for more words on an under-specified brief produces a longer generic email, and often padding rather than substance."}, {"letter": "E", "text": "Add a sentence asking Claude to write in a warm voice that readers will genuinely enjoy.", "correct": false, "explanation": "Style adjectives such as \"warm\" and \"enjoyable\" do not supply any of the missing product facts, so the draft stays generic and simply changes register."}], "correct": ["B", "C"], "select": 2}, {"id": "a1-037", "domain": 1, "task_id": "1.4", "objective": "Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)", "situation": "A communications team has a reliable recipe for drafting: name the audience, set the format and attach examples of accepted work. Applied to a question about why complaint volumes doubled, it returns a weak answer.", "question": "What is missing?", "options": [{"letter": "A", "text": "Examples of complaint analyses that the team has previously accepted, so the shape of a good answer is clear.", "correct": false, "explanation": "Examples convey the shape and register of a finished piece, which is why they work for drafting. They cannot supply the figures this question turns on."}, {"letter": "B", "text": "A stated format for the answer, such as a short summary followed by the supporting detail underneath it.", "correct": false, "explanation": "Format organises an answer, and the problem here is that there is no substance for a format to organise."}, {"letter": "C", "text": "A more precisely named audience, since an answer written for the board differs from one written for the team.", "correct": false, "explanation": "A sharper audience improves how the answer is pitched, and an answer pitched perfectly at the board can still be built on nothing."}, {"letter": "D", "text": "The complaint data itself, along with the specific question about it that the answer is expected to settle.", "correct": true, "explanation": "Drafting is constrained by audience, format and voice; analysis is constrained by evidence and by the question being asked. The recipe supplies all of the first set and none of the second, so it produces well-presented speculation."}], "correct": "D", "select": 1}, {"id": "a1-038", "domain": 1, "task_id": "1.3", "objective": "Iterate prompts to improve output quality", "situation": "Over eight rounds of revision, a team notices that fixing the newest complaint often brings back a problem they had solved three rounds earlier. Nothing about their requirements is written down.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The team will spend more time on revisions than they would have spent writing the document themselves.", "correct": false, "explanation": "Time spent is a real cost, but the team would accept it if the rounds were converging. The problem is that they are not."}, {"letter": "B", "text": "The wording of the request will become long and repetitive as fixes for each separate complaint accumulate.", "correct": false, "explanation": "An accumulating request is an untidy side effect, and it does not by itself cause earlier fixes to be lost."}, {"letter": "C", "text": "The eighth version will read less naturally than the first, because it has been adjusted so many times over.", "correct": false, "explanation": "Prose quality drifting under repeated revision is worth watching, but it is a milder issue than requirements silently reversing."}, {"letter": "D", "text": "With no record of what the document must satisfy, each round can undo an earlier fix and nobody notices.", "correct": true, "explanation": "Without a written list of everything the document must satisfy, each round is checked only against the latest complaint. Requirements settled earlier are not re-tested, so the work circles instead of converging."}], "correct": "D", "select": 1}, {"id": "a1-039", "domain": 1, "task_id": "1.1", "objective": "Create effective prompts for business and technical tasks", "situation": "A communications lead has twenty minutes to produce a holding statement about a service outage. Details are still arriving from the operations team and some are unconfirmed.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Ask what similar organisations usually say during an outage, and adapt the closest example to this situation.", "correct": false, "explanation": "Borrowed wording describes someone else's incident. Adapting it risks asserting a cause or a restoration time that has not been established here."}, {"letter": "B", "text": "Wait until operations has confirmed every detail, then ask for a single accurate statement for publication.", "correct": false, "explanation": "A holding statement exists precisely because confirmation is incomplete, so waiting for certainty means missing the twenty-minute window."}, {"letter": "C", "text": "Write down which facts are confirmed and which are not, and give Claude both lists with the request itself.", "correct": true, "explanation": "Separating confirmed from unconfirmed is the judgement only she can make, and supplying both lists lets the statement assert what is known while explicitly not committing to what is not."}, {"letter": "D", "text": "Ask for a general outage holding statement now, and add the organisation's specific details once they are confirmed.", "correct": false, "explanation": "A generic statement drafted without the confirmed facts has to be rewritten once they arrive, and it invites unconfirmed details to be slotted in under deadline pressure."}], "correct": "C", "select": 1}, {"id": "a1-040", "domain": 1, "task_id": "1.4", "objective": "Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)", "situation": "A product marketer needs a summary of what three competitors announced during the past quarter.", "question": "Which addition to the request matters most for this kind of task?", "options": [{"letter": "A", "text": "A request that each announcement carry the source it came from and the date, so recent claims can be checked.", "correct": true, "explanation": "Recent events are exactly where an unsupported summary is most likely to go wrong, so a research task of this kind needs each item traceable to where it was reported and when. Sources and dates are what make the summary checkable."}, {"letter": "B", "text": "A request that the summary be organised competitor by competitor rather than being organised by announcement date.", "correct": false, "explanation": "Grouping by competitor makes the summary easier to navigate, and it does nothing to establish that the announcements listed were real."}, {"letter": "C", "text": "A request that the summary group the announcements by the product area each of the three announcements affects.", "correct": false, "explanation": "Grouping by product area is a useful lens for her team, but it arranges claims that have not yet been shown to be accurate."}, {"letter": "D", "text": "A request that the summary be written in the same style as the internal updates her team already circulates.", "correct": false, "explanation": "Matching house style makes the summary comfortable to read, which is a drafting concern rather than a research one."}], "correct": "A", "select": 1}, {"id": "a1-041", "domain": 1, "task_id": "1.4", "objective": "Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)", "situation": "An events team wants a wide field of ideas for a member engagement campaign before narrowing down. Which two prompt choices best suit this brainstorming stage?", "question": "Select two.", "options": [{"letter": "A", "text": "Ask for the single strongest idea straight away so the team can start work sooner.", "correct": false, "explanation": "Jumping to one recommendation skips the divergent step entirely, and it hides the alternatives the team asked for precisely so they could compare them."}, {"letter": "B", "text": "Provide one finished campaign write-up and ask for close variations on that model.", "correct": false, "explanation": "Anchoring on a finished example narrows the output to near-copies, which is useful when matching a template but counterproductive when seeking range."}, {"letter": "C", "text": "State the limits any idea must respect, such as a small budget and six weeks.", "correct": true, "explanation": "Constraints serve a different purpose from breadth: they keep the wide field usable by excluding options the team could never actually run."}, {"letter": "D", "text": "Fix the tone, house style and final format so every idea arrives ready to publish.", "correct": false, "explanation": "Publication-ready styling belongs to the drafting stage. Applied now it wastes effort on ideas that will mostly be discarded and slows the divergent pass."}, {"letter": "E", "text": "Ask for a spread of clearly different directions, each given a short label and a line.", "correct": true, "explanation": "Brainstorming values coverage over polish. Short labelled options let the team scan a wide field cheaply, which polished prose actively discourages."}], "correct": ["C", "E"], "select": 2}, {"id": "a1-042", "domain": 1, "task_id": "1.2", "objective": "Apply task decomposition techniques to structure complex requests", "situation": "An events lead must recommend a new format for her organisation's annual conference to the board, and plans to work through it with Claude in stages rather than in one request.", "question": "Which sequence is correct?", "options": [{"letter": "A", "text": "Assemble last year's delegate feedback and the cost figures for each format; compare the candidate formats against those constraints; agree the decision the board must make and the budget and venue constraints; draft the recommendation the board will read.", "correct": false, "explanation": "The constraints arrive after the comparison is already finished, so they cannot shape it."}, {"letter": "B", "text": "Assemble last year's delegate feedback and the cost figures for each format; agree the decision the board must make and the budget and venue constraints; compare the candidate formats against those constraints; draft the recommendation the board will read.", "correct": false, "explanation": "Gathering evidence before agreeing the decision and constraints collects material that may turn out to be irrelevant."}, {"letter": "C", "text": "Agree the decision the board must make and the budget and venue constraints; compare the candidate formats against those constraints; assemble last year's delegate feedback and the cost figures for each format; draft the recommendation the board will read.", "correct": false, "explanation": "Comparing the formats before the feedback and costs are assembled leaves the comparison resting on nothing."}, {"letter": "D", "text": "Agree the decision the board must make and the budget and venue constraints; assemble last year's delegate feedback and the cost figures for each format; compare the candidate formats against those constraints; draft the recommendation the board will read.", "correct": true, "explanation": "The decision and its constraints come first, because they determine which evidence matters; evidence is then gathered, the formats are compared against the constraints, and only then is the recommendation written."}], "correct": "D", "select": 1}, {"id": "a1-043", "domain": 1, "task_id": "1.2", "objective": "Apply task decomposition techniques to structure complex requests", "situation": "A researcher has produced separate findings for each of five regional markets and now wants them combined into one view.", "question": "What should the combining request specify?", "options": [{"letter": "A", "text": "That the five regions should each appear in the combined view in order of the revenue they contribute to the business.", "correct": false, "explanation": "Revenue order affects presentation only, and it can bury a serious finding from a small market at the end of the document."}, {"letter": "B", "text": "That the combined view should be shorter than the five sets of findings added together, to keep it readable.", "correct": false, "explanation": "A length target shapes the summary but says nothing about which findings survive, so the reduction happens on unstated grounds."}, {"letter": "C", "text": "How to handle findings that overlap and findings that conflict, so those decisions are made openly and not quietly.", "correct": true, "explanation": "Merging is where a genuine disagreement between regions can be smoothed into a single confident sentence. Saying in advance how overlaps and conflicts are treated keeps those differences visible, since a conflict between regions is often the most useful finding."}, {"letter": "D", "text": "That the strongest of the five sets of findings should be used as the template the other four are all rewritten into.", "correct": false, "explanation": "Forcing four regions into the shape of a fifth discards whatever each of them found that does not fit that template."}], "correct": "C", "select": 1}, {"id": "a1-044", "domain": 1, "task_id": "1.3", "objective": "Iterate prompts to improve output quality", "situation": "An analyst has refined the wording she uses for supplier reviews across many rounds, using the same supplier's file throughout. She now wants to adopt it as her team's standard wording.", "question": "What should she do before that?", "options": [{"letter": "A", "text": "Record the reasoning behind each instruction, so that colleagues understand why the wording is written this way.", "correct": false, "explanation": "Documenting the reasoning helps the team maintain the wording later, but it assumes the wording already generalises."}, {"letter": "B", "text": "Shorten the wording so that colleagues who have not followed the refinement can still take it in quickly.", "correct": false, "explanation": "Shortening improves how approachable the wording is, and it may remove instructions whose value has not been established."}, {"letter": "C", "text": "Ask colleagues to read the wording through and confirm that they find the instructions in it clear enough.", "correct": false, "explanation": "A readability check confirms colleagues can follow the instructions, not that the instructions work on a supplier file she never tested."}, {"letter": "D", "text": "Try it on two or three other suppliers' files, since it may have been shaped to fit the one file she used.", "correct": true, "explanation": "Wording refined against a single example tends to encode that example's quirks, such as its headings or its level of detail. Trying it on other files is what separates a genuine improvement from one that only fits the case it was built on."}], "correct": "D", "select": 1}, {"id": "a1-045", "domain": 1, "task_id": "1.3", "objective": "Iterate prompts to improve output quality", "situation": "After four rounds, an internal newsletter item is accurate and well structured, but two headings use the wrong department names. The item is due in ten minutes.", "question": "What should the editor do?", "options": [{"letter": "A", "text": "Rewrite her original request to include the department names, and produce a fresh version of the whole item.", "correct": false, "explanation": "Regenerating the item risks the accuracy and structure that four rounds established, with ten minutes left to verify them."}, {"letter": "B", "text": "Correct the two headings herself, since the remaining gap is smaller than the effort of another full round.", "correct": true, "explanation": "Iteration is worth continuing while it is cheaper than doing the work by hand. Two headings are a few seconds of typing, so this is the point to stop iterating and finish it herself."}, {"letter": "C", "text": "Publish the item as it stands and issue a short correction about the department names in next week's edition.", "correct": false, "explanation": "Publishing known errors and correcting them later costs the team credibility over a fix that takes moments."}, {"letter": "D", "text": "Run one more round asking for the department names to be corrected, and read the returned version closely.", "correct": false, "explanation": "Another round can fix two words, but it returns a whole document that has to be re-read to confirm nothing else moved."}], "correct": "B", "select": 1}, {"id": "a1-046", "domain": 1, "task_id": "1.1", "objective": "Create effective prompts for business and technical tasks", "situation": "A prompt describes a multi-step setup process, such as configuring an environment and then running a script, but Claude sometimes performs the steps out of order.", "question": "What change would make the required order unambiguous?", "options": [{"letter": "A", "text": "Provide the instructions as a numbered list of steps in the exact order they must run", "correct": true, "explanation": "Providing instructions as sequential steps using numbered lists or bullet points is recommended specifically when the order or completeness of the steps matters, removing ambiguity about sequence."}, {"letter": "B", "text": "Combine all of the steps into a single long sentence separated only by commas", "correct": false, "explanation": "A single comma-separated sentence obscures step boundaries and order even more than a plain paragraph would, making the sequence less clear rather than more."}, {"letter": "C", "text": "Describe the steps in a random order and let Claude infer the correct sequence", "correct": false, "explanation": "Describing steps out of order and relying on inference is the opposite of making the required sequence unambiguous."}, {"letter": "D", "text": "Remove the step numbers and rely only on paragraph breaks to imply the correct order", "correct": false, "explanation": "Relying only on paragraph breaks without numbering removes the explicit ordering cue that numbered lists provide for steps where sequence matters."}], "correct": "A", "select": 1}, {"id": "a1-047", "domain": 1, "task_id": "1.2", "objective": "Apply task decomposition techniques to structure complex requests", "situation": "A researcher has fifteen sets of interview notes and must produce a findings section for a report.", "question": "Which sequence of stages should she follow?", "options": [{"letter": "A", "text": "Check each theme against the notes; pull the relevant extracts from each set of notes; write the findings section; group the extracts into themes.", "correct": false, "explanation": "Checking themes that have not yet been formed is not possible, and grouping after the section is written leaves the writing unsupported."}, {"letter": "B", "text": "Group the extracts into themes; check each theme against the notes; write the findings section; pull the relevant extracts from each set of notes.", "correct": false, "explanation": "Themes cannot be grouped before the extracts have been pulled, so this ordering starts from material that has not been gathered."}, {"letter": "C", "text": "Pull the relevant extracts from each set of notes; group the extracts into themes; check each theme against the notes; write the findings section.", "correct": true, "explanation": "Extracts come out of the notes first, grouping turns them into themes, the check confirms each theme is genuinely supported before it becomes a claim, and the section is written last from verified material."}, {"letter": "D", "text": "Write the findings section; group the extracts into themes; pull the relevant extracts from each set of notes; check each theme against the notes.", "correct": false, "explanation": "Findings written before the extracts exist are assertions in search of evidence, and the checking stage arrives after the section is already written."}], "correct": "C", "select": 1}, {"id": "a1-048", "domain": 1, "task_id": "1.4", "objective": "Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)", "situation": "A product team is brainstorming new feature ideas and wants Claude to generate a wide range of genuinely different concepts rather than several variations on the same idea.", "question": "What should the team do when providing example ideas in the prompt?", "options": [{"letter": "A", "text": "Include no examples at all so nothing biases the range of ideas produced", "correct": false, "explanation": "While examples aren't required, omitting them does not itself guarantee variety; well-chosen diverse examples are the recommended way to actively support a broad range of ideas."}, {"letter": "B", "text": "Include a small set of deliberately varied examples so Claude avoids one narrow pattern", "correct": true, "explanation": "Examples should be diverse, covering edge cases and varying enough that Claude does not pick up an unintended narrow pattern, which supports a wide range of brainstormed ideas."}, {"letter": "C", "text": "Include just one example and ask Claude to generate close variations of it", "correct": false, "explanation": "Asking for close variations of one example directly narrows the output to minor tweaks instead of a wide range of distinct ideas."}, {"letter": "D", "text": "Include many examples that all follow the exact same structure as just one clear pattern", "correct": false, "explanation": "Uniform examples that all share the same structure and theme encourage Claude to converge on that single pattern rather than exploring genuinely different concepts."}], "correct": "B", "select": 1}, {"id": "a1-049", "domain": 1, "task_id": "1.2", "objective": "Apply task decomposition techniques to structure complex requests", "situation": "A team's breakdown of a supplier review ends with a stage described as 'format the findings and add the recommended supplier to the summary page'.", "question": "What is the greatest concern with this stage?", "options": [{"letter": "A", "text": "Formatting and writing in the same stage makes the stage take longer than the other stages in the breakdown.", "correct": false, "explanation": "Uneven stage durations are a planning detail rather than a risk to the integrity of the review."}, {"letter": "B", "text": "The recommended supplier will appear on the summary page before it appears in the body of the review itself.", "correct": false, "explanation": "Ordering within the document is an editorial choice, and the recommendation appearing early is normal in a review of this kind."}, {"letter": "C", "text": "The summary page may end up formatted differently from the rest of the review, which readers will find jarring.", "correct": false, "explanation": "Inconsistent formatting is a cosmetic defect that a single pass fixes, and it does not affect which supplier is chosen."}, {"letter": "D", "text": "A decision that should be made and reviewed on its own is folded into a step everyone treats as presentation.", "correct": true, "explanation": "Naming the recommendation is the consequential act of the whole review, and burying it in a formatting step means nobody schedules a review of it. Decisions should sit in stages of their own, where they can be challenged."}], "correct": "D", "select": 1}, {"id": "a1-050", "domain": 1, "task_id": "1.2", "objective": "Apply task decomposition techniques to structure complex requests", "situation": "A business analyst is deciding whether to send one large request or break it into stages. Which two features of a request are the strongest signals that it should be decomposed?", "question": "Select two.", "options": [{"letter": "A", "text": "Later parts are only worth producing if the earlier factual work is correct.", "correct": true, "explanation": "Dependency is a separate signal from bundling: when step two builds on step one, a single pass propagates any early error into everything downstream unchecked."}, {"letter": "B", "text": "It runs to a long prompt because a good deal of background has been pasted in with it.", "correct": false, "explanation": "A long prompt is not the same as an overloaded one. Pasting rich source material is good practice and often makes a single-step request work better."}, {"letter": "C", "text": "It bundles several distinct deliverables that are meant for different audiences and uses.", "correct": true, "explanation": "Multiple deliverables with different readers pull the output in different directions, and one response has to compromise. Splitting lets each part be specified properly."}, {"letter": "D", "text": "It asks for the final answer to be presented as a table rather than as flowing paragraphs.", "correct": false, "explanation": "Output format is a presentation choice. A table can be produced in one step just as easily as prose, so it is not a reason to break the work up."}, {"letter": "E", "text": "It concerns a subject area the analyst has not personally worked in for several years.", "correct": false, "explanation": "Unfamiliarity affects how carefully the analyst must review the result, but it says nothing about whether the request itself is structurally overloaded."}], "correct": ["A", "C"], "select": 2}, {"id": "a1-051", "domain": 1, "task_id": "1.4", "objective": "Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)", "situation": "A team asks for 'a polished two-page write-up explaining why our regional sales fell'. They receive a confident, elegantly written explanation.", "question": "What is the greatest risk in how they framed this?", "options": [{"letter": "A", "text": "The write-up may cover several causes at once when the audience would prefer one clear headline reason.", "correct": false, "explanation": "Several causes may well be the honest answer, and choosing a single headline reason is an editorial decision rather than a hazard."}, {"letter": "B", "text": "The write-up may be longer than the audience for a regional sales question is realistically willing to read.", "correct": false, "explanation": "Two pages is a modest length for this kind of question, and length can be adjusted without changing whether the explanation is sound."}, {"letter": "C", "text": "The write-up may use vocabulary the regional sales team does not share, which slows down their response.", "correct": false, "explanation": "Unshared vocabulary is a presentation problem, and it is corrected by naming the audience rather than by changing the task."}, {"letter": "D", "text": "They asked for a finished piece of writing about a question that first needs working out from the figures.", "correct": true, "explanation": "The request is shaped as a drafting task when the underlying work is analysis. Because polish was what was asked for, polish is what arrives, and a fluent explanation of an unexamined cause is very hard for a reader to tell apart from a real finding."}], "correct": "D", "select": 1}, {"id": "a1-052", "domain": 1, "task_id": "1.4", "objective": "Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)", "situation": "A content team wants Claude to draft a press release, have the draft checked against their style guide, and then produce a revised version based on that check, with each stage logged separately for review.", "question": "How should this drafting workflow be set up?", "options": [{"letter": "A", "text": "Ask Claude to draft the release ten different ways and pick the best one", "correct": false, "explanation": "Generating many parallel drafts without a review step against the style guide does not fulfill the requirement to check drafts against specific criteria before revising."}, {"letter": "B", "text": "Ask for the draft, the style review, and the revision all together in a single response", "correct": false, "explanation": "Bundling all three stages into one response does not give the team separately logged and inspectable outputs at each stage."}, {"letter": "C", "text": "Chain three calls: draft the release, review it against the style guide, then refine it", "correct": true, "explanation": "This is the self-correction chaining pattern for drafting tasks: generate a draft, review it against criteria, then refine, with each step as a separate, inspectable call."}, {"letter": "D", "text": "Skip the review step and have Claude produce only a final polished draft", "correct": false, "explanation": "Skipping the style guide review removes the criteria check the team explicitly wants applied before finalizing the draft."}], "correct": "C", "select": 1}, {"id": "a1-053", "domain": 1, "task_id": "1.4", "objective": "Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)", "situation": "A team wants Claude to brainstorm names for a new product from the perspective of a branding specialist, aiming for tone and instincts consistent with that professional background throughout the session.", "question": "What is a simple, effective way to set this up?", "options": [{"letter": "A", "text": "Set a role in the system prompt describing Claude as a branding specialist", "correct": true, "explanation": "Setting a role in the system prompt, even a single descriptive sentence, focuses Claude's behavior and tone for the use case, which fits framing a brainstorming session around a branding specialist perspective."}, {"letter": "B", "text": "Ask Claude to name the product with absolutely no role framing given at all", "correct": false, "explanation": "Skipping role framing leaves the tone and instincts of the response to chance rather than steering them toward a branding specialist perspective."}, {"letter": "C", "text": "Repeat the word 'creative' many times instead of describing any actual role", "correct": false, "explanation": "Repeating a vague adjective does not establish the professional framing or tone the team wants for the brainstorming session."}, {"letter": "D", "text": "Provide the branding background only in the very last sentence", "correct": false, "explanation": "Burying the role description at the end of a long message is a weaker way to establish framing than setting it clearly, such as in the system prompt."}], "correct": "A", "select": 1}, {"id": "a1-054", "domain": 1, "task_id": "1.1", "objective": "Create effective prompts for business and technical tasks", "situation": "A project manager's prompt is written with heavy markdown formatting, bullet points, and bold text, and asks Claude to \"respond in plain text only.\" Claude keeps returning markdown-formatted output anyway.", "question": "What is the most likely fix?", "options": [{"letter": "A", "text": "Ask a follow-up question in a second turn, repeating the instruction using bullet points", "correct": false, "explanation": "Repeating the instruction using bullet points again models the exact formatting style the manager wants to avoid, reinforcing the problem rather than fixing it."}, {"letter": "B", "text": "Add the words \"no markdown\" in bold text at the very top of the prompt so it stands out", "correct": false, "explanation": "Emphasizing the instruction with bold text keeps the prompt itself in markdown, which works against the stated goal of getting a plain-text response."}, {"letter": "C", "text": "Increase the maximum token count so Claude has much more room to comply with the plain-text request", "correct": false, "explanation": "Token limits control response length, not formatting style, so raising the limit would not resolve the markdown issue."}, {"letter": "D", "text": "Rewrite the prompt itself without markdown, since Claude's style tends to mirror the prompt's style", "correct": true, "explanation": "The formatting style used in the prompt can influence Claude's response style, so matching the prompt's style to the desired output style, such as removing markdown from the prompt, helps reduce unwanted markdown in the response."}], "correct": "D", "select": 1}, {"id": "a1-055", "domain": 1, "task_id": "1.4", "objective": "Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)", "situation": "A membership team must recommend one new member benefit to its board.", "question": "Which sequence of task types should they work through?", "options": [{"letter": "A", "text": "Write the recommendation; generate candidate benefits; look up what comparable bodies offer; weigh the shortlist against their budget.", "correct": false, "explanation": "Writing the recommendation first fixes the conclusion, and the generation and research stages then serve to justify a choice already made."}, {"letter": "B", "text": "Look up what comparable bodies offer; write the recommendation; weigh the shortlist against their budget; generate candidate benefits.", "correct": false, "explanation": "Research before generation is defensible, but the recommendation is written second here, ahead of the weighing and the generation it depends on."}, {"letter": "C", "text": "Generate candidate benefits; look up what comparable bodies offer; weigh the shortlist against their budget; write the recommendation.", "correct": true, "explanation": "Each task type feeds the next: open generation widens the field, research grounds the candidates in what others actually do, weighing against the budget narrows them, and only then is there a decision worth writing up."}, {"letter": "D", "text": "Weigh the shortlist against their budget; write the recommendation; generate candidate benefits; look up what comparable bodies offer.", "correct": false, "explanation": "A shortlist cannot be weighed before any candidates have been generated, so this ordering starts from something that does not yet exist."}], "correct": "C", "select": 1}, {"id": "a1-056", "domain": 1, "task_id": "1.2", "objective": "Apply task decomposition techniques to structure complex requests", "situation": "A manager has broken a customer-refund policy rewrite into five stages and wants one point at which she personally reviews the work.", "question": "Where should that review sit?", "options": [{"letter": "A", "text": "After the stage that settles which refund cases are covered, because every later stage is written to that decision.", "correct": true, "explanation": "A checkpoint is worth most immediately after the decision that constrains everything downstream. If the covered cases are wrong, the wording, the examples and the exceptions are all wrong too, and catching that later means rewriting all of them."}, {"letter": "B", "text": "After the first stage, since reviewing early gives her the longest possible time to act on anything she finds.", "correct": false, "explanation": "Reviewing the first stage is cheap but often premature, since the first stage may be preparatory and not yet contain the decision that matters."}, {"letter": "C", "text": "At whichever stage happens to fall in the middle, which keeps the amount of unreviewed work on each side even.", "correct": false, "explanation": "Splitting by position balances the volume of unreviewed work but ignores which stage actually carries the consequential decision."}, {"letter": "D", "text": "At the very end, once all five stages are complete, so that she reviews the finished policy as customers will read it.", "correct": false, "explanation": "A final review catches errors at the point where correcting them means redoing the stages that followed the mistake."}], "correct": "A", "select": 1}, {"id": "a1-057", "domain": 1, "task_id": "1.1", "objective": "Create effective prompts for business and technical tasks", "situation": "A programme manager is writing a single request asking Claude to turn a supplier's quarterly report into a short briefing note for her steering group.", "question": "In what order should she assemble the request?", "options": [{"letter": "A", "text": "State the length and sections needed; name the task and who reads the note; say what to do where the report is silent; paste the supplier report and label it.", "correct": false, "explanation": "Length and sections are decisions about presentation. Fixing them before the task and the source are on the table means the shape is chosen before the content is known."}, {"letter": "B", "text": "Say what to do where the report is silent; state the length and sections needed; paste the supplier report and label it; name the task and who reads the note.", "correct": false, "explanation": "Handling for gaps is meaningless until the task and the material are defined, so this ordering asks for edge-case rules before the main case exists."}, {"letter": "C", "text": "Name the task and who reads the note; paste the supplier report and label it; state the length and sections needed; say what to do where the report is silent.", "correct": true, "explanation": "The task and audience frame everything that follows, the labelled source supplies the substance, format then shapes it, and the instruction about gaps covers what the source cannot answer. Each step depends on the one before it."}, {"letter": "D", "text": "Paste the supplier report and label it; say what to do where the report is silent; state the length and sections needed; name the task and who reads the note.", "correct": false, "explanation": "Leading with the pasted report and holding back the task until the end means the material arrives with no stated purpose, and gap handling is decided before anyone knows what the note is for."}], "correct": "C", "select": 1}, {"id": "a1-058", "domain": 1, "task_id": "1.3", "objective": "Iterate prompts to improve output quality", "situation": "A bid writer keeps receiving long proposal drafts organised in a way her reviewers dislike, and each rewrite costs a full round.", "question": "Which sequence should she follow instead?", "options": [{"letter": "A", "text": "Have the draft written to the agreed outline; ask for a section outline; correct the outline with reviewers; check the draft against the outline.", "correct": false, "explanation": "The draft is written before any outline exists, so the expensive step happens first and the outline is produced after the structure has already been committed to."}, {"letter": "B", "text": "Correct the outline with reviewers; ask for a section outline; check the draft against the outline; have the draft written to the agreed outline.", "correct": false, "explanation": "Reviewers cannot correct an outline that has not been requested yet, and the draft is written last, after the check that was meant to test it."}, {"letter": "C", "text": "Check the draft against the outline; have the draft written to the agreed outline; ask for a section outline; correct the outline with reviewers.", "correct": false, "explanation": "Checking a draft that has not been produced is impossible, and the outline work arrives after both the draft and its review."}, {"letter": "D", "text": "Ask for a section outline; correct the outline with reviewers; have the draft written to the agreed outline; check the draft against the outline.", "correct": true, "explanation": "The disagreement is about structure, and an outline is the cheapest thing that exposes structure. Reviewers settle the shape in minutes, the full draft is written only once, and the final check confirms it followed the agreed outline."}], "correct": "D", "select": 1}, {"id": "a1-059", "domain": 1, "task_id": "1.3", "objective": "Iterate prompts to improve output quality", "situation": "A communications officer receives a first draft that is accurate but far too shallow for the intended briefing. Which two follow-up actions are most likely to improve the second attempt?", "question": "Select two.", "options": [{"letter": "A", "text": "Send the identical prompt again in the hope that the next attempt comes out stronger.", "correct": false, "explanation": "Resending an unchanged prompt is the least productive iteration. The specification is identical, so any improvement is chance rather than something that can be relied on."}, {"letter": "B", "text": "Show a passage from last quarter's briefing as an example of the depth wanted.", "correct": true, "explanation": "An example communicates a standard that adjectives cannot, and it works even where the officer would struggle to describe the wanted depth in words."}, {"letter": "C", "text": "Append the words \"be thorough and detailed\" to the end of the original request text.", "correct": false, "explanation": "Generic intensifiers are among the weakest edits available. \"Thorough\" is undefined, so it gives Claude no target it can aim the second draft at."}, {"letter": "D", "text": "Supply the figures and case details the briefing must cover, which were not in the prompt.", "correct": true, "explanation": "Shallowness is usually an information problem. Claude cannot write to a depth the prompt never supplied the material for, so adding the substance is the direct fix."}, {"letter": "E", "text": "Reply that the draft was disappointing and ask Claude to try considerably harder now.", "correct": false, "explanation": "Negative feedback without specifics tells Claude that something was wrong but not what, so the next draft varies at random rather than converging."}], "correct": ["B", "D"], "select": 2}, {"id": "a1-060", "domain": 1, "task_id": "1.3", "objective": "Iterate prompts to improve output quality", "situation": "For six weeks a customer-service lead has typed the same three corrections about tone and terminology into every new chat in her team's project.", "question": "What should she do about that?", "options": [{"letter": "A", "text": "Keep the three corrections in a note she can copy, so retyping them into each new chat takes only a moment.", "correct": false, "explanation": "A copyable note reduces the typing but keeps the requirement dependent on her remembering, and colleagues in the same project still do without it."}, {"letter": "B", "text": "Move the three corrections into the project's standing instructions, so chats there start with them applied.", "correct": true, "explanation": "A correction needed in every chat is standing context rather than a per-chat instruction. Project instructions apply to every conversation in that project, which makes the fix durable and available to the whole team."}, {"letter": "C", "text": "Add the three corrections at the end of each chat instead, so they are the most recent thing in the conversation.", "correct": false, "explanation": "Moving corrections to the end of a chat applies them after the work has been produced, so each conversation still starts from the wrong footing."}, {"letter": "D", "text": "Ask at the start of each chat whether the three usual corrections are understood before the real work begins.", "correct": false, "explanation": "A confirmation exchange at the start costs a turn and establishes nothing that survives into the next conversation."}], "correct": "B", "select": 1}, {"id": "a1-061", "domain": 1, "task_id": "1.3", "objective": "Iterate prompts to improve output quality", "situation": "A grants officer has a draft that is right in structure and evidence, but its second section is too technical for her readers.", "question": "What is the most efficient next step?", "options": [{"letter": "A", "text": "Rewrite her original request to mention the reader's level, and ask for the whole document to be produced again.", "correct": false, "explanation": "Regenerating everything puts the sections she was happy with back at risk, and she then has to re-review the whole document."}, {"letter": "B", "text": "Accept the draft as it stands and rewrite the technical section herself before the document goes any further.", "correct": false, "explanation": "Rewriting it herself is a reasonable fallback, but doing it before asking for a revision gives up the cheapest improvement available."}, {"letter": "C", "text": "Ask for three alternative versions of the whole document, and choose whichever of them reads most accessibly.", "correct": false, "explanation": "Producing three full alternatives multiplies the reading and comparison work to fix a problem confined to one section."}, {"letter": "D", "text": "Ask for the second section to be rewritten for a non-specialist reader, leaving the rest of the draft as it is.", "correct": true, "explanation": "Only one part is wrong, so only that part needs to change. A targeted revision keeps the structure and evidence she has already accepted, and it makes the improvement easy to check."}], "correct": "D", "select": 1}, {"id": "a1-062", "domain": 1, "task_id": "1.3", "objective": "Iterate prompts to improve output quality", "situation": "A training officer wants a disciplined way to improve the wording she uses for weekly course summaries.", "question": "Which sequence should she follow on each round?", "options": [{"letter": "A", "text": "Compare the new result against the previous one; name what was wrong with the result; keep the change or undo it; make one change.", "correct": false, "explanation": "There is nothing to compare until a change has been made, so this ordering evaluates a result that the round has not yet produced."}, {"letter": "B", "text": "Make one change; compare the new result against the previous one; name what was wrong with the result; keep the change or undo it.", "correct": false, "explanation": "Changing something before naming the defect means the change is aimed at nothing in particular, and the diagnosis arrives after it has been made."}, {"letter": "C", "text": "Keep the change or undo it; name what was wrong with the result; compare the new result against the previous one; make one change.", "correct": false, "explanation": "Deciding whether to keep a change before making it inverts the round entirely, and the comparison happens with nothing new to compare."}, {"letter": "D", "text": "Name what was wrong with the result; make one change; compare the new result against the previous one; keep the change or undo it.", "correct": true, "explanation": "Naming the defect gives the round a target, one change isolates the cause, the comparison shows whether it moved in the right direction, and the keep-or-undo decision makes the improvement deliberate rather than accidental."}], "correct": "D", "select": 1}, {"id": "a1-063", "domain": 1, "task_id": "1.2", "objective": "Apply task decomposition techniques to structure complex requests", "situation": "A project coordinator has listed eight tasks for a product launch and wants to work through them with Claude in a sensible order.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Work out which tasks need another task's output before they can start, since that constrains the possible orders.", "correct": true, "explanation": "Dependencies are the only hard constraint on ordering: everything else is preference. Identifying which tasks consume another's output shows what must be sequenced and what can be done in any order."}, {"letter": "B", "text": "Start with the two tasks she finds hardest, so the difficult work is finished while there is still time to spare.", "correct": false, "explanation": "Tackling the hardest work early is sound scheduling advice, but a hard task that depends on an unfinished one simply cannot be completed yet."}, {"letter": "C", "text": "Put the eight tasks in the order they appear on the launch plan she was given, and work down the list from the top.", "correct": false, "explanation": "A plan's list order usually reflects how it was written rather than what depends on what, so following it can put a task before the input it needs."}, {"letter": "D", "text": "Group the eight tasks by which colleague owns them, and work through one colleague's tasks before the next.", "correct": false, "explanation": "Ownership is useful for assigning work, not for ordering it, and grouping this way routinely places a dependent task ahead of its input."}], "correct": "A", "select": 1}, {"id": "a1-064", "domain": 1, "task_id": "1.3", "objective": "Iterate prompts to improve output quality", "situation": "A knowledge-management lead has asked Claude three times for a \"brief, on-brand\" summary of the weekly internal newsletter, and each attempt comes back at a different length and in a different register. Which two revisions to her prompt are most likely to fix this?", "question": "Select two.", "options": [{"letter": "A", "text": "Replace \"brief\" with a checkable limit: no more than 150 words in three bullets.", "correct": true, "explanation": "\"Brief\" is interpreted afresh each time, which is exactly why the length keeps moving. A word count and a bullet count are constraints Claude can meet and she can verify at a glance."}, {"letter": "B", "text": "Paste one past summary the team accepted and one it rejected, with a note on why.", "correct": true, "explanation": "and independently of the word limit: it fixes register rather than length. \"On-brand\" is undefined in the abstract, and a worked accepted-versus-rejected pair shows the voice she means far better than any adjective."}, {"letter": "C", "text": "Ask for a more formal register, since inconsistency usually reads as informality.", "correct": false, "explanation": "Formality is one register among several, not a cure for varying between registers. She would be swapping an undefined target for a slightly narrower undefined target, and length would still drift."}, {"letter": "D", "text": "Resend the identical request with a line saying the earlier attempts fell short.", "correct": false, "explanation": "Resending the same wording with a complaint attached communicates nothing new about what the summary should look like. Repeating a prompt with added emphasis is the classic non-fix for a vague specification."}, {"letter": "E", "text": "Restate the same two adjectives across a longer sentence to give them more weight.", "correct": false, "explanation": "Expanding \"brief\" and \"on-brand\" into more words leaves both terms exactly as unmeasurable as before. Length of instruction is not the same thing as precision of instruction."}], "correct": ["A", "B"], "select": 2}, {"id": "a1-065", "domain": 1, "task_id": "1.2", "objective": "Apply task decomposition techniques to structure complex requests", "situation": "A team lead has adopted a habit of splitting every request into stages. Today he needs a three-sentence note telling the team that a meeting has moved.", "question": "How should he handle it?", "options": [{"letter": "A", "text": "Ask for the note in a single request, since the task is small enough that splitting it adds steps without adding value.", "correct": true, "explanation": "Decomposition earns its keep when a task is large, has dependent stages, or needs checking before the next step. A short factual note has none of those properties, so one request is the right shape."}, {"letter": "B", "text": "Ask for several versions of the note in separate stages, then run one more stage to choose between the versions.", "correct": false, "explanation": "Generating and comparing variants is worth the effort when wording carries real weight, which a routine scheduling note does not."}, {"letter": "C", "text": "Ask for an outline of the note first, approve that outline, and only then request the three sentences themselves.", "correct": false, "explanation": "An outline of a three-sentence note is longer than the note. The checkpoint exists to catch direction errors that are expensive later, and here there is nothing expensive to catch."}, {"letter": "D", "text": "Split it into drafting, then tone review and a final polish, so the habit is applied consistently to all writing tasks.", "correct": false, "explanation": "Applying the habit uniformly ignores what breaking work up is for. Three sentences can be read and judged at a glance, so the extra stages cost time and return nothing."}], "correct": "A", "select": 1}, {"id": "a1-066", "domain": 1, "task_id": "1.2", "objective": "Apply task decomposition techniques to structure complex requests", "situation": "An analyst breaks a market review into six stages and feeds each result straight into the next, reviewing nothing until the final report is written.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "A mistake in an early stage is carried into every later stage, and the final report is built on top of it.", "correct": true, "explanation": "In a chained breakdown each stage treats the previous output as settled fact. A misread figure or a wrong assumption at stage two is not questioned again, and by stage six it is woven through the analysis and expensive to unpick."}, {"letter": "B", "text": "Later stages will repeat wording from earlier ones, so the finished report reads as somewhat repetitive.", "correct": false, "explanation": "Repetition across stages is a tidying problem that a final editing pass resolves, and it does not affect whether the conclusions hold."}, {"letter": "C", "text": "The six stages will take longer in total than simply asking for the whole market review in one single request.", "correct": false, "explanation": "Staged work does take more elapsed time, but that cost is the point of the approach and is small next to acting on a flawed report."}, {"letter": "D", "text": "The final report will be longer than the audience wants, because every stage adds material of its own.", "correct": false, "explanation": "Length is easily controlled by stating limits, and an overlong report is a presentation issue rather than a correctness one."}], "correct": "A", "select": 1}, {"id": "a1-067", "domain": 1, "task_id": "1.2", "objective": "Apply task decomposition techniques to structure complex requests", "situation": "A finance analyst has broken a budget review into stages. The next stage compares actual spend against forecast, but she has not yet received the final month's figures from two departments.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Run the comparison on the figures she has now, and then repeat the whole stage once the outstanding numbers arrive.", "correct": false, "explanation": "Running twice doubles the work and, worse, produces an interim comparison that circulates and gets quoted before the real figures arrive."}, {"letter": "B", "text": "Chase the two departments for the missing figures, because this stage cannot produce a usable answer without them.", "correct": true, "explanation": "A stage that depends on an input it does not have cannot be rescued by restructuring the request. Securing the missing figures is the step that unblocks the sequence."}, {"letter": "C", "text": "Skip the comparison stage entirely and move to the commentary, returning to the numbers if time allows later.", "correct": false, "explanation": "Commentary written without the comparison has no basis, and deferring the numbers means the conclusions are formed before the evidence exists."}, {"letter": "D", "text": "Ask Claude to estimate the two missing departments from their spending pattern over the preceding months.", "correct": false, "explanation": "Estimated departmental spend inserted into a budget review looks exactly like reported spend, and a variance driven by an estimate is indistinguishable from a real one."}], "correct": "B", "select": 1}, {"id": "a1-068", "domain": 1, "task_id": "1.1", "objective": "Create effective prompts for business and technical tasks", "situation": "A team's prompt says \"Do not use markdown in your response,\" but Claude's output still includes headings and bullet lists inconsistently.", "question": "What rewording is more effective at controlling the format?", "options": [{"letter": "A", "text": "Repeat the negative instruction \"do not use markdown\" several more times in the prompt", "correct": false, "explanation": "Repeating the same negative phrasing does not change the underlying issue that negative instructions are less effective than positive ones for controlling format."}, {"letter": "B", "text": "Ask Claude to generate the response twice and manually delete any markdown afterward", "correct": false, "explanation": "Manually stripping markdown after the fact is a workaround rather than a prompting fix, and doesn't address why the instruction isn't being followed."}, {"letter": "C", "text": "State that the response should be composed of smoothly flowing prose paragraphs instead", "correct": true, "explanation": "Telling Claude what to do instead of what not to do, such as specifying that the response should be smoothly flowing prose, is a more effective way to steer output formatting than a negative instruction."}, {"letter": "D", "text": "Set the temperature to zero so Claude follows negative instructions more reliably", "correct": false, "explanation": "Temperature controls randomness in word choice, not adherence to formatting instructions, so it would not reliably fix markdown usage."}], "correct": "C", "select": 1}, {"id": "a1-069", "domain": 1, "task_id": "1.4", "objective": "Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)", "situation": "A legal team wants Claude to draft a contract clause and needs the tone to closely match previously approved clauses in structure and phrasing. They have three approved clauses available.", "question": "What is the best way to use these in the prompt?", "options": [{"letter": "A", "text": "Include only one of the three approved clauses and drop the other two", "correct": false, "explanation": "Dropping two of the three approved examples reduces the diversity of reference material, risking a narrower or less representative match to approved style."}, {"letter": "B", "text": "Summarize the general topic of the three clauses in one sentence instead of their text", "correct": false, "explanation": "A one-sentence summary discards the actual structure and phrasing details that the legal team wants the new clause to match."}, {"letter": "C", "text": "Describe the desired tone abstractly without showing any approved clause", "correct": false, "explanation": "An abstract tone description without concrete examples is a weaker way to steer close structural and phrasing alignment than showing the actual approved clauses."}, {"letter": "D", "text": "Include the three approved clauses as examples wrapped in tags to mirror phrasing", "correct": true, "explanation": "Providing the relevant approved clauses as examples wrapped in example tags mirrors the actual use case and lets Claude closely match their structure and phrasing, the recommended approach for steering tone and format."}], "correct": "D", "select": 1}, {"id": "a1-070", "domain": 1, "task_id": "1.4", "objective": "Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)", "situation": "A writer drafting a long-form article wants Claude to avoid starting the response with phrases like 'Here is the article you requested' and instead jump straight into the content.", "question": "What is the recommended way to achieve this with current models?", "options": [{"letter": "A", "text": "Add an instruction telling Claude to respond without preamble or opening phrases", "correct": true, "explanation": "Current models support direct instructions in the system prompt telling Claude to respond without preamble and avoid phrases like 'Here is...', which is the recommended replacement for older prefill-based techniques."}, {"letter": "B", "text": "Ask the writer to manually delete the opening sentence from every single response", "correct": false, "explanation": "Manually trimming every response after the fact is unnecessary post-processing when a direct instruction can prevent the preamble in the first place."}, {"letter": "C", "text": "Provide a prefilled assistant message that begins the article for Claude to continue from", "correct": false, "explanation": "Prefilled assistant messages on the last turn are no longer supported on current models, so this is not a viable way to eliminate preambles."}, {"letter": "D", "text": "Leave the request as-is, since preamble phrasing cannot be steered at all in prompts", "correct": false, "explanation": "Preamble phrasing can be steered directly through prompt instructions, so this is not the recommended path."}], "correct": "A", "select": 1}, {"id": "a1-071", "domain": 1, "task_id": "1.4", "objective": "Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)", "situation": "A startup needs to brainstorm low-cost marketing tactics under a tight $500 budget.", "question": "Which prompt will generate the most actionable ideas?", "options": [{"letter": "A", "text": "Ask Claude to brainstorm a list of innovative and unusual marketing ideas, emphasizing wild creativity and viral potential, while deliberately ignoring budget limits or resource constraints to encourage unfiltered ideas.", "correct": false, "explanation": "Asking for innovative and unusual ideas while ignoring budget limits encourages unfiltered creativity. However, many suggestions may be too costly or resource-intensive, making them impractical for the startup's $500 constraint."}, {"letter": "B", "text": "Instruct Claude to generate a wide range of creative marketing tactics, covering everything from social media campaigns to guerrilla marketing stunts, while encouraging unconventional approaches without filtering by cost or practicality.", "correct": false, "explanation": "This prompt instructs Claude to generate ideas without filtering by cost or practicality. The resulting wide range of creative tactics may include expensive or unrealistic ideas that cannot be executed within the tight $500 budget."}, {"letter": "C", "text": "Prompt Claude to devise a list of practical, low-cost marketing ideas with a strict $500 budget, emphasizing tactics that utilize social media, local partnerships, and guerrilla marketing techniques.", "correct": true, "explanation": "By specifying a strict $500 budget and requesting practical, low-cost marketing ideas, this prompt directly constrains Claude to generate feasible tactics. Emphasizing social media, local partnerships, and guerrilla marketing techniques further focuses the output on actionable, budget-friendly channels."}, {"letter": "D", "text": "Direct Claude to find and compile a list of real-world examples of successful low-budget marketing campaigns, detailing the specific strategies, outcomes, and budget constraints, to derive actionable insights.", "correct": false, "explanation": "Directing Claude to compile real-world examples focuses on summarization of past campaigns rather than generating new, actionable ideas. The startup needs tailored, low-cost tactics, not just case studies that may not fit their specific budget and context."}], "correct": "C", "select": 1}, {"id": "a1-072", "domain": 1, "task_id": "1.2", "objective": "Apply task decomposition techniques to structure complex requests", "situation": "A team summarises each of twenty customer interviews, then works only from those summaries to decide which product problems to fix.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The summaries will use the team's own vocabulary rather than the words the interviewed customers actually used.", "correct": false, "explanation": "Vocabulary drift is a genuine annoyance and one reason to keep quotes, but it is a milder form of the detail loss described above."}, {"letter": "B", "text": "Summarising twenty interviews separately takes longer than reading through all twenty of the transcripts once.", "correct": false, "explanation": "Time spent is a cost of the approach, not a threat to the decision the team is about to make."}, {"letter": "C", "text": "The twenty summaries will vary in length, so some interviews will appear to carry more weight than the others do.", "correct": false, "explanation": "Uneven summary lengths are worth normalising, but they signal how much was said rather than distorting what the problems are."}, {"letter": "D", "text": "Details that would identify the real problem are dropped at the summary stage and cannot be recovered later.", "correct": true, "explanation": "A summary keeps what seemed important without knowing which decision it will serve. The concrete detail that distinguishes a pricing objection from a trust objection is exactly what gets compressed away, and once the team works only from summaries the loss is invisible."}], "correct": "D", "select": 1}, {"id": "a1-073", "domain": 1, "task_id": "1.3", "objective": "Iterate prompts to improve output quality", "situation": "A fundraising officer has asked twice for donor thank-you letters in her charity's own voice. Her request already gives the audience, the length and a description of the tone, and the donation details are pasted in, but both drafts read like generic marketing copy. She has a folder of letters her director approved last year.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Ask for a batch of further drafts in the hope that one of them lands on the right voice.", "correct": false, "explanation": "Generating more variations of an under-specified request is slow and leaves the gap unaddressed."}, {"letter": "B", "text": "Describe the tone she wants at greater length, listing more adjectives for the voice she is after.", "correct": false, "explanation": "More adjectives restate a description that has already failed twice to convey the voice."}, {"letter": "C", "text": "Include one or two of the approved letters in her request as examples of the voice she wants.", "correct": true, "explanation": "The spec and the context are already there, so the missing ingredient is a concrete target; a worked example shows the voice far more precisely than adjectives can describe it."}, {"letter": "D", "text": "Repeat the request and stress that the tone is the most important part of the task.", "correct": false, "explanation": "Emphasis adds no new information, and repeating a prompt unchanged rarely improves the result."}], "correct": "C", "select": 1}, {"id": "a1-074", "domain": 1, "task_id": "1.4", "objective": "Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)", "situation": "A manager asks to see the intermediate steps when working out which cost lines drove a budget overspend, but not when generating names for a new staff award.", "question": "Why is that difference reasonable?", "options": [{"letter": "A", "text": "Because a numerical answer is more likely to be wrong than a creative answer produced for the same team.", "correct": false, "explanation": "This compares how error-prone the two tasks are, when the real difference is whether an answer can be wrong in a way that inspection would expose."}, {"letter": "B", "text": "Because working through a naming task takes longer than working through a numerical one of the same size.", "correct": false, "explanation": "Relative effort is not what decides this. Steps are asked for because they can be checked, not because the task is slow."}, {"letter": "C", "text": "Because the overspend answer can be wrong in ways only the steps reveal, while a name is judged as it stands.", "correct": true, "explanation": "Analysis produces a conclusion that depends on how it was reached, so the steps are what make it checkable. A proposed name either works or does not, and there is no chain of reasoning behind it to verify."}, {"letter": "D", "text": "Because name suggestions are shorter, so the reasoning behind each of them is easier to reconstruct afterwards.", "correct": false, "explanation": "Length is incidental. Even if reconstructing the thinking behind a name were easy, doing so would not tell the manager whether the name is any good."}], "correct": "C", "select": 1}, {"id": "a1-075", "domain": 1, "task_id": "1.1", "objective": "Create effective prompts for business and technical tasks", "situation": "An operations lead must give all staff a one-page summary of a new expense policy. The full twenty-page policy document is open on her screen.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Ask for a friendly one-page summary now, and attach the policy document only if the first result looks wrong.", "correct": false, "explanation": "She has no way to tell whether the first result is wrong without the policy in front of Claude, because a fluent summary of the wrong content still reads well."}, {"letter": "B", "text": "Paste the policy text itself into the request, so the summary is built from the wording staff will be held to.", "correct": true, "explanation": "The source material has to be in the request before anything useful can be written from it. Pasting the actual policy is what makes the summary reflect the rules staff are accountable to, rather than a plausible-sounding approximation."}, {"letter": "C", "text": "Type out the handful of policy points she remembers as important, and ask Claude to expand each of them.", "correct": false, "explanation": "Working from memory reintroduces exactly the errors a summary is meant to remove, and it silently narrows the summary to the points she happened to recall."}, {"letter": "D", "text": "Ask Claude what expense rules organisations of a similar size normally apply, and adapt that answer for staff.", "correct": false, "explanation": "General practice at comparable organisations is not this organisation's policy. Staff will be held to the document, and a summary built from typical rules will differ from it in ways nobody catches."}], "correct": "B", "select": 1}, {"id": "a1-076", "domain": 1, "task_id": "1.4", "objective": "Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)", "situation": "An engineer asks Claude to analyze why a distributed system intermittently drops requests, a problem that requires working through several interacting causes before reaching a conclusion.", "question": "Which approach best fits this analytical task?", "options": [{"letter": "A", "text": "Let Claude reason through the multi-step causal analysis before its final answer", "correct": true, "explanation": "Reasoning capabilities are especially helpful for tasks involving complex multi-step reasoning, such as working through several interacting causes of an intermittent distributed systems issue before concluding."}, {"letter": "B", "text": "Ask Claude to list every possible cause without linking it to the symptom", "correct": false, "explanation": "An unconnected list of possible causes does not perform the causal analysis needed to explain how the causes interact to produce the intermittent symptom."}, {"letter": "C", "text": "Ask Claude to guess the most likely cause without examining interactions", "correct": false, "explanation": "Guessing a single cause immediately bypasses the careful analysis of interacting causes that this kind of problem requires."}, {"letter": "D", "text": "Ask Claude to answer in just one sentence without reasoning through any of the causes", "correct": false, "explanation": "A single-sentence answer skips the multi-step reasoning needed to work through interacting causes of an intermittent, hard-to-diagnose issue."}], "correct": "A", "select": 1}, {"id": "a1-077", "domain": 1, "task_id": "1.4", "objective": "Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)", "situation": "A business analyst needs to use Claude to analyze a dataset containing customer satisfaction scores, demographic information, and purchase history. The analyst wants to identify key drivers of customer dissatisfaction.", "question": "Which prompting approach would most effectively guide Claude to produce accurate, step-by-step analysis?", "options": [{"letter": "A", "text": "Instruct Claude to generate a comprehensive statistical report: perform regression analysis, hypothesis testing, and create visualizations to determine causation, and then draft a formal business report with recommendations using the provided data.", "correct": false, "explanation": "This prompt asks Claude to perform regression analysis, hypothesis testing, and create visualizations, tasks it cannot reliably execute without external tools. Additionally, it incorrectly assumes causation can be determined from observational data, which is methodologically unsound."}, {"letter": "B", "text": "Guide Claude to first examine the distribution of satisfaction scores, then explore simple correlations with demographic variables, and summarize the most likely drivers based on observed patterns, without performing deep statistical tests.", "correct": false, "explanation": "This approach avoids deep statistical tests and does not require citing specific data points, which may lead to a superficial summary that misses the true drivers. It also omits exploration of purchase history, limiting the analysis."}, {"letter": "C", "text": "Ask Claude to brainstorm a list of possible key reasons for dissatisfaction based on the data provided, relying only on qualitative judgment, then select the most plausible ones, and very briefly explain why they might be relevant to the business context.", "correct": false, "explanation": "This approach is suited for idea generation rather than rigorous data analysis, as it relies on qualitative judgment without a structured, quantitative methodology. The subjective selection process weakens its ability to accurately identify key drivers of dissatisfaction."}, {"letter": "D", "text": "Prompt Claude to conduct a step-by-step analysis: first, analyze satisfaction score distributions; next, correlate dissatisfaction with demographics and purchase history; then identify the top predictive factors, citing specific data points and patterns.", "correct": true, "explanation": "This prompt guides Claude through a logical, step-by-step process, including examining distributions, correlating dissatisfaction with both demographics and purchase history, and identifying top predictive factors. Requiring citation of specific data points and patterns ensures analytical rigor and accuracy."}], "correct": "D", "select": 1}, {"id": "a1-078", "domain": 1, "task_id": "1.4", "objective": "Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)", "situation": "A knowledge base tool asks Claude to answer users' questions using only information retrieved from company documents, and the team wants every claim traceable to the exact retrieved passage it came from.", "question": "Which technique best supports this analysis and answering task?", "options": [{"letter": "A", "text": "Have Claude answer directly from memory instead of using the retrieved documents at all", "correct": false, "explanation": "Answering from memory instead of the retrieved documents does not produce claims traceable to specific retrieved passages."}, {"letter": "B", "text": "Have Claude answer first and only fetch supporting documents afterward instead", "correct": false, "explanation": "Answering before retrieving documents reverses the order needed to ground the answer in the actual retrieved passages."}, {"letter": "C", "text": "Have Claude paraphrase only the general topic of the documents, not any passage", "correct": false, "explanation": "A general paraphrase without quoting specific passages does not give the traceability to exact source text that the team needs."}, {"letter": "D", "text": "Have Claude quote the specific passages from the retrieved documents before answering", "correct": true, "explanation": "Grounding responses in quotes, having Claude quote relevant passages before answering, helps produce traceable, source-grounded answers, which fits the requirement that every claim be traceable to a passage."}], "correct": "D", "select": 1}, {"id": "a1-079", "domain": 1, "task_id": "1.4", "objective": "Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)", "situation": "A marketing team wants Claude to draft product descriptions in a specific brand voice and needs the tone and structure to stay consistent across many descriptions.", "question": "What is the most reliable way to steer this drafting task?", "options": [{"letter": "A", "text": "Describe the brand voice in one adjective and let Claude infer the rest", "correct": false, "explanation": "A one-word tone descriptor without examples leaves too much of the format and structure to be inferred rather than steered."}, {"letter": "B", "text": "Give no examples and repeat the word 'consistent' several times instead", "correct": false, "explanation": "Repeating a vague adjective like 'consistent' does not give Claude concrete examples of the actual tone, format, or structure desired."}, {"letter": "C", "text": "Provide three to five diverse examples wrapped in tags that mirror the brand voice", "correct": true, "explanation": "Using three to five relevant, diverse examples wrapped in example tags, few-shot prompting, is one of the most reliable ways to steer output format, tone, and structure."}, {"letter": "D", "text": "Provide a single long example and ask Claude to make all future ones roughly similar", "correct": false, "explanation": "A single example risks Claude picking up unintended, overly narrow patterns instead of generalizing the brand voice across varied product types."}], "correct": "C", "select": 1}, {"id": "a1-080", "domain": 1, "task_id": "1.1", "objective": "Create effective prompts for business and technical tasks", "situation": "A communications officer pastes a ten-page partner agreement into the chat and types her instructions directly underneath it, with nothing marking where the pasted document ends and her request begins.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The conversation will become too long to continue, so she will have to start the whole exchange again shortly.", "correct": false, "explanation": "A ten-page agreement is well within what a single chat handles, and length is a separate concern from failing to separate instructions from source material."}, {"letter": "B", "text": "The signed agreement will be reformatted as it is pasted, so its clauses arrive in a different order from the original.", "correct": false, "explanation": "Some layout is lost when text is pasted, but clause order is preserved, and reordering is not what makes an unlabelled paste risky."}, {"letter": "C", "text": "Claude will answer only from the closing paragraphs, because material earlier in a long message is set aside.", "correct": false, "explanation": "Claude works from the whole message rather than only its final paragraphs, so this misstates how a long pasted document is handled."}, {"letter": "D", "text": "Claude may read wording inside the agreement as directions to follow, and act on it instead of her actual request.", "correct": true, "explanation": "Contracts contain imperative language such as 'the supplier shall notify' or 'provide a summary within ten days'. With no boundary between the pasted material and her instructions, that wording can be taken as part of the task, and the output drifts away from what she asked for."}], "correct": "D", "select": 1}, {"id": "a1-081", "domain": 1, "task_id": "1.4", "objective": "Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)", "situation": "A product team wants to brainstorm features for a new fitness app. They want a wide range of creative ideas without early criticism.", "question": "Which prompting strategy fosters the most innovative brainstorming session with Claude?", "options": [{"letter": "A", "text": "Guide Claude to compile a list of the 10 most prevalent features in top-rated fitness apps by referencing common market knowledge, then use that list as a foundation to brainstorm derivative features, ensuring the team has a concrete starting point for ideation.", "correct": false, "explanation": "This approach starts by compiling existing prevalent features, which anchors the session in conventional ideas rather than pushing beyond them. Brainstorming derivative features from a known list limits true innovation and lacks the explicit deferral of judgment essential for creative brainstorming."}, {"letter": "B", "text": "Ask Claude to develop a user survey to gather preferences for fitness app features, including crafting open-ended and multiple-choice questions, and use the responses to spark creative feature ideas during a team brainstorming session focused on novelty.", "correct": false, "explanation": "Developing a user survey is a data-gathering method that collects preferences rather than generating novel concepts. Even if responses later inspire ideas, the process shifts focus to validation and user input, which stifles the spontaneous, judgment-free atmosphere required for truly innovative brainstorming."}, {"letter": "C", "text": "Instruct Claude to conduct a competitive analysis: identify major fitness apps, catalog their features, and present findings in a structured matrix comparing feature sets, noting which are standard versus differentiators, and use the matrix to spark ideas for novel features.", "correct": false, "explanation": "Competitive analysis is an analytical, research-oriented task that catalogs existing features and competitor differentiators. Although it may inspire ideas, the structured, comparative focus constrains open-ended creative generation and does not embody the unfettered ideation needed for innovation."}, {"letter": "D", "text": "Prompt Claude to brainstorm 20 innovative and unconventional feature ideas for the fitness app, explicitly encouraging far-fetched concepts and deferring feasibility analysis; emphasize that quantity and creativity are the primary goals of this exercise.", "correct": true, "explanation": "This prompt explicitly requests a large quantity of unconventional, far-fetched ideas and defers feasibility analysis, directly encouraging divergent thinking. By prioritizing creativity and volume over early criticism, it aligns perfectly with proven brainstorming principles that foster innovation."}], "correct": "D", "select": 1}, {"id": "a1-082", "domain": 1, "task_id": "1.4", "objective": "Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)", "situation": "A programme officer needs to answer the question 'what are our reporting obligations for this funding stream'.", "question": "What should she settle first?", "options": [{"letter": "A", "text": "How much reasoning to ask for, since obligations of this kind are usually explained across several steps.", "correct": false, "explanation": "Asking for more explanation makes the answer easier to follow without making it any more grounded in the right source."}, {"letter": "B", "text": "Whether the answer must come from her own funding documents or from material published outside them.", "correct": true, "explanation": "The two cases call for different approaches. If her grant agreement governs, the documents should be supplied and the answer confined to them; if the obligation is set publicly, the task becomes a research one where cited sources matter. Everything else follows from that choice."}, {"letter": "C", "text": "Whether to ask the question in one go or split it across several smaller and more specific questions.", "correct": false, "explanation": "Splitting the question is a useful technique for large tasks, but it does not determine what the answer will be based on."}, {"letter": "D", "text": "How long the answer should be and which colleagues it will need to be circulated to once it is finished.", "correct": false, "explanation": "Length and audience shape how the answer is presented, and they can be decided once she knows where the answer is coming from."}], "correct": "B", "select": 1}, {"id": "a1-083", "domain": 1, "task_id": "1.1", "objective": "Create effective prompts for business and technical tasks", "situation": "A prompt engineering team wants a quick sanity check to decide whether a draft prompt is clear enough before sending it to Claude in production.", "question": "What check reflects the recommended \"golden rule\" for prompt clarity?", "options": [{"letter": "A", "text": "Show it to a colleague with minimal context and see if they could follow it correctly", "correct": true, "explanation": "The golden rule for prompt clarity is to show the prompt to a colleague with minimal context on the task and ask them to follow it; if they'd be confused, Claude will be too."}, {"letter": "B", "text": "Ask Claude itself to guess what the prompt author must have originally intended", "correct": false, "explanation": "Having Claude guess the original intent does not test whether the instructions themselves are clear and specific enough for a new reader to follow."}, {"letter": "C", "text": "Send the same prompt to several people and average their reactions numerically", "correct": false, "explanation": "Averaging numerical reactions from multiple people is not the documented golden rule and adds unnecessary process without checking for actual confusion or ambiguity."}, {"letter": "D", "text": "Translate the prompt into another language and back to test for consistency", "correct": false, "explanation": "Round-trip translation tests linguistic stability, not whether the instructions are explicit and unambiguous to someone unfamiliar with the task."}], "correct": "A", "select": 1}, {"id": "a1-084", "domain": 1, "task_id": "1.1", "objective": "Create effective prompts for business and technical tasks", "situation": "An events manager is preparing an attendee FAQ. She has last year's FAQ, the signed contract for this year's venue, and an email thread with the caterer.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Ask Claude to draft the FAQ from last year's version, then check it against the contract and the caterer emails.", "correct": false, "explanation": "Drafting from the superseded document first anchors the FAQ on last year's arrangements, and checking afterwards catches only the differences she happens to look for."}, {"letter": "B", "text": "Paste all three documents together and ask Claude to work out which of the details in them are the most current ones.", "correct": false, "explanation": "Nothing in the documents states which is newest or which governs, so this hands a decision she alone can make to a source that cannot see the contractual hierarchy."}, {"letter": "C", "text": "Summarise each of the three documents on her own, and paste only those summaries so the request stays compact.", "correct": false, "explanation": "Her own summaries drop the specifics attendees ask about, such as exact times and access rules, and the precedence problem is still unresolved."}, {"letter": "D", "text": "Decide which document is authoritative this year, and label each as she pastes it so Claude knows which one wins.", "correct": true, "explanation": "The three sources disagree by design: the contract supersedes last year's FAQ, and the email thread may supersede both on catering. Establishing precedence and labelling each source is what stops last year's details being presented as this year's."}], "correct": "D", "select": 1}, {"id": "a1-085", "domain": 1, "task_id": "1.4", "objective": "Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)", "situation": "A team wants Claude to research an open-ended market question, tracking multiple competing explanations and its confidence in each as it gathers evidence over many searches.", "question": "What approach best supports this research task?", "options": [{"letter": "A", "text": "Run one broad search and answer directly from that result with no further checks", "correct": false, "explanation": "A single unverified search skips the source verification that complex research questions need before drawing conclusions."}, {"letter": "B", "text": "Avoid recording any intermediate findings so the final answer reflects only the very last search", "correct": false, "explanation": "Discarding intermediate findings loses the calibration and transparency that a persisted research notes file is meant to provide across a long research task."}, {"letter": "C", "text": "Develop competing hypotheses, track confidence in notes, and self-critique while searching", "correct": true, "explanation": "For complex, open-ended research, Claude should develop competing hypotheses, track confidence in progress notes, and self-critique regularly to work through evidence methodically."}, {"letter": "D", "text": "Commit to a single explanation immediately so later searches confirm that theory", "correct": false, "explanation": "Committing to one explanation immediately biases the research toward confirming a single theory instead of weighing competing hypotheses against gathered evidence."}], "correct": "C", "select": 1}, {"id": "a1-086", "domain": 1, "task_id": "1.1", "objective": "Create effective prompts for business and technical tasks", "situation": "A facilities manager wants Claude to turn her site-visit notes into short summaries that engineers read on their phones between jobs, so each summary must stay under 150 words. She will make this same request many times with different notes.", "question": "How should she express the length limit?", "options": [{"letter": "A", "text": "Ask for the shortest summary she can get, and then cut anything back that still runs over the word limit.", "correct": false, "explanation": "An unspecified target produces inconsistent lengths and moves the work back onto the manager."}, {"letter": "B", "text": "State the 150-word limit and the reason for it: engineers read the summaries on a phone between jobs.", "correct": true, "explanation": "Giving the reason behind a constraint lets Claude apply the underlying intent sensibly to new notes, rather than treating 150 words as an arbitrary rule to satisfy."}, {"letter": "C", "text": "State the 150-word limit on its own, as any explanation around it only lengthens her request.", "correct": false, "explanation": "The limit alone works, but without the reason Claude cannot judge what to keep when notes vary."}, {"letter": "D", "text": "Repeat the 150-word limit in capital letters at the start and again at the end of the request.", "correct": false, "explanation": "Emphasis and repetition do not add information; the constraint was already clear the first time."}], "correct": "B", "select": 1}, {"id": "a1-087", "domain": 1, "task_id": "1.3", "objective": "Iterate prompts to improve output quality", "situation": "After rewriting an instruction to say \"write in flowing prose, avoid bullet points,\" a team still finds Claude's responses returning to bulleted lists in some outputs. They notice their own prompt itself is written using bullet points and headers throughout.", "question": "Which additional iteration is most likely to help?", "options": [{"letter": "A", "text": "Increase the temperature parameter so Claude's output becomes more random, discouraging the rigid structure that leads to bullet points.", "correct": false, "explanation": "Increasing temperature affects randomness in word choice and phrasing, but it does not discourage structural elements like lists or bullet points, which are determined more by the prompt's formatting."}, {"letter": "B", "text": "Add even more bullet points to the prompt, perhaps using numbered lists, to make the instruction to avoid bullets impossible for Claude to overlook.", "correct": false, "explanation": "Adding more bullet points would reinforce, not reduce, the bullet-point pattern the team is trying to eliminate, making it even harder for Claude to avoid that structure."}, {"letter": "C", "text": "Move the anti-bullet-point instruction into a separate, later user turn right before the final request, so it is more recent in the conversation.", "correct": false, "explanation": "The core issue is not instruction recency but the prompt's own style. Moving the anti-bullet instruction to a later turn does not address the underlying influence of the prompt's bulleted formatting."}, {"letter": "D", "text": "Remove the bullet points and headers from the prompt itself so the prompt's own style matches the desired flowing-prose output.", "correct": true, "explanation": "Anthropic notes that the formatting style used in the prompt influences Claude's response style, so removing bullet points and headers from the prompt can help the model adopt the desired flowing-prose style."}], "correct": "D", "select": 1}, {"id": "a1-088", "domain": 1, "task_id": "1.4", "objective": "Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)", "situation": "A legal assistant needs to summarize a 50-page contract, highlighting obligations, risks, and key dates.", "question": "Which prompting strategy yields the most precise and actionable summary?", "options": [{"letter": "A", "text": "Instruct Claude to review the contract and draft a summary focusing on what it determines to be the central points, including obligations and dates it selects based on its own analysis, organizing the output with headings it deems appropriate.", "correct": false, "explanation": "This strategy leaves too much to Claude's discretion, which may result in missing key obligations, risks, or dates, or organizing them in an unstructured way. Without explicit guidance, the summary may not be sufficiently precise or actionable for legal purposes."}, {"letter": "B", "text": "Ask Claude to read the contract and produce an abstract that emphasizes the overall intent and identifies the key parties, describing the general nature of the agreement without going into specific obligations, risk allocations, or dates.", "correct": false, "explanation": "Like option B, this produces a general abstract that omits the critical specifics. It does not extract the actionable details (obligations, risks, dates) necessary for the assistant's task."}, {"letter": "C", "text": "Guide Claude to methodically parse the contract: first list party obligations, then identify risk allocation clauses (e.g., indemnities, limitations), then extract key dates into a timeline, and organize under these headings.", "correct": true, "explanation": "This prompt provides a structured, methodical approach that ensures all required elements (obligations, risks, dates) are systematically extracted and organized. It directly addresses the need for precision and actionability by breaking down the task into clear steps and explicit headings."}, {"letter": "D", "text": "Prompt Claude to generate a high-level overview that broadly covers the contract's purpose, parties, and general terms, while explicitly avoiding detailed extractions such as specific obligations, risk allocation clauses, and key dates.", "correct": false, "explanation": "This approach explicitly avoids the detailed extractions required, making it unsuitable for a precise and actionable summary. It will not provide the specific obligations, risks, or dates needed for legal analysis."}], "correct": "C", "select": 1}, {"id": "a1-089", "domain": 1, "task_id": "1.2", "objective": "Apply task decomposition techniques to structure complex requests", "situation": "An HR team is using Claude to rework the onboarding checklist for four departments at once. Which two practices will make the staged approach hold together?", "question": "Select two.", "options": [{"letter": "A", "text": "Begin from the final recommendation and ask Claude to justify it across the four teams.", "correct": false, "explanation": "Starting from a conclusion invites the evidence to be arranged to fit it, which is the reverse of the order a decomposed analysis should follow."}, {"letter": "B", "text": "Put a human check after the stage where an unnoticed error would carry into later work.", "correct": true, "explanation": "This addresses a different risk from the hand-off format: checkpoints stop a wrong fact being built upon, which is the main cost of chaining steps."}, {"letter": "C", "text": "Keep all four departments inside one prompt so that the wording stays perfectly aligned.", "correct": false, "explanation": "One prompt covering four departments is what caused the shallow result in the first place. Consistency is better handled by reusing a stated format across stages."}, {"letter": "D", "text": "State the concrete output each stage must hand to the next, such as a named gap list.", "correct": true, "explanation": "Naming the artefact each stage produces gives the next stage something concrete to consume, instead of a vague hand-off the team has to interpret."}, {"letter": "E", "text": "Ask Claude to pick the stages itself and work straight through them without showing them.", "correct": false, "explanation": "Hidden stages remove the team's ability to catch a wrong plan early, and produce a long output whose reasoning cannot be inspected or corrected."}], "correct": ["B", "D"], "select": 2}, {"id": "a1-090", "domain": 1, "task_id": "1.1", "objective": "Create effective prompts for business and technical tasks", "situation": "A school communications officer needs a short newsletter item about a new pick-up procedure.", "question": "Which request is most likely to produce copy she can publish after light editing?", "options": [{"letter": "A", "text": "Write a short, engaging newsletter item announcing the change to the new school pick-up procedure that starts next term.", "correct": false, "explanation": "Short and engaging leave the length open and the content unspecified, so what comes back is a guess at what the new procedure involves."}, {"letter": "B", "text": "Using the procedure pasted below, write 120 words for parents covering when it starts, where to queue and who to ask.", "correct": true, "explanation": "It supplies the source material, names the reader, fixes the length, and lists the questions the item must answer, which is what makes the output publishable rather than merely well written."}, {"letter": "C", "text": "Summarise the new pick-up procedure for our newsletter, keeping the tone warm and the wording as simple as possible.", "correct": false, "explanation": "Asking for a summary of a procedure that has not been provided leaves the substance to invention, and tone guidance does not fill that gap."}, {"letter": "D", "text": "Write a newsletter item about our new pick-up procedure that parents will find clear, friendly and easy to follow.", "correct": false, "explanation": "Clear and friendly describe a feeling rather than a specification, and with no procedure supplied the details of the change have to be invented."}], "correct": "B", "select": 1}, {"id": "a1-091", "domain": 1, "task_id": "1.2", "objective": "Apply task decomposition techniques to structure complex requests", "situation": "In an earlier stage a bid team agreed three criteria for evaluating partner organisations. They now start the scoring stage in a fresh conversation.", "question": "What should they include in the scoring request?", "options": [{"letter": "A", "text": "A note that criteria were agreed earlier, and a request that Claude apply the same ones it used in that stage.", "correct": false, "explanation": "A reference to work done elsewhere cannot be relied on to carry the agreed criteria into this stage, so the scoring may proceed against criteria that were never the ones the team settled."}, {"letter": "B", "text": "The name of the earlier conversation, so that the reasoning from that stage is picked up and continued here.", "correct": false, "explanation": "Naming a previous conversation is not a dependable way to bring its contents forward, so the stage can run without the criteria it is meant to apply."}, {"letter": "C", "text": "A request to propose fresh criteria for scoring partners, which they will then compare against the three they agreed.", "correct": false, "explanation": "Regenerating criteria reopens a decision the team already made, and comparing two sets adds work without improving the scores."}, {"letter": "D", "text": "The three agreed criteria written out in full, together with the partner information the scoring is to be based on.", "correct": true, "explanation": "A new conversation starts with nothing from the previous stage, so anything the next step depends on has to be carried across explicitly. Restating the criteria and supplying the partner information makes the scoring reproducible and traceable."}], "correct": "D", "select": 1}, {"id": "a1-092", "domain": 1, "task_id": "1.4", "objective": "Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)", "situation": "A brand team is running a brainstorming session for a new ad campaign and wants Claude to keep generating fresh concepts rather than settling on the first workable idea.", "question": "Which instruction best supports this goal?", "options": [{"letter": "A", "text": "Ask Claude to keep generating distinct concepts after producing a workable one", "correct": true, "explanation": "Explicitly asking for continued generation of distinct concepts keeps a brainstorming session open rather than converging on the first workable answer."}, {"letter": "B", "text": "Ask Claude to rate its own first concept highly to bias later comparisons", "correct": false, "explanation": "Pre-rating the first concept highly biases the comparison against later ideas rather than encouraging open generation of fresh concepts."}, {"letter": "C", "text": "Ask Claude to stop once it produces a concept fitting the campaign brief nicely", "correct": false, "explanation": "Stopping at the first workable concept defeats the purpose of a brainstorming session meant to surface a range of fresh ideas."}, {"letter": "D", "text": "Ask Claude to expand only on the first concept in greater detail", "correct": false, "explanation": "Elaborating only on the first concept narrows the session into refining one idea instead of generating a variety of new ones."}], "correct": "A", "select": 1}, {"id": "a1-093", "domain": 1, "task_id": "1.2", "objective": "Apply task decomposition techniques to structure complex requests", "situation": "A customer-success manager writes one request asking Claude to summarise thirty support tickets, rank the underlying problems, draft an apology to the largest affected account, and outline a slide for the leadership review. Every part comes back thin.", "question": "How should she restructure the request?", "options": [{"letter": "A", "text": "Keep the single request as it stands and attach the tickets as a file so that there is more to work from.", "correct": false, "explanation": "The tickets were never the constraint here; the request is doing four different jobs at once."}, {"letter": "B", "text": "Send the same four-part request again, this time asking for much more detail in each part.", "correct": false, "explanation": "The problem is that one request carries four jobs, so repeating it with more emphasis changes little."}, {"letter": "C", "text": "Keep the single request but split the tickets into three batches and send it once per batch.", "correct": false, "explanation": "This multiplies the overloaded request rather than separating the work it is trying to do at once."}, {"letter": "D", "text": "Establish the summary and the ranking first, then ask for the apology and the slide as separate steps.", "correct": true, "explanation": "The four parts are one analysis feeding three different deliverables; settling the analysis first gives each later step something solid and specific to work from."}], "correct": "D", "select": 1}, {"id": "a1-094", "domain": 1, "task_id": "1.2", "objective": "Apply task decomposition techniques to structure complex requests", "situation": "A procurement lead has four supplier proposals and must recommend one to her director.", "question": "Which sequence of stages should she work through with Claude?", "options": [{"letter": "A", "text": "Summarise each proposal against them; compare the summaries side by side; write the recommendation; agree the criteria that matter.", "correct": false, "explanation": "Summarising against criteria that have not been agreed is not possible, and settling them last leaves the completed comparison resting on nothing."}, {"letter": "B", "text": "Compare the proposals side by side; agree the criteria that matter; write the recommendation; summarise each proposal against them.", "correct": false, "explanation": "Comparing before criteria exist means each proposal is described on whatever terms it chose for itself, and the summaries arrive after the decision is already written."}, {"letter": "C", "text": "Write the recommendation; agree the criteria that matter; summarise each proposal against them; compare the summaries side by side.", "correct": false, "explanation": "Writing the recommendation first turns every later stage into a search for support, which is how a preferred supplier gets justified rather than chosen."}, {"letter": "D", "text": "Agree the criteria that matter; summarise each proposal against them; compare the summaries side by side; write the recommendation.", "correct": true, "explanation": "Criteria fix what counts as a good proposal, the per-proposal summaries then say the same things about each one, the comparison becomes like-for-like, and the recommendation follows from it."}], "correct": "D", "select": 1}, {"id": "a1-095", "domain": 1, "task_id": "1.1", "objective": "Create effective prompts for business and technical tasks", "situation": "A procurement officer needs a supplier comparison that must fit on one page and must not name any supplier outside the approved shortlist.", "question": "Where do these two limits belong?", "options": [{"letter": "A", "text": "After the first draft arrives, so the comparison is not narrowed before she has seen the full range of options.", "correct": false, "explanation": "Seeing the wider market may be useful for her own understanding, but it is a separate task. Applied here it produces a draft that has to be rebuilt around a different set of suppliers."}, {"letter": "B", "text": "In the original request, since both limits change what belongs in the comparison, not merely how it is presented.", "correct": true, "explanation": "Both constraints determine which suppliers are compared and how much can be said about each, so stating them up front shapes the analysis instead of leaving her to reconstruct it by hand."}, {"letter": "C", "text": "In neither place, since limits of this kind are best handled by trimming and correcting the draft she gets back.", "correct": false, "explanation": "Trimming to one page after the fact means deciding what to cut without the original weighting, and the shortlist rule is not a trimming problem at all."}, {"letter": "D", "text": "Only the page limit belongs in the request; the shortlist restriction is simpler to apply while editing the draft.", "correct": false, "explanation": "Splitting the limits leaves the harder one to manual editing, and removing an off-shortlist supplier from a comparison changes the surrounding analysis rather than a single line."}], "correct": "B", "select": 1}, {"id": "a1-096", "domain": 1, "task_id": "1.2", "objective": "Apply task decomposition techniques to structure complex requests", "situation": "One stage of a service review is described as 'look into how customers feel about our support'. It keeps returning broad material the team cannot act on.", "question": "How should the stage be redefined?", "options": [{"letter": "A", "text": "As a stage answering one question, such as which support issues drive customers to contact the team twice.", "correct": true, "explanation": "A stage is only well formed when it has a single answerable question and a recognisable finishing point. Naming the specific behaviour to explain gives the stage a boundary and gives the team a result they can act on."}, {"letter": "B", "text": "As two stages of the same broad description, so the material arrives in halves that are easier to read through.", "correct": false, "explanation": "Halving a vague stage produces two vague stages, since the problem is the absence of a specific question rather than the volume."}, {"letter": "C", "text": "As a stage that gathers as much material as possible on support quality, which the team then filters down itself.", "correct": false, "explanation": "Collecting more of the same material moves the filtering work onto the team without making the stage's purpose any clearer."}, {"letter": "D", "text": "As a stage covering customer feeling about support, pricing and product together, giving the findings context.", "correct": false, "explanation": "Widening the scope makes the output broader still, and the team already cannot act on what the narrower version returns."}], "correct": "A", "select": 1}, {"id": "a1-097", "domain": 1, "task_id": "1.1", "objective": "Create effective prompts for business and technical tasks", "situation": "An operations manager writes a prompt for a weekly incident summary that the executive team reads every Monday, and a partner firm receives the same summary. Which two elements should the prompt include to make the output usable without rework?", "question": "Select two.", "options": [{"letter": "A", "text": "A request that the summary be written so that it sounds authoritative and well informed.", "correct": false, "explanation": "Asking for an authoritative tone shapes how the text sounds, not what it contains, and can encourage confident phrasing over accurate reporting of the week."}, {"letter": "B", "text": "A note that she is short of time this week and would appreciate a quick turnaround here.", "correct": false, "explanation": "Claude's output does not change because the requester is busy. This adds words to the prompt without adding any requirement the draft can be judged against."}, {"letter": "C", "text": "The output shape she needs: one page, three headed sections, and no more than six bullets.", "correct": true, "explanation": "Stating the format and its limits up front is what removes the reformatting step. \"One page, three sections, six bullets\" is checkable, unlike \"keep it short\"."}, {"letter": "D", "text": "An instruction telling Claude to use its own judgment about what executives care about.", "correct": false, "explanation": "Delegating the specification back to Claude is the opposite of describing intent. It guarantees a plausible guess rather than the summary this audience actually needs."}, {"letter": "E", "text": "Why customer names must be excluded, since a partner firm also reads it.", "correct": true, "explanation": "Giving the reason behind a constraint lets Claude generalise it: knowing an external partner reads the summary, it will also handle account numbers and site names sensibly."}], "correct": ["C", "E"], "select": 2}, {"id": "a1-098", "domain": 1, "task_id": "1.4", "objective": "Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)", "situation": "A marketing manager needs to draft a persuasive email campaign to launch a new product. The email should capture attention, highlight benefits, and include a call to action.", "question": "Which prompting approach is best?", "options": [{"letter": "A", "text": "Prompt Claude to draft an email announcing the new product, leaving the structure and content entirely to the model's discretion without providing any specific guidelines on audience, tone, or key product benefits to highlight.", "correct": false, "explanation": "This approach provides no guidance on audience, tone, or key benefits, leaving the output too open-ended. Without clear direction, the email may lack persuasive elements and fail to meet marketing goals."}, {"letter": "B", "text": "Instruct Claude to write a thorough product description report covering all features, specifications, technical details, and pricing info, formatted as a formal document that is suitable for internal stakeholders rather than a marketing audience.", "correct": false, "explanation": "This instruction results in a formal internal report, not a marketing email. It emphasizes technical specifications rather than attention-grabbing copy, benefits, and a call to action."}, {"letter": "C", "text": "Guide Claude to create a marketing email: first, write an attention-grabbing subject line; then, open with a hook; next, present three key benefits with persuasive language; and finally, include a clear call to action linked to the product page.", "correct": true, "explanation": "This prompt gives step-by-step guidance tailored to marketing email creation, ensuring inclusion of a subject line, hook, benefits, and call to action. It's a structured approach that yields a persuasive, complete email."}, {"letter": "D", "text": "Ask Claude to produce several email drafts with distinct tones (formal, casual, urgent), then assess each draft's persuasiveness, clarity, and brand alignment, and choose the best one for final edits, and then iterate if needed.", "correct": false, "explanation": "While producing multiple drafts can be useful for exploration, it is less efficient for a single drafting task. Option C provides a more direct and structured method to get a well-crafted email in one go."}], "correct": "C", "select": 1}, {"id": "a1-099", "domain": 1, "task_id": "1.2", "objective": "Apply task decomposition techniques to structure complex requests", "situation": "A compliance officer must update an internal policy to reflect a new regulation.", "question": "Which sequence of stages should she work through?", "options": [{"letter": "A", "text": "Pull the obligations out of the current policy; list what the regulation now requires; mark the clauses affected; draft the replacement wording.", "correct": true, "explanation": "Establishing what the policy says today and what the regulation now demands gives the two lists that make a gap visible, the affected clauses fall out of that comparison, and only then is there something specific to redraft."}, {"letter": "B", "text": "Mark the clauses affected; draft the replacement wording; pull the obligations out of the current policy; list what the regulation now requires.", "correct": false, "explanation": "Clauses cannot be marked as affected until both the existing obligations and the new requirements have been set out to compare."}, {"letter": "C", "text": "List what the regulation now requires; draft the replacement wording; pull the obligations out of the current policy; mark the clauses affected.", "correct": false, "explanation": "Drafting immediately after reading the regulation ignores what the policy already covers, so it produces new text that duplicates or contradicts existing clauses."}, {"letter": "D", "text": "Draft the replacement wording; list what the regulation now requires; mark the clauses affected; pull the obligations out of the current policy.", "correct": false, "explanation": "Replacement wording written before either the current position or the new requirement is known has nothing to replace and nothing to satisfy."}], "correct": "A", "select": 1}, {"id": "a1-100", "domain": 1, "task_id": "1.1", "objective": "Create effective prompts for business and technical tasks", "situation": "A benefits administrator asks Claude to answer staff questions using the benefits handbook she pastes in. Some of the questions concern situations the handbook simply does not cover.", "question": "What should her request say about those?", "options": [{"letter": "A", "text": "That any question the handbook does not settle should be skipped, returning only the ones it covers clearly.", "correct": false, "explanation": "Silently dropping questions leaves staff with no reply and leaves her with no list of the gaps she needs to escalate."}, {"letter": "B", "text": "That every question should receive a full answer, drawing on general knowledge of employee benefits where needed.", "correct": false, "explanation": "General benefits practice is not this employer's scheme. An answer built that way looks authoritative and can commit the organisation to entitlements it does not offer."}, {"letter": "C", "text": "That uncovered questions should still be answered, with a note attached asking the reader to confirm the details.", "correct": false, "explanation": "A caveat attached to an invented answer still puts a specific figure or rule in front of the reader, and readers act on the answer rather than the caveat."}, {"letter": "D", "text": "That where the handbook is silent, Claude should say so plainly instead of supplying an answer of its own.", "correct": true, "explanation": "Naming the boundary of the source turns a gap into visible information she can act on, and it prevents plausible general knowledge being presented as her employer's policy."}], "correct": "D", "select": 1}, {"id": "a1-101", "domain": 1, "task_id": "1.3", "objective": "Iterate prompts to improve output quality", "situation": "A prompt asks Claude to solve multi-step math word problems, but a review of outputs shows occasional arithmetic slips that go uncaught. The team wants to iterate on the prompt to catch these errors before the final answer is returned, without switching to a different evaluation pipeline.", "question": "Which addition to the prompt best addresses this?", "options": [{"letter": "A", "text": "Ask Claude to supply only the final numeric answer without any intermediate steps, so no arithmetic is shown to check.", "correct": false, "explanation": "Suppressing intermediate work hides the reasoning chain, preventing both the model and reviewers from spotting where arithmetic errors happen. It eliminates any chance for self-correction and does not add a verification mechanism."}, {"letter": "B", "text": "Remove all mention of the word \"steps\" from the prompt so Claude solves the problem in a single pass.", "correct": false, "explanation": "Removing instruction to show steps discourages structured, step-by-step reasoning that helps maintain accuracy on multi-step math problems. Without explicit reasoning, arithmetic slips are more likely to occur and go unnoticed."}, {"letter": "C", "text": "Append a step instructing Claude to verify its answer against the problem constraints before giving the final answer.", "correct": true, "explanation": "Adding an explicit self-verification step prompts Claude to check its solution against the original constraints, which can catch arithmetic errors before the final output. This technique directly addresses the need for error detection without altering the evaluation pipeline."}, {"letter": "D", "text": "Instruct Claude to solve the problem twice using the exact same approach both times, then submit the first result.", "correct": false, "explanation": "Repeating the same approach twice does not introduce a genuine check, because identical reasoning is likely to reproduce the same mistake. Since only the first result is used, the second computation provides no error-catching benefit."}], "correct": "C", "select": 1}, {"id": "a1-102", "domain": 1, "task_id": "1.1", "objective": "Create effective prompts for business and technical tasks", "situation": "A member services officer must explain a change to membership fees. The explanation goes to members of fifteen years and to people who joined last month.", "question": "Which request is most likely to produce copy she can use?", "options": [{"letter": "A", "text": "Ask for two versions of the notice, stating what each group already knows about fees and what each has to do.", "correct": true, "explanation": "Naming each audience and what it already knows lets the request specify different starting points and different required actions, which is what actually separates copy that lands from copy that has to be rewritten."}, {"letter": "B", "text": "Ask for one notice aimed at long-standing members, since newer members will accept whatever the notice tells them.", "correct": false, "explanation": "Writing for the group with the most history assumes background the newest members do not have, and it treats their confusion as acceptable rather than as a cost."}, {"letter": "C", "text": "Ask for one notice in neutral wording, and let each reader work out for themselves which parts of it apply to them.", "correct": false, "explanation": "A notice that serves neither group well is the usual result: long-standing members wade through explanations they do not need, and new members meet references they cannot place."}, {"letter": "D", "text": "Ask for a single notice, plus a glossary at the end explaining every term that appears in the fee schedule.", "correct": false, "explanation": "A glossary bolts definitions onto the end without changing the body of the notice, so the parts that assume familiarity still assume it."}], "correct": "A", "select": 1}, {"id": "a1-103", "domain": 1, "task_id": "1.2", "objective": "Apply task decomposition techniques to structure complex requests", "situation": "Three colleagues divide a customer research report between them. Each works on their own section in their own separate chat, and the sections are combined the day before the deadline.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "Each section will be about the same length, giving the combined report an unhelpfully even balance across topics.", "correct": false, "explanation": "Even section lengths are a cosmetic outcome, and easily adjusted once the material is in front of them."}, {"letter": "B", "text": "The three colleagues will each spend longer on their section than they would have spent working all together.", "correct": false, "explanation": "Splitting the work reduces individual effort rather than increasing it, and effort is not what fails here."}, {"letter": "C", "text": "The sections will rest on different assumptions and definitions, and the combined report will not hold together.", "correct": true, "explanation": "Nothing is shared between the three chats, so each section quietly settles its own time period, customer segment definition and success measure. Those conflicts only surface at the merge, when there is a day left to resolve them."}, {"letter": "D", "text": "The finished report will contain three different writing styles, which readers will notice as they move through it.", "correct": false, "explanation": "Inconsistent voice is real and worth an editing pass, but it is visible on the surface and does not undermine the report's conclusions."}], "correct": "C", "select": 1}, {"id": "a1-104", "domain": 1, "task_id": "1.1", "objective": "Create effective prompts for business and technical tasks", "situation": "A project manager's prompt says only \"never use the word delay in the status report\". Claude complies literally but the reports still alarm the sponsor. Which two revisions would make the instruction work as intended?", "question": "Select two.", "options": [{"letter": "A", "text": "Say how slipped dates should be reported, such as a revised date and the cause.", "correct": true, "explanation": "A prohibition alone leaves Claude to guess a substitute. Defining the positive replacement is an independent fix that specifies the desired output rather than the reason for it."}, {"letter": "B", "text": "Explain that the sponsor escalates on schedule language, so Claude applies the intent widely.", "correct": true, "explanation": "Supplying the reason turns a single banned word into a general principle, so Claude also handles \"slippage\", \"behind plan\" and similar phrasing the way the sponsor needs."}, {"letter": "C", "text": "Ask Claude to confirm that it has understood the instruction before it writes anything.", "correct": false, "explanation": "A confirmation step surfaces the same literal reading back to the manager and consumes a turn without improving the specification behind it."}, {"letter": "D", "text": "Add the phrase \"this is extremely important\" directly in front of the existing sentence.", "correct": false, "explanation": "Emphasis phrases do not clarify intent. The word was already avoided, so raising the stakes on an obeyed rule changes nothing about the report."}, {"letter": "E", "text": "Repeat the banned word instruction three times so that Claude treats it as a firm rule.", "correct": false, "explanation": "Repetition adds emphasis, not information. The instruction was already followed; the problem is that following it literally does not achieve the underlying goal."}], "correct": ["A", "B"], "select": 2}, {"id": "a1-105", "domain": 1, "task_id": "1.4", "objective": "Adapt prompting strategies based on task type (analysis, research, drafting, brainstorming)", "situation": "A researcher writing an executive summary wants Claude's analysis of a quarterly report to note both the strongest supporting evidence and the weakest counterpoints, rather than a one-sided take.", "question": "Which instruction best supports this balanced analysis?", "options": [{"letter": "A", "text": "Ask Claude to list every sentence without judging its significance", "correct": false, "explanation": "Listing every sentence without judgment does not distinguish the strongest evidence from the weakest counterpoints as asked."}, {"letter": "B", "text": "Ask Claude only for evidence supporting the report's existing headline conclusion", "correct": false, "explanation": "Asking only for supporting evidence produces a one-sided analysis and ignores the counterpoints the researcher explicitly wants surfaced."}, {"letter": "C", "text": "Explicitly ask for both the strongest evidence and the weakest counterpoints", "correct": true, "explanation": "Being clear and direct about wanting both the strongest evidence and the weakest counterpoints tells Claude exactly what balanced analytical output is expected, rather than leaving the balance to inference."}, {"letter": "D", "text": "Ask Claude to summarize the report without evaluating any evidence", "correct": false, "explanation": "A plain summary without evaluating evidence does not perform the comparative analysis of strengths and weaknesses that was requested."}], "correct": "C", "select": 1}, {"id": "a2-001", "domain": 2, "task_id": "2.5", "objective": "Edit, adapt, refine, and compare outputs for the intended audience", "situation": "A consultant asks Claude to turn a dense technical whitepaper into a one-page executive brief, but the first attempt still restates most sentences from the whitepaper nearly verbatim.", "question": "What refinement would best address this?", "options": [{"letter": "A", "text": "Ask Claude to identify decisions that matter most to executives and write fresh prose around them.", "correct": true, "explanation": "This refinement directs Claude to synthesize the content around high-impact decisions, moving away from verbatim restatement. Writing fresh prose ensures the output is tailored to executive needs rather than a compressed copy of the original."}, {"letter": "B", "text": "Ask Claude to keep the same original sentences but remove all adjectives and adverbs to save space.", "correct": false, "explanation": "Stripping modifiers leaves the original sentence structures and verbatim phrasing intact, failing to synthesize the material into key takeaways. Executives require distilled insights, not trimmed versions of the same text."}, {"letter": "C", "text": "Ask Claude to spread the whitepaper content across two pages instead of one so that nothing is cut.", "correct": false, "explanation": "Expanding to two pages avoids the need for synthesis and decision-making about what truly matters. An executive brief demands conciseness and focus, not simply more space for undifferentiated content."}, {"letter": "D", "text": "Ask Claude to reduce the font size so the dense text still fits on one page despite its length.", "correct": false, "explanation": "Adjusting font size is a cosmetic change that does not address the underlying problem of undigested, sentence-by-sentence repetition. The content remains dense and unsuitable for an executive summary."}], "correct": "A", "select": 1}, {"id": "a2-002", "domain": 2, "task_id": "2.5", "objective": "Edit, adapt, refine, and compare outputs for the intended audience", "situation": "You asked Claude two different prompts to generate a summary of a research article. Output A is factually correct but misses the study's limitations. Output B mentions the limitations but misstates one key statistic. You need to provide the most accurate summary to a peer-reviewed journal.", "question": "What should you do?", "options": [{"letter": "A", "text": "Select Output A and manually incorporate the missing limitations into the summary.", "correct": false, "explanation": "While Output A is factually correct, it lacks the study's limitations, which are important for a scholarly summary. Manually incorporating limitations without cross-validating against the original article risks inconsistencies or omissions. This approach does not guarantee the completeness and accuracy required for a peer-reviewed journal."}, {"letter": "B", "text": "Reject both and compose a new summary from the original article to guarantee accuracy.", "correct": false, "explanation": "While composing a new summary from the original article would guarantee accuracy, it is unnecessary when a merge-and-validate approach can achieve the same result more efficiently. The task is to work with the available outputs, not to discard them entirely."}, {"letter": "C", "text": "Select Output B because covering limitations outweighs the minor statistical error.", "correct": false, "explanation": "Output B includes limitations but contains a misstated statistic, which is a critical error that undermines credibility. For a peer-reviewed journal, factual accuracy cannot be sacrificed for completeness; a statistical mistake is not minor in this context."}, {"letter": "D", "text": "Merge both: use Output A's statistics and Output B's limitations, then validate all content.", "correct": true, "explanation": "Merging the accurate statistics from Output A with the limitations from Output B, then validating the combined content against the original article, ensures both factual correctness and completeness. This efficient approach meets the precision standards of peer-reviewed submissions."}], "correct": "D", "select": 1}, {"id": "a2-003", "domain": 2, "task_id": "2.1", "objective": "Evaluate Claude-generated outputs for accuracy and completeness", "situation": "A user asks Claude about a company earnings announcement that happened yesterday, and Claude responds confidently with financial figures.", "question": "Given that Claude's knowledge comes from training data with a fixed cutoff date, what should the user do before trusting this answer?", "options": [{"letter": "A", "text": "Verify the figures against a current source, since recent events fall outside Claude's training cutoff", "correct": true, "explanation": "Because a model's knowledge is only reliable up to its training cutoff, users should verify claims about very recent events against a current, authoritative source before trusting them."}, {"letter": "B", "text": "Assume the figures are correct because Claude always declines to answer questions about recent events", "correct": false, "explanation": "Claude does not always decline questions about recent events; it may respond confidently even when the information falls outside its reliable knowledge, which is exactly why independent verification is needed."}, {"letter": "C", "text": "Trust the figures as written, since Claude's training data is continuously updated in real time", "correct": false, "explanation": "Claude's training data has a fixed cutoff and is not updated in real time between releases, so events occurring after that cutoff are not reliably reflected in its knowledge."}, {"letter": "D", "text": "Ignore the response entirely, since Claude cannot provide any useful information about financial topics", "correct": false, "explanation": "Ignoring the response entirely overstates the limitation; the concern is specifically about information beyond the training cutoff, not a blanket inability to discuss financial topics."}], "correct": "A", "select": 1}, {"id": "a2-004", "domain": 2, "task_id": "2.5", "objective": "Edit, adapt, refine, and compare outputs for the intended audience", "situation": "A marketing analyst asks Claude for a sales summary for executives who want only key trends and decisions, not methodology, but the first draft includes several paragraphs on data cleaning.", "question": "How should the analyst refine the prompt for this audience?", "options": [{"letter": "A", "text": "Ask Claude to keep the structure but shorten every paragraph by a similar amount", "correct": false, "explanation": "Uniformly shortening every paragraph keeps the irrelevant methodology section in the summary instead of removing it."}, {"letter": "B", "text": "Ask Claude to move the data cleaning paragraphs to a footnote at the end of the summary", "correct": false, "explanation": "Moving the methodology to a footnote still includes content the executives do not want instead of leading with what they need first."}, {"letter": "C", "text": "Ask Claude to expand the cleaning explanation so executives can verify it themselves", "correct": false, "explanation": "Expanding the methodology section moves the summary further from what a time-constrained executive audience wants."}, {"letter": "D", "text": "Ask Claude to lead with the trends and decisions and drop the data cleaning explanation", "correct": true, "explanation": "Executives want trends and decisions, so restructuring the summary to foreground that content and dropping methodology detail matches the stated audience need."}], "correct": "D", "select": 1}, {"id": "a2-005", "domain": 2, "task_id": "2.4", "objective": "Determine when human review or additional verification is required", "situation": "An operations analyst has a Claude-generated list of two thousand customer accounts that supposedly meet the company's dormancy rule and are due to be closed. Closing an account cannot be reversed.", "question": "What should the analyst do first?", "options": [{"letter": "A", "text": "Proceed with the closures, because the dormancy rule the list was built from is the company's own written rule.", "correct": false, "explanation": "A sound rule does not guarantee sound application of it. The question is whether this list correctly reflects the rule, and that has not been checked."}, {"letter": "B", "text": "Ask Claude to go over its own list a second time and drop any account it is no longer confident belongs there.", "correct": false, "explanation": "Asking the model to re-examine its own output gives no independent evidence. The list can only be validated against the source records it was supposedly derived from."}, {"letter": "C", "text": "Close the smallest accounts on the list first, so that any mistake would affect the least valuable customers.", "correct": false, "explanation": "Sequencing by account value does not verify anything; it only rations the damage. Wrongly closing a small customer's account is still an irreversible error affecting a real person."}, {"letter": "D", "text": "Check a sample of the listed accounts against the source records to confirm the rule was applied correctly.", "correct": true, "explanation": "The action is irreversible and affects real customers, so verification has to happen before anything is executed. Sampling the list against the underlying records tests whether the rule was actually applied as intended, which is the one thing that cannot be fixed afterwards."}], "correct": "D", "select": 1}, {"id": "a2-006", "domain": 2, "task_id": "2.1", "objective": "Evaluate Claude-generated outputs for accuracy and completeness", "situation": "A press officer receives a draft article containing a striking quotation attributed by name to a named industry figure. The quotation is exactly what the article needs.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Find the quotation in a source that carries it, since authoritative-sounding quotations can be invented.", "correct": true, "explanation": "Fabricated quotations are a well-known failure mode and they read as convincingly as real ones. Locating the words in a source that carries them is the only check that distinguishes the two."}, {"letter": "B", "text": "Check that the wording is consistent with positions that this industry figure is already known to hold.", "correct": false, "explanation": "Plausibility is what makes an invented quotation dangerous, because it is generated to fit the person's known positions."}, {"letter": "C", "text": "Ask for the date and setting of the quotation, and include those details alongside it in the published article.", "correct": false, "explanation": "Requesting more surrounding detail can produce more invented detail. Nothing here establishes that the quotation was ever said."}, {"letter": "D", "text": "Paraphrase the quotation instead of quoting it directly, which avoids the need to attribute exact wording.", "correct": false, "explanation": "Paraphrasing still asserts that the person expressed this view, so it attributes an unverified statement while making it harder for readers to check."}], "correct": "A", "select": 1}, {"id": "a2-007", "domain": 2, "task_id": "2.6", "objective": "Organize and curate information and select appropriate output formats (artifacts, inline, structured data)", "situation": "A team records every decision it makes only inside chat messages, spread across many separate conversations.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The decisions will be phrased inconsistently, which makes the team's writing style vary a lot.", "correct": false, "explanation": "Inconsistent phrasing is cosmetic next to colleagues acting on conflicting versions of the same decision."}, {"letter": "B", "text": "The conversations will each reach their length limit, forcing the team to start new ones often.", "correct": false, "explanation": "Restarting conversations is an inconvenience rather than the substantive problem with keeping decisions only in chat."}, {"letter": "C", "text": "There is no single current version, so different people act on different versions of a decision.", "correct": true, "explanation": "Decisions scattered across conversations have no canonical home, so colleagues find whichever message they find and proceed on out-of-date or superseded information."}, {"letter": "D", "text": "The team will spend more of its usage allowance rereading conversations than it saves by chatting.", "correct": false, "explanation": "Rereading a conversation is cheap, and cost is not what makes scattered decisions dangerous."}], "correct": "C", "select": 1}, {"id": "a2-008", "domain": 2, "task_id": "2.6", "objective": "Organize and curate information and select appropriate output formats (artifacts, inline, structured data)", "situation": "Claude has generated a project proposal as a Markdown document artifact. The user wants to tighten the wording of just the second paragraph without regenerating the whole document or losing the original version.", "question": "What should the user do?", "options": [{"letter": "A", "text": "Ask Claude to generate a second, separate artifact containing only the revised paragraph, as artifacts are static after creation and lack any editing interface.", "correct": false, "explanation": "Artifacts are not static; they have an editing interface. Asking for a separate artifact would clutter the workspace and miss the versioning capability of the built-in editor."}, {"letter": "B", "text": "Use the inline \"Edit with Claude\" feature on the Markdown artifact to revise the paragraph; the feature creates a new version and keeps the original accessible.", "correct": true, "explanation": "The inline \"Edit with Claude\" feature on a Markdown artifact allows you to make targeted revisions like tightening a single paragraph. It creates a new version while preserving the original, so the user can compare or revert if needed."}, {"letter": "C", "text": "Delete the artifact and ask Claude to regenerate the entire proposal. Artifacts are locked once created and cannot accept paragraph modifications.", "correct": false, "explanation": "Artifacts are not locked once created; they support inline editing. Deleting and regenerating the entire proposal is unnecessary for a single paragraph change."}, {"letter": "D", "text": "Copy the entire document into the chat as plain text and manually retype the corrected paragraph. Artifacts cannot accept partial edits once created.", "correct": false, "explanation": "The artifact does not need to be copied out and manually retyped. Users can edit artifacts directly within the interface, so partial edits are possible without leaving the artifact."}], "correct": "B", "select": 1}, {"id": "a2-009", "domain": 2, "task_id": "2.3", "objective": "Apply fact-checking and validation techniques", "situation": "A facilities manager used Claude with web search on and received an answer citing a national standard for emergency lighting inspection intervals. She plans to rewrite the depot inspection schedule around that interval.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Open the source that Claude cited and read what the standard itself says about intervals.", "correct": true, "explanation": "Where web search produced the answer, reviewing the cited source is the recommended check, because reading the original reveals anything Claude has misinterpreted. With a safety-related schedule at stake, the standard's own wording is what she needs."}, {"letter": "B", "text": "Search for other online articles that happen to mention the same inspection interval.", "correct": false, "explanation": "Secondary articles may all be repeating the same misreading, and agreement between them proves nothing about what the standard requires. The cited source is closer to the truth than any commentary on it."}, {"letter": "C", "text": "Apply the new interval at one depot first and see whether any problem shows up there.", "correct": false, "explanation": "Piloting a safety interval treats real depots as the test, and a lengthened interval would show its consequences only when something goes wrong. Reversibility is low here, so this is the wrong order."}, {"letter": "D", "text": "Ask Claude to confirm that the interval it has given her is correct before she relies on it.", "correct": false, "explanation": "Claude confirming its own claim adds confidence without adding evidence. Self-confirmation is not verification, and the citation offers a much stronger check that costs no more effort."}], "correct": "A", "select": 1}, {"id": "a2-010", "domain": 2, "task_id": "2.5", "objective": "Edit, adapt, refine, and compare outputs for the intended audience", "situation": "A manager takes one Claude-drafted announcement about a service change and, in three separate conversations, has it reworked for staff, for customers and for the regulator. She sends all three without ever reading them side by side.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The three versions adopt slightly different tones of voice, so the organisation sounds less consistent than its style guide requires.", "correct": false, "explanation": "Tone variation between audiences is largely intended, and an inconsistent voice is a presentational problem. It is a much smaller concern than the underlying facts differing between the three versions."}, {"letter": "B", "text": "The three versions end up at noticeably different lengths, so some audiences receive a considerably fuller account of the change than others.", "correct": false, "explanation": "Different lengths for different audiences are usually appropriate rather than a defect; a regulator and a customer legitimately need different amounts of detail. Length alone does not create a contradiction."}, {"letter": "C", "text": "The three versions drift apart on the substance, so different audiences are told different things and an outsider finds the discrepancy.", "correct": true, "explanation": "Adapting independently lets the facts, dates and commitments diverge, and because nobody compared the versions the inconsistency surfaces externally, in the worst case in front of the regulator."}, {"letter": "D", "text": "The three conversations consume more of her weekly allowance than adapting the announcement once inside a single conversation would.", "correct": false, "explanation": "Three conversations use somewhat more allowance than one, but that is a modest efficiency cost. It is far outweighed by the possibility of telling a regulator something different from what customers were told."}], "correct": "C", "select": 1}, {"id": "a2-011", "domain": 2, "task_id": "2.2", "objective": "Identify hallucinations, inconsistencies, and biases in responses", "situation": "An analyst must decide which of four Claude tasks to check most carefully for invented content. Web search is switched off in all four.", "question": "Which task carries the highest risk of fabricated specifics?", "options": [{"letter": "A", "text": "Summarising a long report she has uploaded, keeping strictly to the points the report makes.", "correct": false, "explanation": "The material is supplied, so claims can be traced back to it. Errors here are usually distortion or omission rather than invention."}, {"letter": "B", "text": "Tightening the wording of a paragraph that she wrote herself and pasted into the conversation.", "correct": false, "explanation": "The source text is present in the conversation, so there is little for Claude to fill in and the result is easy to compare against the original."}, {"letter": "C", "text": "Producing a list of the deals announced in her industry over the past three weeks, with dates.", "correct": true, "explanation": "Recent, specific events sit outside what the model reliably knows and no source has been supplied, so plausible-looking names and dates can be generated to fill the gap."}, {"letter": "D", "text": "Suggesting a set of possible names for an internal working group that she will choose between.", "correct": false, "explanation": "Suggestions are meant to be invented, and she evaluates them directly, so there is no factual claim to be wrong about."}], "correct": "C", "select": 1}, {"id": "a2-012", "domain": 2, "task_id": "2.1", "objective": "Evaluate Claude-generated outputs for accuracy and completeness", "situation": "A communications officer reviews a well-written briefing produced from documents she has not read closely herself. It is confident, tidy and reads exactly like the briefings her team normally circulates.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "How well the briefing reads is taken as evidence that its contents are right, so wrong details go unchallenged.", "correct": true, "explanation": "Fluency and accuracy are independent: a wrong figure or a misattributed statement is written in exactly the same assured style as a right one. Reviewers who use readability as their signal have no way to see the errors."}, {"letter": "B", "text": "The briefing may repeat points from the source documents almost word for word, rather than putting them in her own words.", "correct": false, "explanation": "Close paraphrasing is worth watching for attribution reasons, but content taken faithfully from the source is not the accuracy problem here."}, {"letter": "C", "text": "The briefing may be longer than colleagues have time to read, so its later sections are skimmed or skipped entirely.", "correct": false, "explanation": "Length affects how much gets read, but a shorter briefing built on the same unchecked material carries the same errors."}, {"letter": "D", "text": "The briefing may adopt a tone that is more formal than the one her team's regular readers have come to expect.", "correct": false, "explanation": "Tone that misses the house style is noticeable on a first read and easily adjusted, and it does not put anything untrue in front of readers."}], "correct": "A", "select": 1}, {"id": "a2-013", "domain": 2, "task_id": "2.4", "objective": "Determine when human review or additional verification is required", "situation": "An AI-generated history lesson states that a famous battle occurred in 1067, but established sources say 1066.", "question": "What should the content creator do?", "options": [{"letter": "A", "text": "Publish the lesson without changes, as the AI may have accessed newer research that supersedes traditional accounts.", "correct": false, "explanation": "Publishing without verification risks spreading misinformation. AI does not inherently have access to newer or more accurate research than established historical records, and assuming so for a well-documented date is unwarranted."}, {"letter": "B", "text": "Add a footnote indicating that historians dispute the exact year, but retain the AI-generated date in the main text.", "correct": false, "explanation": "Retaining an incorrect date in the main text, even with a footnote, misleads readers. The primary content should reflect verified facts; a footnote does not resolve the factual error."}, {"letter": "C", "text": "Check the date against several trusted records, fix the mistake, and scan for other potential factual discrepancies.", "correct": true, "explanation": "The creator should verify the fact using multiple trusted historical sources, correct any error, and then review surrounding content for additional inaccuracies. This aligns with best practices for fact-checking AI-generated content and ensuring reliability."}, {"letter": "D", "text": "Delete the line with the date entirely and replace it with a note that the date is uncertain and needs review later.", "correct": false, "explanation": "Removing the date avoids the specific error but does not address the need to verify and correct factual information. The better practice is to confirm the accurate date and update the lesson accordingly, not to omit the detail."}], "correct": "C", "select": 1}, {"id": "a2-014", "domain": 2, "task_id": "2.3", "objective": "Apply fact-checking and validation techniques", "situation": "A compliance officer must validate a Claude summary of a contract that was uploaded to the conversation.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "List the claims that carry weight; find the passage behind each one; mark the claims with no passage; decide what to do about those.", "correct": true, "explanation": "Scoping the review, tracing each claim to the contract, and isolating what could not be traced is what turns a summary into something a compliance officer can sign off."}, {"letter": "B", "text": "List the claims that carry weight; decide what to do about those; mark the claims with no passage; find the passage behind each one.", "correct": false, "explanation": "The decision is made before any evidence has been gathered, which makes the remaining steps ceremonial."}, {"letter": "C", "text": "Mark the claims with no passage; list the claims that carry weight; find the passage behind each one; decide what to do about those.", "correct": false, "explanation": "Nothing can be marked as unsupported until the search for supporting passages has actually been carried out."}, {"letter": "D", "text": "Find the passage behind each one; list the claims that carry weight; decide what to do about those; mark the claims with no passage.", "correct": false, "explanation": "Passages cannot be traced before the claims have been listed, and decisions are taken before the untraceable claims are identified."}], "correct": "A", "select": 1}, {"id": "a2-015", "domain": 2, "task_id": "2.6", "objective": "Organize and curate information and select appropriate output formats (artifacts, inline, structured data)", "situation": "", "question": "During a conversation about database design, a user asks Claude, \"What does the acronym ACID stand for?\" What is the most appropriate way for Claude to answer?", "options": [{"letter": "A", "text": "Create an artifact with a Mermaid diagram illustrating each property, since diagrams are always preferred for technical terminology", "correct": false, "explanation": "A one-line factual answer does not warrant a diagram, and diagrams are not the default treatment for terminology questions."}, {"letter": "B", "text": "Respond inline in the conversation, since the answer is a short, self-contained fact that does not need to be edited or reused separately", "correct": true, "explanation": "A brief factual answer with no reuse or editing need fits naturally as an inline conversational response rather than a standalone artifact."}, {"letter": "C", "text": "Create an artifact containing the definition, since any factual answer should be stored outside the conversation for later reference", "correct": false, "explanation": "Artifacts are reserved for substantial, self-contained content; a short acronym definition does not meet that threshold."}, {"letter": "D", "text": "Ask the user to confirm whether they want the answer as an artifact before responding, since format choice requires explicit approval", "correct": false, "explanation": "Claude selects the output format automatically based on content characteristics; it does not need to ask permission for a short factual reply."}], "correct": "B", "select": 1}, {"id": "a2-016", "domain": 2, "task_id": "2.5", "objective": "Edit, adapt, refine, and compare outputs for the intended audience", "situation": "A teacher asks Claude to explain photosynthesis for a group of 8-year-olds and separately for high school biology students.", "question": "What is the best approach to producing output suited to each group?", "options": [{"letter": "A", "text": "Request 2 explanations, using simple analogies for the younger group and precise terms for the older group.", "correct": true, "explanation": "Requesting two separate explanations allows you to tailor vocabulary, analogies, and depth specifically for each audience, making the content genuinely appropriate for both 8-year-olds and high school biology students."}, {"letter": "B", "text": "Give the high-school-level explanation to both groups because its detailed terms convey the basics to younger children.", "correct": false, "explanation": "Giving the high-school-level explanation to both groups disregards the younger children's limited vocabulary and conceptual readiness, rendering it ineffective for the intended 8-year-old audience."}, {"letter": "C", "text": "Give the child-friendly explanation to both groups to ensure no content exceeds the younger children's comprehension.", "correct": false, "explanation": "Using the child-friendly explanation for both groups would oversimplify the material for high school students, omitting precise terminology and the depth of understanding required at their level."}, {"letter": "D", "text": "Use one explanation written at an intermediate level, balancing simple analogies for younger kids with key terms for older ones.", "correct": false, "explanation": "A single intermediate-level explanation that balances simple analogies with key terms would likely be too complex for 8-year-olds while lacking the scientific rigor expected by high school biology students."}], "correct": "A", "select": 1}, {"id": "a2-017", "domain": 2, "task_id": "2.1", "objective": "Evaluate Claude-generated outputs for accuracy and completeness", "situation": "A facilities coordinator pasted a list of twelve depots and asked for a condition summary of each. The reply is well organised and covers nine of them without saying that three are absent.", "question": "What check would have caught this?", "options": [{"letter": "A", "text": "Comparing the length of this reply against the length of the depot summaries she received in previous months.", "correct": false, "explanation": "Overall length varies with how much there was to report each month, so it is far too coarse a signal to reveal three missing entries."}, {"letter": "B", "text": "Asking whether every depot on the list has been covered, and accepting the answer that comes back in reply.", "correct": false, "explanation": "Asking whether coverage is complete invites the same process that produced the gap to confirm there is no gap."}, {"letter": "C", "text": "Reading the summary closely for depots described in less detail than the others, which suggests thin coverage.", "correct": false, "explanation": "Thin descriptions signal weak coverage of depots that are present. A depot left out entirely produces no description to notice."}, {"letter": "D", "text": "Counting the entries in the reply against the twelve depots on the list she supplied in her original request.", "correct": true, "explanation": "Omissions leave no mark in the text, so they are only visible when the output is counted back against the input. A list she supplied gives her an exact expected number to check against."}], "correct": "D", "select": 1}, {"id": "a2-018", "domain": 2, "task_id": "2.5", "objective": "Edit, adapt, refine, and compare outputs for the intended audience", "situation": "A recruiter needs a job posting to sound approachable and conversational for candidates, but the first draft reads like a formal internal policy memo.", "question": "Which instruction best adapts the tone for this audience?", "options": [{"letter": "A", "text": "Lengthen the posting with more details about its internal reporting structures.", "correct": false, "explanation": "Lengthening the posting with internal reporting details makes it more bureaucratic and formal. It moves further away from an approachable, conversational tone."}, {"letter": "B", "text": "Add bullet points that list the required qualifications in the same formal tone.", "correct": false, "explanation": "Adding bullet points only reformats the content without changing the formal tone. The posting would remain unapproachable for candidates."}, {"letter": "C", "text": "Insert the company's mission statement at the very top of the formal draft.", "correct": false, "explanation": "Inserting the mission statement adds new content but leaves the formal tone unchanged. The overall voice stays unfriendly to candidates."}, {"letter": "D", "text": "Rewrite the posting in a warm, conversational tone that appeals to candidates.", "correct": true, "explanation": "This instruction directly addresses the tone mismatch by asking for a warm, conversational style tailored to candidates. It transforms the formal memo into an approachable, inviting job posting."}], "correct": "D", "select": 1}, {"id": "a2-019", "domain": 2, "task_id": "2.4", "objective": "Determine when human review or additional verification is required", "situation": "A corporate legal team uses Claude to flag unusual clauses in a batch of vendor contracts and draft a risk summary for each.", "question": "What is the most appropriate way to use these summaries before any contract is signed?", "options": [{"letter": "A", "text": "File the AI-drafted summaries as the official legal record alongside the signed contracts themselves", "correct": false, "explanation": "Treating an unverified AI draft as the official legal record skips the verification step the scenario requires before any binding action."}, {"letter": "B", "text": "Rerun the request with extended thinking enabled and treat the new draft as the final legal answer", "correct": false, "explanation": "Regenerating the summary with more reasoning may improve clarity but still comes from the same model and cannot replace independent legal judgment."}, {"letter": "C", "text": "Send the summaries directly to the vendor's counsel as the company's opening negotiating position", "correct": false, "explanation": "Sending an unreviewed internal risk assessment to the counterparty could expose the company's negotiating weaknesses and is not a verification step at all."}, {"letter": "D", "text": "Have a licensed attorney review the flagged clauses and full contract before signing is authorized", "correct": true, "explanation": "Contract execution carries binding legal and financial consequences, so a qualified attorney must review the actual clauses and confirm the risk assessment before signing."}], "correct": "D", "select": 1}, {"id": "a2-020", "domain": 2, "task_id": "2.2", "objective": "Identify hallucinations, inconsistencies, and biases in responses", "situation": "An HR partner suspects that Claude's candidate summaries describe some applicants more warmly than others for reasons unrelated to their experience.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Compare the wording across the results; agree what an even summary should contain; run closely comparable profiles; decide whether the difference is systematic.", "correct": false, "explanation": "There is nothing to compare until the comparable profiles have been run, so this ordering puts the comparison before the material it needs."}, {"letter": "B", "text": "Run closely comparable profiles; agree what an even summary should contain; decide whether the difference is systematic; compare the wording across the results.", "correct": false, "explanation": "Deciding whether a difference is systematic before the wording has been compared reverses the two steps that carry the evidence."}, {"letter": "C", "text": "Agree what an even summary should contain; compare the wording across the results; decide whether the difference is systematic; run closely comparable profiles.", "correct": false, "explanation": "The comparable profiles are the test itself, so running them last means the conclusion was reached without them."}, {"letter": "D", "text": "Agree what an even summary should contain; run closely comparable profiles; compare the wording across the results; decide whether the difference is systematic.", "correct": true, "explanation": "A standard has to exist before wording can be judged against it, comparable profiles create a fair test, and only then does a difference in wording mean anything."}], "correct": "D", "select": 1}, {"id": "a2-021", "domain": 2, "task_id": "2.4", "objective": "Determine when human review or additional verification is required", "situation": "A marketing team asks Claude to generate an original jingle and slogan closely inspired by a currently popular ad campaign from a competitor.", "question": "Before using the output publicly, what should happen?", "options": [{"letter": "A", "text": "Publish the jingle immediately across all channels since Claude was instructed to make the output original", "correct": false, "explanation": "An instruction to be original does not guarantee the output avoids meaningful similarity to the source campaign it was inspired by."}, {"letter": "B", "text": "The team shortens the slogan so it is less likely to resemble the competitor's campaign", "correct": false, "explanation": "Shortening the slogan does not substantively address whether the underlying concept or expression is too similar to the competitor's work."}, {"letter": "C", "text": "The team asks Claude to confirm in writing that the output does not infringe on anything", "correct": false, "explanation": "A model's self-declaration about infringement is not a reliable legal determination and cannot replace expert review."}, {"letter": "D", "text": "Legal and creative reviewers assess the output for similarity to the existing campaign before it is used", "correct": true, "explanation": "Content closely inspired by an existing campaign carries intellectual property risk, so legal and creative reviewers should assess similarity before public use."}], "correct": "D", "select": 1}, {"id": "a2-022", "domain": 2, "task_id": "2.5", "objective": "Edit, adapt, refine, and compare outputs for the intended audience", "situation": "An internal communications lead asks Claude to adapt a global announcement for staff at a site that is closing. The draft keeps the upbeat, celebratory framing of the original.", "question": "What is the most appropriate refinement?", "options": [{"letter": "A", "text": "Rewrite it in a register that fits what this audience is being told, keeping the facts intact.", "correct": true, "explanation": "Adapting for an audience means matching the register to their situation, not just their reading level, while every factual statement stays the same."}, {"letter": "B", "text": "Add a paragraph acknowledging the closure at the end of the otherwise unchanged announcement text.", "correct": false, "explanation": "Appending an acknowledgement after celebratory content reads as an afterthought and does not fix the framing of the piece."}, {"letter": "C", "text": "Shorten the announcement so that the celebratory passages take up less space in the message.", "correct": false, "explanation": "Trimming the upbeat passages leaves the same mismatch in a smaller form and still ignores what the audience is receiving."}, {"letter": "D", "text": "Keep the celebratory framing, since consistency with the global message matters most here.", "correct": false, "explanation": "Consistency applies to the facts of the announcement, not to a tone that misreads what this audience is being told."}], "correct": "A", "select": 1}, {"id": "a2-023", "domain": 2, "task_id": "2.5", "objective": "Edit, adapt, refine, and compare outputs for the intended audience", "situation": "A customer success manager asks Claude to rewrite a churn-risk analysis, originally written for data scientists, so it can be shared with account executives who need to know which accounts to prioritize and what actions to take.", "question": "What change best reflects an effective refinement here?", "options": [{"letter": "A", "text": "Keep the same structure and only replace the word 'churn' with 'account attrition' throughout", "correct": false, "explanation": "Swapping a single term does not change the underlying methodology-heavy structure that is not suited to an executive audience focused on action."}, {"letter": "B", "text": "Add more detail on the model's feature weights so executives understand the scoring fully", "correct": false, "explanation": "Adding more statistical modeling detail moves the document further from what account executives need, which is prioritization, not model internals."}, {"letter": "C", "text": "Remove all account names so the analysis focuses purely on aggregate customer trends", "correct": false, "explanation": "Removing account names would eliminate the exact information executives need to know which specific accounts to prioritize."}, {"letter": "D", "text": "Reorganize around prioritized accounts and recommended actions, moving methodology to an appendix", "correct": true, "explanation": "Account executives need actionable priorities and next steps up front, so restructuring around that need and demoting methodology detail matches their purpose for reading."}], "correct": "D", "select": 1}, {"id": "a2-024", "domain": 2, "task_id": "2.1", "objective": "Evaluate Claude-generated outputs for accuracy and completeness", "situation": "An estates officer asked for a maintenance summary covering all eight regional depots, plus a cost estimate and a recommended priority order for each. The reply is polished and reads convincingly.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Read the summary closely for anything that sounds implausible, since a fluent draft rarely omits requested material.", "correct": false, "explanation": "Plausibility screening catches odd-sounding claims but is blind to omission: a section that was never written cannot sound wrong. Fluency is not evidence of coverage, so this misses the most likely defect in a multi-part request."}, {"letter": "B", "text": "Forward the summary to the depot managers so that each of them can tell her whether their own section is missing.", "correct": false, "explanation": "Distributing the draft pushes a basic completeness check onto colleagues and risks an incomplete document circulating as if it were finished. It is also slower than simply comparing the reply with the request she wrote."}, {"letter": "C", "text": "Ask Claude to confirm that its summary addressed every part of the request, and rely on the answer it gives.", "correct": false, "explanation": "Claude is not a reliable auditor of its own coverage, and treating its self-assessment as confirmation makes it a singular source of truth on the very point in doubt. The request itself is the standard to compare against."}, {"letter": "D", "text": "Lay her original request beside the reply and mark which of the depots and required elements are actually covered.", "correct": true, "explanation": "Completeness is judged against the original request, not against how good the answer sounds. Checking the reply element by element against what she asked for immediately exposes any depot or required component that was dropped."}], "correct": "D", "select": 1}, {"id": "a2-025", "domain": 2, "task_id": "2.1", "objective": "Evaluate Claude-generated outputs for accuracy and completeness", "situation": "A market summary states that three customer segments account for 45, 38 and 29 percent of revenue. The analyst notices the three add to more than a hundred.", "question": "What should she conclude?", "options": [{"letter": "A", "text": "That the largest segment is the one in error, since a figure of 45 percent is the least likely of the three to be right.", "correct": false, "explanation": "Nothing identifies which figure is wrong. Picking one on the basis of its size is a guess, and the other two may be equally unfounded."}, {"letter": "B", "text": "That rounding explains the excess, so the figures can be used once each one is rounded to the nearest whole number.", "correct": false, "explanation": "Rounding moves totals by a point or two at most. A twelve-point excess is far outside what rounding can produce, so this explanation does not fit the evidence."}, {"letter": "C", "text": "That the segments must overlap, so the summary is correct and needs only a footnote explaining the overlap to readers.", "correct": false, "explanation": "Overlapping segments are possible, but that is an assumption invented to rescue the numbers. If overlap were intended the summary would have to say so, and it does not."}, {"letter": "D", "text": "That the numeric section is unreliable as a whole, and every figure in it needs checking against the source data.", "correct": true, "explanation": "The inconsistency shows that the figures were not derived from a single reconciled source, which says nothing reassuring about any individual number. One visible contradiction is a reason to re-check the whole set."}], "correct": "D", "select": 1}, {"id": "a2-026", "domain": 2, "task_id": "2.4", "objective": "Determine when human review or additional verification is required", "situation": "A health-tech startup uses Claude to draft summaries of patient symptoms from intake forms, which are then used to suggest possible next steps to patients.", "question": "Before any suggestion reaches a patient, what should the team implement to manage this high-stakes use case appropriately?", "options": [{"letter": "A", "text": "Increase the model's thinking budget so it reasons longer before producing the suggestion", "correct": false, "explanation": "More reasoning effort can improve response quality but does not provide the independent verification needed for a safety-critical, patient-facing decision."}, {"letter": "B", "text": "Require a licensed clinician to review and approve every suggestion before it reaches a patient", "correct": true, "explanation": "Patient-facing medical guidance is a high-stakes domain where an incorrect or misread output can cause real harm, so a licensed clinician must review and approve suggestions before they reach patients."}, {"letter": "C", "text": "Have a second instance of Claude cross-check the summary against the intake form before sending it", "correct": false, "explanation": "A second Claude pass can catch internal inconsistencies but cannot substitute for qualified human judgment about clinical appropriateness and patient safety."}, {"letter": "D", "text": "Post a visible disclaimer noting the message content was generated by an AI system, not a clinician", "correct": false, "explanation": "A disclaimer informs patients about the source of the content but does nothing to catch an unsafe or inaccurate suggestion before it is delivered."}], "correct": "B", "select": 1}, {"id": "a2-027", "domain": 2, "task_id": "2.6", "objective": "Organize and curate information and select appropriate output formats (artifacts, inline, structured data)", "situation": "A programme coordinator is deciding whether a piece of Claude's output should be produced as an artifact rather than as an ordinary reply in the chat. Which two characteristics point towards an artifact?", "question": "Select two.", "options": [{"letter": "A", "text": "The reply needs to be read once during the discussion and then not referred to again.", "correct": false, "explanation": "Something read once and then discarded fails the test of being referred back to, which is one of the reasons for creating an artifact in the first place."}, {"letter": "B", "text": "The coordinator wants Claude to talk through how it reached its earlier suggestion.", "correct": false, "explanation": "Reasoning about the discussion is conversational by nature. It is tied to the exchange and would not be meaningful as a standalone document."}, {"letter": "C", "text": "The content is something the team will edit and reuse well outside this conversation.", "correct": true, "explanation": "Content meant to be edited, iterated on and reused elsewhere is exactly what an artifact is for, since it gives the work a durable home outside the chat."}, {"letter": "D", "text": "The answer is a two-sentence clarification of a term used earlier in the discussion.", "correct": false, "explanation": "A brief clarification belongs inline. Putting two sentences into a separate document adds friction without giving anyone something worth returning to."}, {"letter": "E", "text": "The content makes sense on its own without the surrounding chat to explain it.", "correct": true, "explanation": "Standing alone is a separate criterion from reuse: content that needs the conversation around it to make sense loses meaning once it is lifted out."}], "correct": ["C", "E"], "select": 2}, {"id": "a2-028", "domain": 2, "task_id": "2.3", "objective": "Apply fact-checking and validation techniques", "situation": "A supplier manager has an answer that names a specific piece of waste-handling regulation and gives its effective date. Web search was switched off for the whole conversation, so no sources were consulted.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Treat the regulation and its date as unverified and look both of them up herself in the official register of legislation.", "correct": true, "explanation": "With web search off there are no cited sources to review, so nothing in the conversation supports the claim. A regulation and its effective date must be confirmed against the authoritative public record."}, {"letter": "B", "text": "Judge the claim by how well it fits the rest of the answer, which was accurate on every point she already knew about.", "correct": false, "explanation": "Internal coherence is not evidence. A fabricated regulation or date will sit comfortably alongside correct material, which is precisely what makes this kind of error hard to spot by reading alone."}, {"letter": "C", "text": "Turn web search on and put the same question again, accepting the second answer if it repeats exactly the same date as before.", "correct": false, "explanation": "A repeated answer from the same assistant is not independent confirmation, and treating agreement between two attempts as verification leaves Claude as the singular source behind a regulatory claim."}, {"letter": "D", "text": "Ask Claude to list the sources behind the regulation and its date, then check the few items it returns in that list.", "correct": false, "explanation": "Asking for sources after the fact invites a list assembled to justify a claim that was made without them, so the references themselves would then need checking. It adds a step without adding reliability."}], "correct": "A", "select": 1}, {"id": "a2-029", "domain": 2, "task_id": "2.4", "objective": "Determine when human review or additional verification is required", "situation": "A team enables extended thinking with a summarized display so they can review Claude's reasoning steps before accepting the answer to a complex multi-step business problem.", "question": "What is the primary benefit of reviewing this reasoning?", "options": [{"letter": "A", "text": "It automatically corrects any flawed reasoning it detects before the final answer is ever shown to reviewers", "correct": false, "explanation": "Extended thinking displays the reasoning trace; it does not automatically fix errors found within it, a human must still catch and correct them."}, {"letter": "B", "text": "It removes the need to check the model's final answer against any outside data source for the problem", "correct": false, "explanation": "Reviewing the reasoning steps does not replace checking the conclusion against external data when the conclusion depends on facts outside the model."}, {"letter": "C", "text": "It proves mathematically that the final answer is correct before anyone reads the generated response", "correct": false, "explanation": "Showing the reasoning process is not a mathematical proof of correctness; it is a transparency aid that still requires human judgment."}, {"letter": "D", "text": "It lets reviewers catch flawed logic or a wrong assumption before the final answer is trusted and acted on", "correct": true, "explanation": "Visible reasoning steps let a reviewer spot a flawed assumption or logic error before the final answer is trusted, which is the main verification benefit of extended thinking."}], "correct": "D", "select": 1}, {"id": "a2-030", "domain": 2, "task_id": "2.2", "objective": "Identify hallucinations, inconsistencies, and biases in responses", "situation": "A team pastes each Claude-written summary into a second Claude conversation and asks it to flag any invented claims. The second conversation is not given the underlying documents.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The second conversation will refuse the request because it cannot see the original documents.", "correct": false, "explanation": "It will answer willingly, which is what makes the arrangement misleading rather than merely useless."}, {"letter": "B", "text": "Without the source material the checker can only judge plausibility, so invented claims pass.", "correct": true, "explanation": "A check with no access to the evidence can assess whether a claim sounds reasonable, and a fabricated claim was generated precisely because it sounds reasonable."}, {"letter": "C", "text": "The second conversation consumes so much allowance that the team runs out before finishing.", "correct": false, "explanation": "The extra cost is real but modest, and it is not what makes this practice unsafe."}, {"letter": "D", "text": "The checker will rewrite the summary instead of assessing it, so the wording changes each time.", "correct": false, "explanation": "Rewriting is easy to notice and easy to correct. The damaging failure is a confident clearance the team believes."}], "correct": "B", "select": 1}, {"id": "a2-031", "domain": 2, "task_id": "2.4", "objective": "Determine when human review or additional verification is required", "situation": "A teacher uses Claude to generate suggested scores and feedback for student essays.", "question": "For essays where the suggested score is borderline between two grade bands, what should the teacher do?", "options": [{"letter": "A", "text": "Automatically assign the higher grade band since the essay's quality is close to that level.", "correct": false, "explanation": "Biasing the score in favor of the student may not align with the actual rubric or the intended grading standards. Responsible AI use dictates that the teacher apply consistent criteria rather than defaulting to arbitrary rules. Anthropic's documentation does not suggest automatically raising scores; instead, it stresses evaluating outputs for appropriateness."}, {"letter": "B", "text": "Accept Claude's suggested score and feedback without modification, as Claude is designed to be helpful, honest, and harmless.", "correct": false, "explanation": "While Claude aims to be HHH (Helpful, Harmless, Honest), no AI model is infallible. Anthropic explicitly states that developers are responsible for reviewing outputs for appropriateness, and for high-stakes tasks like grading, human oversight is critical. Blindly trusting the model contradicts the principle of human review recommended in Anthropic's guidelines."}, {"letter": "C", "text": "Manually review the essay and use professional judgment to determine the final score, using Claude's suggestions as one input.", "correct": true, "explanation": "Anthropic's terms of service and responsible AI practices emphasize that developers (and users) must evaluate API outputs for appropriateness, especially in high-stakes decisions like grading. Human review is essential for borderline cases where the model's confidence may be low, ensuring fairness and accuracy. Research indicates that AI models can sometimes generate inconsistent or irrelevant outputs, so final decisions should incorporate human oversight."}, {"letter": "D", "text": "Ask Claude to regenerate the score and feedback until it converges on a definitive grade band.", "correct": false, "explanation": "While prompt engineering can improve outputs, regenerating responses may not resolve fundamental ambiguity and could lead to inconsistent results. The model's output is probabilistic, and borderline cases reflect inherent uncertainty; relying solely on regeneration without human judgment is not a recommended best practice per Anthropic's emphasis on human review for critical applications."}], "correct": "C", "select": 1}, {"id": "a2-032", "domain": 2, "task_id": "2.6", "objective": "Organize and curate information and select appropriate output formats (artifacts, inline, structured data)", "situation": "A user asks Claude to write a 400-word case study about a customer's product rollout, noting they plan to revise the wording several times before sharing it with stakeholders.", "question": "How should Claude format the response?", "options": [{"letter": "A", "text": "As a Markdown artifact, because the case study is long-form, self-contained, and meant for iterative editing before external sharing.", "correct": true, "explanation": "The case study is a long-form, self-contained document meant for multiple rounds of revision. A Markdown artifact provides an editable, standalone file that the user can refine and later share with stakeholders, perfectly matching the described use case."}, {"letter": "B", "text": "As a strict-tool-validated response, because tool schemas automatically enforce the output to match the requested 400-word length exactly.", "correct": false, "explanation": "Strict tool validation ensures that tool call parameters meet a schema, not that prose output meets a specific word count. It does not provide any formatting advantage for a case study that needs to be revised."}, {"letter": "C", "text": "Inline in the chat, because inline responses are easier to scroll back to than artifacts when a user wants to review prior wording.", "correct": false, "explanation": "Inline responses are embedded in the chat, making them harder to isolate and edit across multiple rounds of revision. An artifact offers a cleaner, more reusable format for iterative work, especially for longer content."}, {"letter": "D", "text": "As a CSV export, because tabular formats are often the standard way to prepare narrative content for stakeholder distribution.", "correct": false, "explanation": "CSV is designed for tabular data, not narrative prose. A case study is a written narrative, and a tabular format is inappropriate for stakeholder distribution of such content."}], "correct": "A", "select": 1}, {"id": "a2-033", "domain": 2, "task_id": "2.4", "objective": "Determine when human review or additional verification is required", "situation": "An operations team is agreeing which Claude-assisted outputs must be reviewed by a person before use. Which two situations clearly require that review?", "question": "Select two.", "options": [{"letter": "A", "text": "A first draft of brainstorming prompts to be discussed at a team workshop later.", "correct": false, "explanation": "Discussion material is reviewed by definition when the workshop happens, and nothing is acted on before people have looked at it together."}, {"letter": "B", "text": "A set of notes summarising an internal retrospective for the team that attended it.", "correct": false, "explanation": "Notes for the people who were in the room are low stakes and self-correcting, because those readers can immediately spot anything that is wrong."}, {"letter": "C", "text": "A rewritten meeting agenda that the chair will read through before the meeting.", "correct": false, "explanation": "The chair reading the agenda before the meeting already supplies the review, and an error in it is trivially correctable on the spot."}, {"letter": "D", "text": "A customer-facing notice that will be published under the organisation's own name.", "correct": true, "explanation": "External publication under the organisation's name is hard to withdraw and carries reputational consequences, so a person must own it before it is released."}, {"letter": "E", "text": "A written assessment feeding into a decision about an employee's contract.", "correct": true, "explanation": "This is a different trigger from publication: decisions affecting a person's employment must be made and owned by a human regardless of how good the draft is."}], "correct": ["D", "E"], "select": 2}, {"id": "a2-034", "domain": 2, "task_id": "2.1", "objective": "Evaluate Claude-generated outputs for accuracy and completeness", "situation": "A retention lead asked why cancellations rose sharply in the last quarter. The answer explains at length what cancellation rates are and how they are generally measured.", "question": "How should she judge this output?", "options": [{"letter": "A", "text": "As accurate but off target, because it explains the topic instead of addressing the rise she asked about.", "correct": true, "explanation": "Everything present may be correct while the question remains unanswered. Accuracy and relevance are separate tests, and this output passes the first and fails the second."}, {"letter": "B", "text": "As adequate for now, because background on how the measure works is a reasonable place to begin the analysis.", "correct": false, "explanation": "Accepting background as progress leaves the actual question open, and the material she now has does not move her any closer to a cause."}, {"letter": "C", "text": "As evidence that the question cannot be answered, since the reply avoided the specific rise she was asking about.", "correct": false, "explanation": "A general answer indicates that the request was read broadly, not that the specific question is unanswerable given her own data."}, {"letter": "D", "text": "As a sound answer that simply needs shortening, since the explanation contains more detail than she required here.", "correct": false, "explanation": "Shortening reduces the volume of material that does not address her question, leaving a briefer answer to the wrong question."}], "correct": "A", "select": 1}, {"id": "a2-035", "domain": 2, "task_id": "2.5", "objective": "Edit, adapt, refine, and compare outputs for the intended audience", "situation": "A product manager asks Claude to explain a new caching feature to both internal engineers and non-technical customers, but the first draft uses the same technical wording for each.", "question": "What is the most effective way to adapt this output for both audiences?", "options": [{"letter": "A", "text": "Keep the one technical draft and let each reader skip the parts that do not apply to them", "correct": false, "explanation": "Leaving one technical draft unchanged forces non-technical customers to filter out jargon themselves, defeating the purpose of adapting content for an audience."}, {"letter": "B", "text": "Request two separate versions, matching vocabulary and detail to each reader's background", "correct": true, "explanation": "Because the two audiences differ in background and needs, producing separate versions tailored to each produces content that is actually usable by both readers."}, {"letter": "C", "text": "Attach a glossary of engineering terms to the technical draft for customers to consult", "correct": false, "explanation": "A glossary still requires customers to translate jargon themselves rather than receiving an explanation already written in language they understand."}, {"letter": "D", "text": "Trim the technical draft slightly so it reads a bit faster for both groups of readers", "correct": false, "explanation": "Trimming length without changing vocabulary or explanation depth does not address the mismatch between a technical explanation and a non-technical reader."}], "correct": "B", "select": 1}, {"id": "a2-036", "domain": 2, "task_id": "2.4", "objective": "Determine when human review or additional verification is required", "situation": "An analyst asks Claude to compute a set of financial ratios from a spreadsheet and explain the results in prose, without using a code execution tool.", "question": "What is the safest way to handle the numeric outputs before they are used in a report?", "options": [{"letter": "A", "text": "Ask Claude to generate a detailed step-by-step explanation of each calculation and include that in the report as justification.", "correct": false, "explanation": "While a step-by-step explanation can help the analyst trace the logic, it does not validate the correctness of the calculations. The underlying numbers could still be wrong, and the explanation itself might contain inaccuracies. The safest approach is to independently verify the outputs."}, {"letter": "B", "text": "Manually verify the ratios by recalculating them with the original spreadsheet data before including them in the report.", "correct": true, "explanation": "Claude can make arithmetic errors when performing calculations without a code execution tool, as the model responds based on patterns rather than executing precise algorithms. Official Anthropic financial analysis solutions recommend human oversight for critical numeric outputs, and recalculating the ratios directly from the source data is the most reliable method to ensure accuracy before the results are used in a report."}, {"letter": "C", "text": "Submit the ratios as-is because Claude's financial analysis capabilities are certified for accuracy in reporting.", "correct": false, "explanation": "While Claude's outputs are often accurate, no AI model is certified to guarantee error‑free financial computations without verification. Anthropic's documentation for financial services emphasizes the importance of human review and validation, especially when working with numerical data outside of controlled, code‑execution environments."}, {"letter": "D", "text": "Use a separate Claude session without code execution to independently compute the ratios and only use the results if both sessions match.", "correct": false, "explanation": "Running a second, independent Claude session can help catch some inconsistencies, but it does not guarantee correctness because both instances may make the same type of arithmetic mistake. Manual verification against the original data is a more dependable safeguard."}], "correct": "B", "select": 1}, {"id": "a2-037", "domain": 2, "task_id": "2.1", "objective": "Evaluate Claude-generated outputs for accuracy and completeness", "situation": "A conference programme coordinator uploaded ten breakout-session proposals and asked Claude to summarise each one with a verdict against the published selection criteria. The reply is fluent and well organised.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Read each verdict carefully and decide whether it sounds reasonable to her on its own terms.", "correct": false, "explanation": "Judging whether each verdict sounds reasonable tests plausibility, not coverage. It would not reveal that a proposal or a criterion had been left out altogether, which is the more likely failure with a batch of ten."}, {"letter": "B", "text": "Forward the reply to the selection panel and invite its members to flag anything that looks wrong to them.", "correct": false, "explanation": "Handing an unchecked summary to the panel outsources the coordinator's own responsibility for the output and risks the panel treating a partial summary as the full field of proposals."}, {"letter": "C", "text": "Ask Claude to shorten the summary so that the panel can work through the material more quickly.", "correct": false, "explanation": "Shortening a summary that may already be missing proposals or criteria makes the gap harder to see, not easier. Length is a presentation choice and it is premature here."}, {"letter": "D", "text": "Check the reply against what she asked for, confirming all ten proposals and every criterion appear.", "correct": true, "explanation": "A fluent reply can still be an incomplete one, and completeness is checked against the original request rather than against how the text reads. Confirming that all ten proposals and all the criteria are actually covered is the cheapest check available and it shapes everything she does next."}], "correct": "D", "select": 1}, {"id": "a2-038", "domain": 2, "task_id": "2.2", "objective": "Identify hallucinations, inconsistencies, and biases in responses", "situation": "A user asks, 'When was the Eiffel Tower completed?' The model responds, 'The Eiffel Tower was completed in 1887.' The actual completion year is 1889.", "question": "How could a reviewer most reliably detect this error?", "options": [{"letter": "A", "text": "Search for any contradictory dates in the model's previous answers.", "correct": false, "explanation": "Searching for contradictory dates in the model’s previous answers only checks for consistency, not truth. A model may be consistently wrong, and a standalone query may have no relevant prior answers to compare against."}, {"letter": "B", "text": "Analyze the response's tone for any uncertainty or hedging.", "correct": false, "explanation": "Analyzing the response's tone for uncertainty or hedging does not reliably indicate factual accuracy. A confident tone can still contain errors, and a hesitant tone can be correct, so tone is not a valid verification method."}, {"letter": "C", "text": "Verify the grammatical correctness of the sentence structure.", "correct": false, "explanation": "Verifying grammatical correctness only checks sentence structure, not factual content. A grammatically perfect sentence can still present false information, so this method does not detect factual errors."}, {"letter": "D", "text": "Compare the response to an encyclopedia entry on the Eiffel Tower.", "correct": true, "explanation": "Comparing the response to a trusted encyclopedia entry directly verifies the factual claim against a reliable external source. This method can detect errors regardless of the model's confidence or consistency."}], "correct": "D", "select": 1}, {"id": "a2-039", "domain": 2, "task_id": "2.3", "objective": "Apply fact-checking and validation techniques", "situation": "An operations manager wants to confirm a claim Claude made about her own organisation's escalation process.", "question": "Where should she check?", "options": [{"letter": "A", "text": "In the organisation's current published procedure, or with the person who owns it.", "correct": true, "explanation": "The authoritative source for an internal process is the organisation's own current documentation or its owner, not anything outside it."}, {"letter": "B", "text": "Against a second Claude answer produced from the same question in a new conversation.", "correct": false, "explanation": "Another generated answer is not an authority on how her organisation actually works."}, {"letter": "C", "text": "Against the recollection of a colleague who followed the process a few years ago.", "correct": false, "explanation": "Memory of an older version of a process is a weak authority when a current written procedure exists."}, {"letter": "D", "text": "Against a web search, since published sources are more independent than internal ones.", "correct": false, "explanation": "Independence is not the criterion here. An internal process is defined internally, and no public source describes it."}], "correct": "A", "select": 1}, {"id": "a2-040", "domain": 2, "task_id": "2.3", "objective": "Apply fact-checking and validation techniques", "situation": "A manager wants to know whether Claude's summary of a report she uploaded is faithful to it.", "question": "What is the most reliable check?", "options": [{"letter": "A", "text": "Check that the summary's length is proportionate to the length of the original report.", "correct": false, "explanation": "Proportion is unrelated to accuracy; a well-sized summary can still distort what the report says."}, {"letter": "B", "text": "Ask Claude to summarise the report again and see whether the two versions agree.", "correct": false, "explanation": "Two summaries can agree and both misrepresent the report, so agreement is not evidence of faithfulness."}, {"letter": "C", "text": "Read the summary's main claims against the relevant sections of the report itself.", "correct": true, "explanation": "The uploaded report is the standard the summary is meant to represent, so a direct comparison is the only check that can settle faithfulness."}, {"letter": "D", "text": "Ask Claude to rate how faithful its own summary is to the report on a numerical scale.", "correct": false, "explanation": "A self-assigned score is generated in the same way as the summary and carries no independent weight."}], "correct": "C", "select": 1}, {"id": "a2-041", "domain": 2, "task_id": "2.3", "objective": "Apply fact-checking and validation techniques", "situation": "An analyst finds that a figure in Claude's summary of a spreadsheet she uploaded does not match what she expected.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Correct the figure in the summary to the value she expected, then carry on with the review.", "correct": false, "explanation": "Her expectation may be the thing that is wrong. Overwriting the summary from memory can introduce an error rather than remove one."}, {"letter": "B", "text": "Re-upload the spreadsheet and ask for the whole summary again in an entirely new conversation.", "correct": false, "explanation": "Regenerating produces another unverified summary and leaves her no better placed to say which figure is right."}, {"letter": "C", "text": "Ask Claude to recalculate the figure and then use whichever of the two answers looks the more plausible.", "correct": false, "explanation": "Choosing between two generated answers on plausibility never touches the spreadsheet, which is the only thing that settles the question."}, {"letter": "D", "text": "Look for the figure in the spreadsheet itself, to see whether the summary or her expectation is wrong.", "correct": true, "explanation": "The source she supplied is the authority here, and checking it distinguishes a summarisation error from a mistaken assumption before anything is changed."}], "correct": "D", "select": 1}, {"id": "a2-042", "domain": 2, "task_id": "2.5", "objective": "Edit, adapt, refine, and compare outputs for the intended audience", "situation": "A communications team has two Claude-generated versions of a member newsletter and must choose one to develop further. Which two practices make that comparison reliable?", "question": "Select two.", "options": [{"letter": "A", "text": "Ask Claude which of its two drafts is stronger and take that as the decision.", "correct": false, "explanation": "Claude has no access to the team's priorities beyond what was stated, so this outsources a judgment the communications team is better placed to make."}, {"letter": "B", "text": "Choose the longer version, since more material means more can be cut later.", "correct": false, "explanation": "Length is not a proxy for quality. A longer draft may simply contain more filler, and cutting it back costs time the team need not spend."}, {"letter": "C", "text": "Score each version against the original brief instead of against the other one.", "correct": true, "explanation": "Comparing each draft to the brief is a distinct safeguard: two versions can be ranked against each other while both fail the actual requirement."}, {"letter": "D", "text": "Write down the criteria that matter before either draft is read at all.", "correct": true, "explanation": "Criteria set in advance stop the team from rationalising a preference formed while reading, which is the main way informal comparisons go wrong."}, {"letter": "E", "text": "Choose the version that reads with more confidence and energy on first pass.", "correct": false, "explanation": "Confident writing is a style feature. It can accompany weaker content, and it is precisely the quality that makes an unreliable draft feel persuasive."}], "correct": ["C", "D"], "select": 2}, {"id": "a2-043", "domain": 2, "task_id": "2.6", "objective": "Organize and curate information and select appropriate output formats (artifacts, inline, structured data)", "situation": "A membership officer asked Claude for the wording of a renewal reminder that her team will revise repeatedly over the next few weeks before sending it. Claude returned it as an ordinary reply in the chat.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Copy the wording into the team's chat channel so that colleagues can suggest changes there.", "correct": false, "explanation": "A chat channel scatters revisions across messages with no single current version, which is the problem she is trying to avoid over several weeks of editing."}, {"letter": "B", "text": "Ask Claude to put the wording into an artifact that the team can go on editing and reusing.", "correct": true, "explanation": "Content that is self-contained, will be iterated on, and is meant to be reused outside the conversation is what an artifact is for. Putting the reminder in one gives the team a single editable version to work on rather than a reply buried in chat history."}, {"letter": "C", "text": "Ask Claude for several more versions of the wording and pick between them at a later point.", "correct": false, "explanation": "More variants add choice, not durability. The team would still be editing text that lives inside a conversation and has to be found again each time."}, {"letter": "D", "text": "Paste the wording into a fresh conversation each time the team wants to change something.", "correct": false, "explanation": "Starting a new conversation for every change loses the context behind earlier edits and produces a trail of near-identical drafts with no authoritative copy."}], "correct": "B", "select": 1}, {"id": "a2-044", "domain": 2, "task_id": "2.2", "objective": "Identify hallucinations, inconsistencies, and biases in responses", "situation": "A team adds an instruction to their project telling Claude never to state anything untrue, and stops checking outputs as a result.", "question": "What is the most reasonable expectation of that instruction?", "options": [{"letter": "A", "text": "It makes Claude refuse every question it cannot verify, so the project returns very few answers.", "correct": false, "explanation": "Permission to admit uncertainty leads Claude to flag the specific gaps it notices, not to refuse every question it cannot verify, and the team's real exposure is the checking they have dropped."}, {"letter": "B", "text": "It removes fabricated claims entirely, because Claude will now withhold unsupported detail.", "correct": false, "explanation": "Anthropic's guidance says such techniques significantly reduce hallucinations but do not eliminate them entirely, so an instruction cannot remove fabricated claims completely."}, {"letter": "C", "text": "It may reduce guessing, but it cannot guarantee accuracy, so checking is still required.", "correct": true, "explanation": "Giving Claude permission to admit uncertainty can sharply cut false statements, but Anthropic's guidance says these techniques do not eliminate hallucinations entirely, so important claims still need validating by a person."}, {"letter": "D", "text": "It has no effect at all, because project instructions do not influence how answers are written.", "correct": false, "explanation": "Project instructions do shape responses across the project. The error is treating that influence as a guarantee."}], "correct": "C", "select": 1}, {"id": "a2-045", "domain": 2, "task_id": "2.1", "objective": "Evaluate Claude-generated outputs for accuracy and completeness", "situation": "A policy researcher used Research and received an answer with citations attached to each main claim. She needs to be sure of one claim before quoting it in a submission.", "question": "What should she do with the citation?", "options": [{"letter": "A", "text": "Ask Claude to restate the claim and see whether the second answer cites the same source as the first one.", "correct": false, "explanation": "A repeated answer citing the same source is consistency, not confirmation, and the same misreading can easily recur."}, {"letter": "B", "text": "Check that the cited organisation is a credible one, which establishes that the claim drawn from it is sound.", "correct": false, "explanation": "A reputable publisher makes the underlying material more trustworthy, but says nothing about whether this particular sentence reflects what that material says."}, {"letter": "C", "text": "Count how many citations support the claim, and treat a claim carrying several citations as confirmed.", "correct": false, "explanation": "Several citations can all be attached to a claim none of them quite makes, so counting them measures volume rather than support."}, {"letter": "D", "text": "Open the cited source and confirm it actually states what the answer attributes to it, in the context given.", "correct": true, "explanation": "A citation shows where a claim is said to come from; only reading the source shows whether it supports the claim as stated. Opening it is the step that turns a citation into verification."}], "correct": "D", "select": 1}, {"id": "a2-046", "domain": 2, "task_id": "2.2", "objective": "Identify hallucinations, inconsistencies, and biases in responses", "situation": "A manager asks the same question about a supplier deadline in two separate chats and receives two different dates. She has given Claude no document containing that deadline.", "question": "What is the most reasonable conclusion?", "options": [{"letter": "A", "text": "The dates differ only in wording, so she should reread both looking for a shared meaning.", "correct": false, "explanation": "Two different calendar dates are a conflict of substance, not a difference in phrasing that closer reading will reconcile."}, {"letter": "B", "text": "The later of the two answers is more reliable, because it was produced most recently.", "correct": false, "explanation": "Order of generation says nothing about accuracy; neither answer had access to the deadline in the first place."}, {"letter": "C", "text": "The disagreement proves that the supplier changed the deadline between the two conversations.", "correct": false, "explanation": "Claude was not consulting the supplier in either chat, so the variation tells her about the answers, not about the supplier."}, {"letter": "D", "text": "Neither date is grounded in anything she supplied, so both should be treated as unsupported.", "correct": true, "explanation": "The disagreement is evidence that no source is anchoring the answer. Without the underlying document, both dates are guesses however confidently they are stated."}], "correct": "D", "select": 1}, {"id": "a2-047", "domain": 2, "task_id": "2.1", "objective": "Evaluate Claude-generated outputs for accuracy and completeness", "situation": "A project manager has received a draft board paper and has time to review it properly.", "question": "In what order should she carry out her review?", "options": [{"letter": "A", "text": "Confirm every requested topic is covered; check the claims that carry the argument; correct the tone and formatting; approve it.", "correct": true, "explanation": "Coverage is settled first because it determines what there is to check, accuracy is then established on the full set of content, and presentation is polished last on text that is staying. Approval comes only after all three."}, {"letter": "B", "text": "Check the claims that carry the argument; correct the tone and formatting; approve it; confirm every requested topic is covered.", "correct": false, "explanation": "Verifying claims before knowing what is missing means the added sections arrive later, unchecked, and after approval in this ordering."}, {"letter": "C", "text": "Correct the tone and formatting; check the claims that carry the argument; confirm every requested topic is covered; approve it.", "correct": false, "explanation": "Polishing first means investing effort in wording that may be deleted once a missing topic is added or an unsupported claim is removed."}, {"letter": "D", "text": "Approve it; confirm every requested topic is covered; correct the tone and formatting; check the claims that carry the argument.", "correct": false, "explanation": "Approving before any review has happened makes the remaining steps a formality, and anything found afterwards has already been signed off."}], "correct": "A", "select": 1}, {"id": "a2-048", "domain": 2, "task_id": "2.2", "objective": "Identify hallucinations, inconsistencies, and biases in responses", "situation": "A researcher is scanning a Claude-written sector briefing for signs that something in it has been fabricated. Which two features are the strongest warning signs?", "question": "Select two.", "options": [{"letter": "A", "text": "A named industry report with a specific year that cannot be located anywhere at all.", "correct": true, "explanation": "A confidently named source that does not exist is the classic fabrication signature, and the specificity of the citation is what makes it convincing."}, {"letter": "B", "text": "Technical vocabulary from the sector used throughout the briefing without definitions.", "correct": false, "explanation": "Sector vocabulary is what a briefing would naturally contain. It affects readability for a lay audience but carries no information about truthfulness."}, {"letter": "C", "text": "Two figures in the same briefing that cannot both be true as they are written there.", "correct": true, "explanation": "Internal contradiction is independent evidence: at least one of the two figures must be wrong, and it is detectable from the text alone with no external source."}, {"letter": "D", "text": "A response noticeably longer than the researcher expected for a briefing of this kind.", "correct": false, "explanation": "Length reflects how the request was framed. Long answers are not more likely to be invented than short ones, so this is not a useful signal by itself."}, {"letter": "E", "text": "Cautious phrasing such as \"it appears that\" appearing in several of the paragraphs.", "correct": false, "explanation": "Hedging language signals uncertainty rather than fabrication. If anything, unhedged confidence about a specific unverifiable fact is the more worrying pattern."}], "correct": ["A", "C"], "select": 2}, {"id": "a2-049", "domain": 2, "task_id": "2.3", "objective": "Apply fact-checking and validation techniques", "situation": "An analyst must validate a Claude-written competitor comparison, produced with web search on, before it informs a pricing meeting tomorrow.", "question": "Which sequence is correct?", "options": [{"letter": "A", "text": "Open the source behind each claim and read it, then look up any unsourced claim independently, then mark whatever is still unconfirmed, then list the claims the decision rests on.", "correct": false, "explanation": "Reading every cited source before deciding which claims matter treats all claims as equally important, and working out what the decision rests on last makes that step unable to guide any of the work."}, {"letter": "B", "text": "Look up every claim independently, then list the claims the decision rests on, then open the sources behind them, then mark whatever is still unconfirmed.", "correct": false, "explanation": "Independent lookup of everything before any prioritisation spends the analyst's limited time evenly across claims that matter and claims that do not, and it ignores the sources already attached to the answer."}, {"letter": "C", "text": "Mark whatever is still unconfirmed, then open the source behind each claim and read it, then list the claims the decision rests on, then look up unsourced claims independently.", "correct": false, "explanation": "Marking unconfirmed material before any checking has happened is meaningless, and identifying the decision-critical claims is left until the third step, after the effort has already been spent."}, {"letter": "D", "text": "List the claims the decision rests on, then open the source behind each one and read it, then look up any unsourced claim independently, then mark whatever is still unconfirmed.", "correct": true, "explanation": "Scope the check to the decision-critical claims, review the cited sources for those first because the original content shows whether anything was misread, then fill the gaps independently and flag what remains unconfirmed."}], "correct": "D", "select": 1}, {"id": "a2-050", "domain": 2, "task_id": "2.1", "objective": "Evaluate Claude-generated outputs for accuracy and completeness", "situation": "A procurement officer pasted a sixty-page supplier contract into Claude and asked for a summary of the commercial terms. Which two evaluation steps should she take before circulating it?", "question": "Select two.", "options": [{"letter": "A", "text": "Accept it if it reads coherently from beginning to end without any obvious contradiction.", "correct": false, "explanation": "Coherence is a property of the writing. A fluent summary of the wrong clauses reads exactly as smoothly as a fluent summary of the right ones."}, {"letter": "B", "text": "Ask Claude whether its summary is accurate and treat a confident yes as confirmation.", "correct": false, "explanation": "A self-assessment is generated the same way the summary was. Claude cannot independently confirm its own reading of the contract, so this adds no evidence."}, {"letter": "C", "text": "Take the precise contractual vocabulary in the summary as a sign of a careful reading.", "correct": false, "explanation": "Precise legal vocabulary is easy to produce and is not evidence of fidelity. Confident phrasing is a common feature of inaccurate summaries."}, {"letter": "D", "text": "Open the contract and read the clauses behind several of the summarised commercial points.", "correct": true, "explanation": "Spot-checking against the source is the only way to catch a term that has been subtly restated, such as a notice period reported as thirty rather than ninety days."}, {"letter": "E", "text": "List the terms she needed covered and check whether any of them are simply absent.", "correct": true, "explanation": "Checking for omissions is a different test from checking what is present: a summary can be entirely accurate and still leave out the termination terms."}], "correct": ["D", "E"], "select": 2}, {"id": "a2-051", "domain": 2, "task_id": "2.6", "objective": "Organize and curate information and select appropriate output formats (artifacts, inline, structured data)", "situation": "An analyst has forty pages of interview notes and has been asked to produce something the team can use in a workshop next week.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Ask Claude to produce the notes as a long document she can then trim down before the workshop.", "correct": false, "explanation": "Generating a deliverable before its purpose is settled usually means reworking it, and length is not the problem she is solving."}, {"letter": "B", "text": "Paste all forty pages into a conversation and ask Claude what the most important themes are.", "correct": false, "explanation": "Extracting themes is useful later, but it does not answer what the workshop needs the material to do."}, {"letter": "C", "text": "Split the notes into separate files by interviewee so that each one can be discussed on its own.", "correct": false, "explanation": "Organising by source is one possible structure, and choosing it now assumes an answer she has not worked out yet."}, {"letter": "D", "text": "Decide what the participants must be able to do with it, since that drives structure and format.", "correct": true, "explanation": "Format follows use. Knowing whether people will read, sort, discuss or annotate the material is what determines whether it should be a document, a table or a diagram."}], "correct": "D", "select": 1}, {"id": "a2-052", "domain": 2, "task_id": "2.3", "objective": "Apply fact-checking and validation techniques", "situation": "A team verified the figures in a Claude-drafted capability statement nine months ago and is about to reissue the document unchanged.", "question": "What should they do?", "options": [{"letter": "A", "text": "Remove the figures entirely, so that the statement will not need to be checked again in future.", "correct": false, "explanation": "Stripping out the evidence to avoid maintenance destroys the value of the statement."}, {"letter": "B", "text": "Recheck the figures that move over time, since verification does not stay valid indefinitely.", "correct": true, "explanation": "Headcounts, prices and volumes change, so a claim's verification has a shelf life and reissuing revives every figure as a current assertion."}, {"letter": "C", "text": "Reissue it as it stands, because the figures were checked against sources at the time.", "correct": false, "explanation": "The check confirmed the figures as they were then; it says nothing about what they are now."}, {"letter": "D", "text": "Ask Claude to confirm that none of the figures in the statement have changed since then.", "correct": false, "explanation": "Claude has no reliable view of how the organisation's own numbers have moved over the past nine months."}], "correct": "B", "select": 1}, {"id": "a2-053", "domain": 2, "task_id": "2.3", "objective": "Apply fact-checking and validation techniques", "situation": "A team's convention is to fact-check only those statements in a Claude output that look surprising.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "Checking only some statements causes Claude to produce more surprising statements over time.", "correct": false, "explanation": "How a team reviews an output does not change how later outputs are generated."}, {"letter": "B", "text": "Surprising statements take the longest to check, so the team's review will overrun its budget.", "correct": false, "explanation": "Review effort is a scheduling concern, not the reason this convention lets errors through."}, {"letter": "C", "text": "The team will end up verifying the same surprising claims repeatedly across different documents.", "correct": false, "explanation": "Duplicated effort is wasteful but harmless compared with an unchecked error reaching a decision."}, {"letter": "D", "text": "Wrong figures that look entirely ordinary pass untouched, and those are the ones people act on.", "correct": true, "explanation": "Fabricated content is generated because it fits expectations, so screening on surprise filters out precisely the errors that are easiest to miss."}], "correct": "D", "select": 1}, {"id": "a2-054", "domain": 2, "task_id": "2.4", "objective": "Determine when human review or additional verification is required", "situation": "A marketing team uses Claude to help draft promotional content that includes important numerical figures (e.g., sales numbers, performance metrics).", "question": "Before publishing the content, what verification step is essential to ensure accuracy?", "options": [{"letter": "A", "text": "Accept the figures as correct because Claude’s explanations sound confident and well-written.", "correct": false, "explanation": "Confidence in tone does not guarantee accuracy. AI models like Claude can produce plausible but incorrect information, including hallucinated numbers. Always independently verify critical data before publication."}, {"letter": "B", "text": "Recompute the key figures using a calculator, spreadsheet, or code execution tool to confirm they are accurate.", "correct": true, "explanation": "Recomputation is a direct method to verify mathematical and factual correctness. Because AI outputs can contain errors, it is a best practice to independently validate important numbers using a reliable tool or secondary source."}, {"letter": "C", "text": "Ask Claude to restate the same numbers in a different way to check consistency.", "correct": false, "explanation": "Restating the same information does not verify the underlying data’s correctness. AI can consistently repeat erroneous information. Independent verification is required to catch mistakes."}, {"letter": "D", "text": "Round all the figures to fewer decimal places to reduce the chance of noticeable errors.", "correct": false, "explanation": "Rounding may obscure mistakes but does not correct underlying inaccuracies. The goal is to publish correct data, not simply hide potential errors. Verification should confirm exact values."}], "correct": "B", "select": 1}, {"id": "a2-055", "domain": 2, "task_id": "2.2", "objective": "Identify hallucinations, inconsistencies, and biases in responses", "situation": "A consultant reads a Claude-written market overview that is fluent, well organised and contains no hedging at all.", "question": "What is the greatest risk in reading that as a sign of reliability?", "options": [{"letter": "A", "text": "Fluent writing takes longer to produce, so the overview is likely to be out of date before delivery.", "correct": false, "explanation": "Production time is not the issue here, and the age of the material is a separate question from whether its claims are supported."}, {"letter": "B", "text": "Confident wording tends to be shorter, so the overview probably leaves out most of the detail.", "correct": false, "explanation": "Confidence and length are not linked in this way, and the overview described is detailed rather than sparse."}, {"letter": "C", "text": "Polished writing is harder to reformat later, so the overview will be costly to adapt for reuse.", "correct": false, "explanation": "Reformatting effort is a minor practical concern and has nothing to do with whether the content can be trusted."}, {"letter": "D", "text": "A fabricated detail reads exactly like a supported one, so fluency carries no signal about accuracy.", "correct": true, "explanation": "Polish and confidence are properties of the writing, not of the evidence behind it, so a smooth answer gives a reader no way to tell invented content from grounded content."}], "correct": "D", "select": 1}, {"id": "a2-056", "domain": 2, "task_id": "2.2", "objective": "Identify hallucinations, inconsistencies, and biases in responses", "situation": "A user asks Claude for a detailed, confident summary of a major news event that occurred after the model's reliable knowledge cutoff. Claude produces specific dates, names, and outcomes that cannot be verified in any source.", "question": "What is the most likely underlying cause of this response?", "options": [{"letter": "A", "text": "Claude hallucinated plausible-sounding details to fill a gap past its knowledge cutoff", "correct": true, "explanation": "When a query falls beyond the model's reliable knowledge cutoff and no external grounding is provided, the model can generate confident but fabricated specifics instead of acknowledging the gap."}, {"letter": "B", "text": "Claude misread the user's question and answered about an unrelated earlier event", "correct": false, "explanation": "The scenario states the summary addressed the event in question, so a topic mismatch does not explain the fabricated details."}, {"letter": "C", "text": "Claude retrieved outdated cached search results and presented them as current, verified facts", "correct": false, "explanation": "No tool use or search was described in the scenario, so there are no cached results to attribute the fabricated details to."}, {"letter": "D", "text": "Claude applied a systematic bias toward whichever outcome seemed most newsworthy", "correct": false, "explanation": "Systematic favoring of certain groups or viewpoints is bias; inventing unverifiable specifics about a recent event is a knowledge-gap hallucination."}], "correct": "A", "select": 1}, {"id": "a2-057", "domain": 2, "task_id": "2.4", "objective": "Determine when human review or additional verification is required", "situation": "A developer uses Claude to generate a database migration script intended for a production system with live customer data.", "question": "What is the appropriate next step before running the script?", "options": [{"letter": "A", "text": "Ask Claude to shorten the script so it executes faster during the production maintenance window", "correct": false, "explanation": "Optimizing for execution speed does not address whether the migration logic is correct or safe to run against live data."}, {"letter": "B", "text": "Run the script directly in production because Claude generated syntactically valid SQL for the target database", "correct": false, "explanation": "Valid syntax only confirms the script will run, not that its logic is correct or safe for the specific production schema and data."}, {"letter": "C", "text": "Have an engineer review and test the script against a staging environment before running it in production", "correct": true, "explanation": "A migration against live production data is an irreversible, high-impact action, so an engineer should review the logic and test it in staging before it ever touches production."}, {"letter": "D", "text": "Paste the script into a chat message so the team can read it before the deployment window closes", "correct": false, "explanation": "Simply pasting the script for others to glance at is not equivalent to a structured review and staging test before a production change."}], "correct": "C", "select": 1}, {"id": "a2-058", "domain": 2, "task_id": "2.2", "objective": "Identify hallucinations, inconsistencies, and biases in responses", "situation": "In a single multi-turn conversation, Claude tells a user early on that a product ships in 3 business days, then later in the same conversation states the same product ships in 7 business days without any new information being provided.", "question": "How should a QA reviewer classify this behavior?", "options": [{"letter": "A", "text": "An inconsistency, since the two statements about the same fact conflict within one chat", "correct": true, "explanation": "An inconsistency is when a model gives conflicting answers to the same question without any change in input, which is exactly what happened here."}, {"letter": "B", "text": "A latency issue, since the delay between turns caused Claude to forget its earlier answer", "correct": false, "explanation": "Response timing does not cause a model to contradict earlier output; the two turns are evaluated on content, not on elapsed time."}, {"letter": "C", "text": "A bias, since Claude favored the later answer over the earlier, correct answer given", "correct": false, "explanation": "Bias refers to skewed treatment of a topic or group, not a reviewer question about which of two conflicting numeric answers is favored."}, {"letter": "D", "text": "A hallucination, since Claude invented a shipping policy that does not exist anywhere", "correct": false, "explanation": "A shipping estimate is plausible and was actually stated twice; the problem is that the two values contradict each other, not that one is fabricated from nothing."}], "correct": "A", "select": 1}, {"id": "a2-059", "domain": 2, "task_id": "2.5", "objective": "Edit, adapt, refine, and compare outputs for the intended audience", "situation": "An incident write-up drafted for engineers now has to go to a board audience with no technical background. The draft refers to internal components by their codenames. Which two edits best adapt it for those readers?", "question": "Select two.", "options": [{"letter": "A", "text": "Replace component codenames with a plain account of what customers experienced.", "correct": true, "explanation": "Board readers cannot evaluate internal codenames, but they can weigh what customers experienced. Describing the impact in plain terms gives them something they can act on, which the codenames alone do not."}, {"letter": "B", "text": "Add a glossary of the technical terms so the original wording can all be retained.", "correct": false, "explanation": "A glossary preserves the engineering framing and asks the board to do the translation, which is the work the author should be doing for them."}, {"letter": "C", "text": "Open with the decision the board is being asked to make and what it depends on.", "correct": true, "explanation": "Reordering for the reader's purpose is a separate change from translating the vocabulary: the board needs the decision first, not after the narrative."}, {"letter": "D", "text": "Shorten every sentence to under fifteen words so the document reads more quickly.", "correct": false, "explanation": "A mechanical sentence-length rule chops explanation into fragments and does nothing about the underlying content being framed for the wrong audience."}, {"letter": "E", "text": "Keep the minute-by-minute chronology, since that is how the incident really unfolded.", "correct": false, "explanation": "An accurate chronology is still the wrong structure here. It buries impact and decisions inside a timeline the board has no reason to follow."}], "correct": ["A", "C"], "select": 2}, {"id": "a2-060", "domain": 2, "task_id": "2.5", "objective": "Edit, adapt, refine, and compare outputs for the intended audience", "situation": "A sales rep wants to compare two Claude-drafted pitches, one emphasizing cost savings and one emphasizing reliability, to see which better fits a prospective client's stated priorities.", "question": "What is the most effective way to make this comparison useful?", "options": [{"letter": "A", "text": "Ask Claude which version is generally more persuasive and use that answer regardless of the client's needs.", "correct": false, "explanation": "Relying on Claude's general persuasiveness judgment bypasses the client's specific needs and priorities, which are the critical criteria for this decision. A general claim of persuasiveness does not account for the particular context of the client's stated preferences."}, {"letter": "B", "text": "Combine both emphases into one pitch that covers cost and reliability so the client hears everything.", "correct": false, "explanation": "Merging both emphases into a single pitch dilutes the messages and prevents a clear comparison of which angle is more persuasive for this client. It also risks overwhelming the client with mixed messaging rather than testing distinct strategic approaches."}, {"letter": "C", "text": "Generate both with the same facts but different emphasis, then compare to the client's priorities.", "correct": true, "explanation": "Generating both pitches with the same factual foundation but different emphases, and then evaluating each against the client's stated priorities, enables a direct comparison that reveals which framing is more aligned. This approach ensures the final choice is data-driven and tailored to the specific audience."}, {"letter": "D", "text": "Generate only the cost-savings version, because price is usually the decisive factor in sales.", "correct": false, "explanation": "Generating only the cost-savings pitch overlooks the client's expressed priorities and assumes price is the sole deciding factor, which may not be true. It skips the comparative analysis needed to determine the best fit for this particular client."}], "correct": "C", "select": 1}, {"id": "a2-061", "domain": 2, "task_id": "2.3", "objective": "Apply fact-checking and validation techniques", "situation": "A knowledge worker needs figures on a market that changed this month, and her manager will ask where each number came from.", "question": "Which approach best supports that?", "options": [{"letter": "A", "text": "Ask the same question three times over and use whichever figures come back most consistently.", "correct": false, "explanation": "Repetition measures how stable an answer is, not whether it is true, and it still yields no source."}, {"letter": "B", "text": "Use Research, so the answer arrives with citations that she can open and check herself.", "correct": true, "explanation": "Research searches current sources and returns citations, which is what makes each figure traceable to something her manager can inspect."}, {"letter": "C", "text": "Ask in a plain chat and request that Claude add the sources it used at the end of the answer.", "correct": false, "explanation": "Sources attached to an answer that was written from general knowledge are chosen to fit the text rather than the text being drawn from them."}, {"letter": "D", "text": "Ask in a plain chat and rely on Claude's general knowledge of how that market has developed.", "correct": false, "explanation": "General knowledge will not reliably cover this month's changes, and it gives her manager nothing to inspect."}], "correct": "B", "select": 1}, {"id": "a2-062", "domain": 2, "task_id": "2.3", "objective": "Apply fact-checking and validation techniques", "situation": "A press officer has forty minutes to check a nine-hundred-word Claude-drafted briefing before it goes out. Which two kinds of claim should be verified first?", "question": "Select two.", "options": [{"letter": "A", "text": "The statements that will be quoted publicly and would be costly to retract.", "correct": true, "explanation": "This weighs consequence rather than logical importance: a claim that leaves the organisation and is hard to withdraw carries risk beyond the argument itself."}, {"letter": "B", "text": "The sentences whose phrasing reads a little unusually compared with the rest.", "correct": false, "explanation": "Odd phrasing is a weak signal and is just as likely to reflect style as error, so it is a poor way to spend a limited checking window."}, {"letter": "C", "text": "The points the team already believes, so the briefing agrees with their view.", "correct": false, "explanation": "Checking what the team already agrees with confirms existing assumptions and leaves the genuinely untested claims unexamined."}, {"letter": "D", "text": "The claims that appear earliest, on the basis that readers reach those first.", "correct": false, "explanation": "Position in the document has no bearing on risk. A wrong number in the final paragraph is exactly as damaging as one in the opening line."}, {"letter": "E", "text": "The figures that the briefing's central recommendation actually depends upon.", "correct": true, "explanation": "Load-bearing figures decide whether the recommendation stands at all, so under time pressure they return the most assurance per minute spent."}], "correct": ["A", "E"], "select": 2}, {"id": "a2-063", "domain": 2, "task_id": "2.3", "objective": "Apply fact-checking and validation techniques", "situation": "A teacher wants students to be able to check the claims in a Claude-drafted study guide for themselves.", "question": "What should she build into the guide?", "options": [{"letter": "A", "text": "A confidence rating beside each claim showing how certain Claude was when it wrote it.", "correct": false, "explanation": "A self-reported confidence score is not something a student can check against anything."}, {"letter": "B", "text": "A note at the front explaining that parts of the guide were drafted with Claude's help.", "correct": false, "explanation": "Disclosure is honest and worth doing, but it gives a student no way to check any particular statement."}, {"letter": "C", "text": "Each claim paired with the specific source and page on which a student can find it.", "correct": true, "explanation": "Claim-level sourcing is what makes verification possible for a reader, because it tells them exactly where to look for each statement."}, {"letter": "D", "text": "A list at the back of every source that was consulted while the guide was being written.", "correct": false, "explanation": "A bibliography shows what was read overall but leaves students unable to tell which source supports which claim."}], "correct": "C", "select": 1}, {"id": "a2-064", "domain": 2, "task_id": "2.5", "objective": "Edit, adapt, refine, and compare outputs for the intended audience", "situation": "A trainer asks Claude for two onboarding guides on the same topic: one for new hires with no industry background and one for experienced hires from a competitor.", "question": "What distinguishes an effective adaptation between the two versions?", "options": [{"letter": "A", "text": "Both versions include identical content, differing only in the order the sections appear", "correct": false, "explanation": "Reordering identical content without adjusting explained context does not address the different background knowledge of each audience."}, {"letter": "B", "text": "The experienced-hire version repeats every foundational term twice for reinforcement", "correct": false, "explanation": "Repeating foundational terms for experienced hires adds unnecessary content for readers who already understand that vocabulary."}, {"letter": "C", "text": "The new-hire version uses a more casual tone while the experienced version drops all punctuation", "correct": false, "explanation": "Removing punctuation does not adapt content to audience background and would harm readability for either group."}, {"letter": "D", "text": "The new-hire version defines foundational terms that the experienced version can assume the reader knows", "correct": true, "explanation": "Adapting for prior background means adjusting how much foundational context is spelled out, assuming less for readers with no experience and more for those who already know it."}], "correct": "D", "select": 1}, {"id": "a2-065", "domain": 2, "task_id": "2.3", "objective": "Apply fact-checking and validation techniques", "situation": "A policy officer has one hour before a board meeting and a Claude-written briefing containing roughly twenty factual claims. She cannot check them all.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Ask Claude which of the claims in the briefing are the ones most likely to be incorrect.", "correct": false, "explanation": "Claude has no independent view of which of its own statements are unsupported, so this ranking is generated the same way the claims were."}, {"letter": "B", "text": "Check the claims that are easiest to look up, so that the largest number get covered.", "correct": false, "explanation": "Maximising the count optimises for the wrong thing; easy claims are often the low-stakes ones."}, {"letter": "C", "text": "Work through the claims in the order they appear until the hour she has available runs out.", "correct": false, "explanation": "Order of appearance has no relationship to importance, so the claims that matter most may sit in the part she never reaches."}, {"letter": "D", "text": "Identify the few claims the board's decision actually rests on and verify those ones first.", "correct": true, "explanation": "When verification time is scarce it has to be spent where an error would change the decision, so triage by consequence comes before any checking begins."}], "correct": "D", "select": 1}, {"id": "a2-066", "domain": 2, "task_id": "2.1", "objective": "Evaluate Claude-generated outputs for accuracy and completeness", "situation": "A team lead asks Claude to review its own draft for errors. It returns a list of three corrections and states that the rest of the draft is accurate.", "question": "How should she treat that statement?", "options": [{"letter": "A", "text": "As grounds for discarding the draft, since a draft needing three corrections is unlikely to be sound elsewhere.", "correct": false, "explanation": "Three corrections in a draft are unremarkable, and discarding otherwise usable work is an overreaction to a normal review result."}, {"letter": "B", "text": "As confirmation that no further checking is needed, because a second pass has already been carried out here.", "correct": false, "explanation": "A statement that the remainder is accurate is produced by the same process as the remainder, so it adds no independent evidence."}, {"letter": "C", "text": "As a useful prompt for her own checking, since a review of its own work is not independent confirmation.", "correct": true, "explanation": "A self-review can surface internal slips such as contradictions and arithmetic, and that makes it a useful first sweep. It cannot confirm anything against the outside world, so her own checking still has to happen."}, {"letter": "D", "text": "As sufficient for the sections it names, allowing her to review only the parts where corrections were reported.", "correct": false, "explanation": "Reviewing only the flagged sections trusts the same judgement that produced the draft to decide where checking is needed."}], "correct": "C", "select": 1}, {"id": "a2-067", "domain": 2, "task_id": "2.4", "objective": "Determine when human review or additional verification is required", "situation": "A marketing operations specialist has built a workflow that will send Claude-drafted emails automatically to a fifty-thousand-person customer list at six tomorrow morning. Nobody has read any of the drafts yet.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Read the first and last few drafts as a sample, and let the send run if those ones look fine.", "correct": false, "explanation": "A sample of the first and last few tells her nothing about the drafts in between, and every one of them still goes to a real customer. Sampling is reasonable for low-stakes internal work, not for an irreversible mass send."}, {"letter": "B", "text": "Turn off the automatic send, so that nothing reaches the list until it has been reviewed.", "correct": true, "explanation": "The output is externally visible, irreversible once sent and has a very wide blast radius, so the first move is to remove the deadline pressure by stopping the automatic send. Review can then happen properly rather than against a clock she does not control."}, {"letter": "C", "text": "Read as many of the drafts as she can before six, and let the rest of them go out unread.", "correct": false, "explanation": "This accepts that unreviewed messages will reach customers anyway. Once an email is sent to fifty thousand people it cannot be recalled, so partial review under a running clock is not enough."}, {"letter": "D", "text": "Ask Claude to check its own drafts overnight and correct any problems that it finds in them.", "correct": false, "explanation": "Leaving the model to check itself keeps the unreviewed send in place. No human has looked at anything, which is precisely the risk here."}], "correct": "B", "select": 1}, {"id": "a2-068", "domain": 2, "task_id": "2.3", "objective": "Apply fact-checking and validation techniques", "situation": "A manager challenges a claim in a Claude answer, Claude replies that it is confident the claim is correct, and the manager accepts it.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The manager's challenge is recorded in the chat, which makes the exchange harder to audit.", "correct": false, "explanation": "A recorded exchange helps an audit rather than hindering it; the problem is that no external source was consulted."}, {"letter": "B", "text": "A restated assurance is not evidence, and a fabricated claim is defended just as readily.", "correct": true, "explanation": "Confidence is produced by the same process that produced the claim, so a challenge answered from inside the conversation adds nothing independent."}, {"letter": "C", "text": "Repeating the question uses allowance that could have been spent on the rest of the review.", "correct": false, "explanation": "The cost of one follow-up message is trivial next to acting on a claim that was never actually checked."}, {"letter": "D", "text": "Claude will always reverse its position when challenged, so its second answer is unreliable.", "correct": false, "explanation": "It does not always reverse, and the manager's error is treating either response as evidence about the outside world."}], "correct": "B", "select": 1}, {"id": "a2-069", "domain": 2, "task_id": "2.6", "objective": "Organize and curate information and select appropriate output formats (artifacts, inline, structured data)", "situation": "An internal communications lead has a month of scattered Claude answers about an office relocation and must turn them into one reference page for staff.", "question": "In which order should these steps be taken?", "options": [{"letter": "A", "text": "Gather the relevant answers and drop the ones that no longer apply; decide what staff need to be able to look up about the move; have Claude organise the surviving material into a single artifact; check that artifact against the questions staff are actually asking.", "correct": false, "explanation": "Curation begins before she knows what the page is for, so relevance is judged against nothing and the decision about what staff need arrives after the material has already been filtered."}, {"letter": "B", "text": "Have Claude organise the surviving material into a single artifact; decide what staff need to be able to look up about the move; gather the relevant answers and drop the ones that no longer apply; check that artifact against the questions staff are actually asking.", "correct": false, "explanation": "The artifact is assembled first, so a month of unfiltered and partly outdated answers is what gets organised, and the culling happens after the page has been built from them."}, {"letter": "C", "text": "Decide what staff need to be able to look up about the move; have Claude organise the surviving material into a single artifact; gather the relevant answers and drop the ones that no longer apply; check that artifact against the questions staff are actually asking.", "correct": false, "explanation": "The scope is set correctly but the artifact is built before the outdated answers have been removed, so superseded relocation details are carried straight into the reference page staff will rely on."}, {"letter": "D", "text": "Decide what staff need to be able to look up about the move; gather the relevant answers and drop the ones that no longer apply; have Claude organise the surviving material into a single artifact; check that artifact against the questions staff are actually asking.", "correct": true, "explanation": "What staff need to look up defines relevance, which is what makes the culling of stale answers meaningful; only then is there a clean set to organise into an artifact; and checking the finished page against real questions confirms it does the job it was scoped to do."}], "correct": "D", "select": 1}, {"id": "a2-070", "domain": 2, "task_id": "2.2", "objective": "Identify hallucinations, inconsistencies, and biases in responses", "situation": "Within a single response, Claude states in the first paragraph that a company was founded in 1998, then states in the closing paragraph that the same company celebrated its 30th anniversary last year, which would imply a different founding year.", "question": "How should this be classified during output review?", "options": [{"letter": "A", "text": "A bias, since Claude favored the more recent-sounding date over the earlier one it gave", "correct": false, "explanation": "Bias involves systematic skew toward a group or viewpoint, not a single self-contradictory pair of dates."}, {"letter": "B", "text": "A formatting error, since the two paragraphs simply use different date presentation styles here", "correct": false, "explanation": "The problem described is a factual contradiction between two claims, not merely a difference in date formatting style."}, {"letter": "C", "text": "A hallucination, since no founding date for the company exists in any known source at all", "correct": false, "explanation": "The scenario does not establish that the founding date is unverifiable or invented; the issue is that the response contradicts itself."}, {"letter": "D", "text": "An internal inconsistency, since the two statements within one response contradict each other", "correct": true, "explanation": "When two statements within the same response conflict with each other, this is an internal inconsistency, regardless of whether either individual fact can be verified externally."}], "correct": "D", "select": 1}, {"id": "a2-071", "domain": 2, "task_id": "2.2", "objective": "Identify hallucinations, inconsistencies, and biases in responses", "situation": "A project manager reviews a risk register that Claude built from incident logs. Every risk attributed to one supplier is rated high and every risk attributed to the other is rated low, although both suppliers had comparable incidents.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Ask Claude to redo the register and use the second version if it gives a more balanced split.", "correct": false, "explanation": "Choosing whichever regeneration looks more comfortable selects on appearance rather than on what the incident records actually show."}, {"letter": "B", "text": "Compare the incident records behind several of the ratings to see whether the pattern is supported.", "correct": true, "explanation": "A systematic split like this is either a real signal in the data or a skew introduced in the response, and only the underlying records can tell her which."}, {"letter": "C", "text": "Ask Claude to raise the ratings for the second supplier so that both suppliers are treated more evenly.", "correct": false, "explanation": "Forcing an even split assumes the answer. If the second supplier really does carry more risk, she has now hidden it."}, {"letter": "D", "text": "Accept the register, because a consistent pattern within a supplier suggests a genuine difference.", "correct": false, "explanation": "Consistency is exactly what a skewed response looks like. A uniform pattern is a reason to check the evidence, not a substitute for checking it."}], "correct": "B", "select": 1}, {"id": "a2-072", "domain": 2, "task_id": "2.1", "objective": "Evaluate Claude-generated outputs for accuracy and completeness", "situation": "A project manager asked for a risk register covering five workstreams and received a neat table of eleven risks. Which two checks best establish that the output is accurate and complete?", "question": "Select two.", "options": [{"letter": "A", "text": "Check that the table renders properly and that the columns line up as they should.", "correct": false, "explanation": "Rendering is a presentation matter. A perfectly formatted table can carry wrong owners and missing workstreams without any visual sign of a problem."}, {"letter": "B", "text": "Check that the wording matches the tone the sponsor prefers in written reporting.", "correct": false, "explanation": "Tone affects how the register reads, not whether its content is right. Editing for the sponsor comes after accuracy has been established."}, {"letter": "C", "text": "Confirm that all five of the named workstreams are actually represented in the table.", "correct": true, "explanation": "Completeness is measured against the request. A polished table can silently cover four workstreams, and only checking coverage item by item exposes that."}, {"letter": "D", "text": "Judge whether eleven risks is about the number a project of this size usually has.", "correct": false, "explanation": "A plausible count is not evidence. Eleven risks can look reasonable while omitting an entire workstream or duplicating two entries from another one."}, {"letter": "E", "text": "Trace the dates and owners in the table back to the project plan they came from.", "correct": true, "explanation": "This tests accuracy rather than coverage: the entries must match the source plan, and a risk with an invented owner is wrong even if the scope is complete."}], "correct": ["C", "E"], "select": 2}, {"id": "a2-073", "domain": 2, "task_id": "2.3", "objective": "Apply fact-checking and validation techniques", "situation": "A team lead is writing a short internal guide on what to do when Claude returns an answer that looks incorrect or misleading. Which two actions does Claude's own guidance recommend?", "question": "Select two.", "options": [{"letter": "A", "text": "Turn off web search so that answers come from the model alone and stay consistent.", "correct": false, "explanation": "Removing web search removes the citations that make checking possible. Consistency is not accuracy, and this makes verification harder rather than easier."}, {"letter": "B", "text": "Use the thumbs-down control on the reply so the problem is reported as feedback.", "correct": true, "explanation": "The feedback control is the documented route for flagging an unhelpful or incorrect response, and it is the step users most often skip."}, {"letter": "C", "text": "Regenerate the answer several times and keep whichever version the team likes most.", "correct": false, "explanation": "Choosing the version the team prefers selects for appeal rather than truth, and repeated generations can repeat the same underlying error."}, {"letter": "D", "text": "Ask Claude to score its own confidence and only accept answers above a chosen level.", "correct": false, "explanation": "A self-reported confidence score is produced in the same way as the answer, so it cannot serve as independent evidence that the answer is sound."}, {"letter": "E", "text": "Open the sources Claude cited and read them, since the originals reveal misreadings.", "correct": true, "explanation": "Reviewing cited sources is a distinct recommendation from giving feedback: checking the original content is what reveals anything that was misinterpreted."}], "correct": ["B", "E"], "select": 2}, {"id": "a2-074", "domain": 2, "task_id": "2.5", "objective": "Edit, adapt, refine, and compare outputs for the intended audience", "situation": "A technical writer generates two drafts of an API integration guide, one for experienced developers and one for low-code platform users, and needs to check them for consistency before publishing.", "question": "What should the writer confirm?", "options": [{"letter": "A", "text": "Both drafts must describe the same steps and results, varying only in terminology and depth.", "correct": true, "explanation": "Both drafts should describe the same underlying steps and outcomes, ensuring consistency across audiences. The only acceptable differences are in terminology and level of detail appropriate for each user group."}, {"letter": "B", "text": "The writer should check that both drafts have identical code samples to ensure uniformity.", "correct": false, "explanation": "Requiring identical code samples overlooks the fact that experienced developers and low-code platform users interact with the API differently. Low-code users may not use code samples at all, or they might need platform-specific snippets that differ from the developer draft."}, {"letter": "C", "text": "The low-code draft should omit authentication, as low-code users only need the base URL.", "correct": false, "explanation": "Low-code users still need authentication information to use the API securely. Omitting authentication entirely and providing only a base URL would leave them with incomplete setup instructions."}, {"letter": "D", "text": "The developer draft should simply be essentially a condensed version of the low-code guide.", "correct": false, "explanation": "The developer draft should not be a condensed version of the low-code guide; each audience requires distinct structure and depth. Developers need in-depth technical details and code examples, while low-code users need simplified configuration steps."}], "correct": "A", "select": 1}, {"id": "a2-075", "domain": 2, "task_id": "2.5", "objective": "Edit, adapt, refine, and compare outputs for the intended audience", "situation": "An operations manager compares two drafts of a policy announcement, one for headquarters staff and one adapted for a regional office, to confirm both convey the same policy without confusing readers.", "question": "What should the manager check for?", "options": [{"letter": "A", "text": "Only the headquarters draft should be used, since any rewrite risks altering policy's meaning.", "correct": false, "explanation": "While rewriting carries some risk, careful adaptation ensures the policy is clear to each audience without changing meaning. Using only the headquarters draft ignores the regional office's needs and may lead to misinterpretation."}, {"letter": "B", "text": "Verify both drafts have the same number of words so neither office feels treated differently.", "correct": false, "explanation": "Word count is irrelevant to whether the drafts communicate the policy consistently. The goal is clarity and accuracy, not equal length, and focusing on word count does not ensure the policy is correctly conveyed."}, {"letter": "C", "text": "Check that the regional draft uses more formal language than headquarters to convey authority.", "correct": false, "explanation": "The formality level should match the audience's expectations, not be used to assert authority. The regional draft may actually require less formal language if that suits the readers, and imposing formal language could hinder understanding."}, {"letter": "D", "text": "That the drafts convey identical rules and deadlines but use wording tailored to each audience.", "correct": true, "explanation": "The drafts should convey identical rules and deadlines, but the wording can be tailored to each audience. The manager should verify that both versions communicate the same policy accurately while using appropriate language for each group."}], "correct": "D", "select": 1}, {"id": "a2-076", "domain": 2, "task_id": "2.5", "objective": "Edit, adapt, refine, and compare outputs for the intended audience", "situation": "An analyst has condensed a twenty-page Claude-produced analysis into a one-page version for an executive committee that will act on it directly.", "question": "Which check matters most before she circulates the shorter version?", "options": [{"letter": "A", "text": "That the one-page version reads in the same house style as the committee papers her team has circulated in previous quarters.", "correct": false, "explanation": "House style helps a paper land well with a familiar audience, but it is a presentation concern. A perfectly styled summary can still have lost the qualification that made the recommendation safe."}, {"letter": "B", "text": "That the one-page version comes in under the length the committee asked for, since executive committee papers are capped at a single side.", "correct": false, "explanation": "Meeting the length limit is a formatting requirement that must be satisfied anyway, and it is easy to check. It says nothing about whether the shortened paper still supports the decision being made."}, {"letter": "C", "text": "That the one-page version is worded differently enough from the long analysis for the two documents to be told apart easily.", "correct": false, "explanation": "Distinguishing the two documents is a filing convenience at best. Deliberately rewording for difference risks introducing changes of meaning rather than protecting the committee from acting on a distorted summary."}, {"letter": "D", "text": "That nothing removed in the condensing changes the recommendation, particularly any condition on which that recommendation depends.", "correct": true, "explanation": "Adapting for an audience must not change what the audience will do. A caveat or condition dropped in the interests of brevity turns a qualified recommendation into an unqualified one for the people acting on it."}], "correct": "D", "select": 1}, {"id": "a2-077", "domain": 2, "task_id": "2.3", "objective": "Apply fact-checking and validation techniques", "situation": "A financial analyst asks a Claude-powered assistant, \"What was Acme Corp's closing stock price yesterday?\" The assistant has no tools enabled and answers with a specific dollar figure stated confidently.", "question": "What is the most likely validation concern with this response?", "options": [{"letter": "A", "text": "The figure is unreliable because stock prices are personal financial data that Claude is restricted from ever discussing", "correct": false, "explanation": "Public stock prices are not restricted personal data, so this is not the relevant validation issue."}, {"letter": "B", "text": "The response is untrustworthy because Claude always declines to answer any question that involves specific numbers", "correct": false, "explanation": "Claude does not categorically refuse numeric answers; the concern here is factual grounding, not refusal behavior."}, {"letter": "C", "text": "Without a tool that retrieves external data, the figure is likely fabricated since Claude cannot know yesterday's actual price", "correct": true, "explanation": "Claude's knowledge comes from training data with a cutoff and does not include live market data. Without a tool like web search to fetch current information, a specific recent stock price is likely a hallucination and should be treated as unverified."}, {"letter": "D", "text": "The response cannot be trusted because Claude's training data updates in real time and may include conflicting price figures", "correct": false, "explanation": "Claude's training data has a fixed cutoff and is not updated in real time, which is precisely why an ungrounded answer about yesterday's price is risky."}], "correct": "C", "select": 1}, {"id": "a2-078", "domain": 2, "task_id": "2.4", "objective": "Determine when human review or additional verification is required", "situation": "A company uses Claude to process customer support tickets that contain personal data and generate draft responses referencing account details.", "question": "Before these drafts are sent, what should the workflow include?", "options": [{"letter": "A", "text": "The company adds a footer noting that personal data may appear in the message before sending it", "correct": false, "explanation": "A disclaimer footer does not verify that the personal data in the draft is accurate or appropriately handled."}, {"letter": "B", "text": "A support agent reviews each draft to confirm the personal data referenced is accurate and handled properly", "correct": true, "explanation": "Responses referencing personal data carry privacy and accuracy risk, so a support agent should review each draft before it reaches the customer."}, {"letter": "C", "text": "The support team increases the model's response length so more context is included automatically", "correct": false, "explanation": "A longer response does not address the accuracy or privacy handling of the personal data referenced in the draft."}, {"letter": "D", "text": "Send the drafts automatically since the account details came straight from the customer's own support ticket", "correct": false, "explanation": "Data originating in the customer's ticket does not guarantee the model referenced or handled it correctly in the generated draft."}], "correct": "B", "select": 1}, {"id": "a2-079", "domain": 2, "task_id": "2.5", "objective": "Edit, adapt, refine, and compare outputs for the intended audience", "situation": "A physician asks Claude to draft the same diagnosis explanation for a patient handout and for a note in the clinical record. The first draft uses identical clinical shorthand in both.", "question": "What is the most appropriate refinement for the patient-facing version?", "options": [{"letter": "A", "text": "Rewrite the handout in plain language for patients, while the clinical note keeps precise terms.", "correct": true, "explanation": "Rewriting the handout in plain language ensures patients can understand and act on the information, while the clinical note retains precise terminology appropriate for clinicians."}, {"letter": "B", "text": "Keep the clinical shorthand in the handout but print it using a larger, easier-to-read font size.", "correct": false, "explanation": "Increasing font size improves readability but does not address the core issue that clinical shorthand is not understandable to patients. They need plain language, not just larger text."}, {"letter": "C", "text": "Shorten both the handout and the clinical note to match in length so they appear equally brief.", "correct": false, "explanation": "Matching the length of the two documents does not make the clinical shorthand understandable to patients. The issue is content clarity, not document length."}, {"letter": "D", "text": "Add a brief plain-language summary above the clinical shorthand to aid patient comprehension.", "correct": false, "explanation": "Adding a plain-language summary does not eliminate the confusing clinical shorthand from the handout. The entire document should be in accessible language for patients."}], "correct": "A", "select": 1}, {"id": "a2-080", "domain": 2, "task_id": "2.6", "objective": "Organize and curate information and select appropriate output formats (artifacts, inline, structured data)", "situation": "An operations manager asks Claude a quick factual question during a meeting and receives a long document in a side panel that she has to open.", "question": "What is the most useful adjustment?", "options": [{"letter": "A", "text": "Copy the document's contents back into the chat so that the answer appears in both places.", "correct": false, "explanation": "Duplicating the same content in two places adds clutter without addressing why the format was wrong."}, {"letter": "B", "text": "Ask for a short answer in the conversation itself, since she needs to read it immediately.", "correct": true, "explanation": "Format should follow how the answer will be used. A one-off answer read on the spot belongs in the conversation, not in a self-contained document meant for reuse."}, {"letter": "C", "text": "Ask Claude to make the document shorter, so that it takes less time for her to read through.", "correct": false, "explanation": "A shorter document is still a document she has to open, which is the friction she is trying to remove."}, {"letter": "D", "text": "Ask the question again later, once the meeting she is currently sitting in has finished.", "correct": false, "explanation": "Delaying the answer does not change the format, and she needed the information during the meeting."}], "correct": "B", "select": 1}, {"id": "a2-081", "domain": 2, "task_id": "2.6", "objective": "Organize and curate information and select appropriate output formats (artifacts, inline, structured data)", "situation": "A designer asks Claude to create a company logo as a scalable vector graphic that they can drop directly into their design tool.", "question": "Which output approach is appropriate?", "options": [{"letter": "A", "text": "An inline text description of the logo's appearance, since design tools can convert plain descriptions into vector paths automatically", "correct": false, "explanation": "Design tools do not automatically convert prose descriptions into vector paths, so plain text would not give the designer a usable asset."}, {"letter": "B", "text": "An artifact containing the SVG markup, since SVG images are a supported artifact content type meant for standalone, reusable visual assets", "correct": true, "explanation": "SVG images are explicitly a supported artifact type, and a logo the designer will reuse in another tool is exactly the standalone-asset case artifacts handle."}, {"letter": "C", "text": "A strict tool call to a design_logo tool, since strict tool use is required whenever an image-like asset needs to be generated", "correct": false, "explanation": "Strict tool use validates parameters passed to a defined tool; it is not a mechanism for generating or delivering image assets."}, {"letter": "D", "text": "A structured JSON response describing the logo's colors and shapes, since JSON outputs are the recommended format for any visual design asset", "correct": false, "explanation": "JSON outputs are meant for structured data fields, not for producing renderable vector graphics that a design tool can import directly."}], "correct": "B", "select": 1}, {"id": "a2-082", "domain": 2, "task_id": "2.6", "objective": "Organize and curate information and select appropriate output formats (artifacts, inline, structured data)", "situation": "In a single conversation, a user has asked Claude to create three separate artifacts: a Python script, an HTML landing page, and a Markdown report. The user now asks Claude to update \"the landing page\" to change its header color.", "question": "What should Claude do?", "options": [{"letter": "A", "text": "Create a fourth, brand-new artifact containing just the updated header, since existing artifacts cannot be selected for further edits", "correct": false, "explanation": "Existing artifacts can be selected and updated directly; a disconnected new artifact would not represent an edit to the original landing page."}, {"letter": "B", "text": "Ask the user to delete the other two artifacts first, since only one artifact can exist in a conversation at any given time", "correct": false, "explanation": "Multiple artifacts can coexist in the same conversation; deleting the others is not required to edit one of them."}, {"letter": "C", "text": "Identify the HTML landing page artifact among the three open artifacts and apply the header color change to that specific one", "correct": true, "explanation": "Claude.ai supports multiple artifacts within one conversation, and Claude can identify which specific artifact the user is referring to and update that one."}, {"letter": "D", "text": "Update all three artifacts with the header color change, since a single conversation can only track edits collectively across every open artifact", "correct": false, "explanation": "Edits target the specific artifact identified by the user's request, not every artifact in the conversation."}], "correct": "C", "select": 1}, {"id": "a2-083", "domain": 2, "task_id": "2.6", "objective": "Organize and curate information and select appropriate output formats (artifacts, inline, structured data)", "situation": "A manager needs a comparison of six suppliers across five criteria. The information is currently spread across several separate Claude answers.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Ask Claude to combine all of the earlier answers into one long document she can read through.", "correct": false, "explanation": "Combining unreconciled answers produces a long document that still contains whatever gaps and conflicts the originals had."}, {"letter": "B", "text": "Agree which suppliers and which criteria belong in the comparison before asking for any output.", "correct": true, "explanation": "A comparison is only as good as its frame. Fixing the rows and columns first is what stops the result being shaped by whatever happened to come up in earlier chats."}, {"letter": "C", "text": "Choose the output format now, so that every further answer can be produced in that same shape.", "correct": false, "explanation": "Format is a downstream decision; choosing it before the content is defined locks in a shape that may not fit."}, {"letter": "D", "text": "Ask Claude to rank the six suppliers straight away and work backwards from the ranking it gives.", "correct": false, "explanation": "A ranking produced before the criteria are agreed reflects unstated assumptions she cannot inspect or defend."}], "correct": "B", "select": 1}, {"id": "a2-084", "domain": 2, "task_id": "2.4", "objective": "Determine when human review or additional verification is required", "situation": "A Claude-drafted notice about a service change is going out to every customer tomorrow.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Have the accountable owner review it; publish the notice to customers; check the facts against the source; obtain the required sign-off.", "correct": false, "explanation": "The owner reviews unverified content, and the notice is published before either the fact check or the sign-off has happened."}, {"letter": "B", "text": "Check the facts against the source; have the accountable owner review it; obtain the required sign-off; publish the notice to customers.", "correct": true, "explanation": "Factual checking comes first because the owner's judgment depends on accurate content, sign-off follows a completed review, and publication is the last irreversible step."}, {"letter": "C", "text": "Publish the notice to customers; check the facts against the source; have the accountable owner review it; obtain the required sign-off.", "correct": false, "explanation": "Publication comes before every check, which means any error reaches customers before anyone has looked for it."}, {"letter": "D", "text": "Obtain the required sign-off; check the facts against the source; publish the notice to customers; have the accountable owner review it.", "correct": false, "explanation": "Sign-off is given before anything has been checked, and the owner's review lands after customers have already received the notice."}], "correct": "B", "select": 1}, {"id": "a2-085", "domain": 2, "task_id": "2.5", "objective": "Edit, adapt, refine, and compare outputs for the intended audience", "situation": "A support lead finds a Claude-drafted troubleshooting article accurate but written at a reading level too advanced for the average customer.", "question": "Which follow-up instruction best refines the draft without sacrificing accuracy?", "options": [{"letter": "A", "text": "Rewrite the article with shorter sentences and everyday words, keeping every step accurate", "correct": true, "explanation": "Simplifying sentence structure and vocabulary while preserving the accurate steps directly targets the readability problem without losing correctness."}, {"letter": "B", "text": "Remove the least common troubleshooting steps so the article reads a bit shorter", "correct": false, "explanation": "Removing steps changes the article's completeness and could omit cases some customers need, which does not fix a reading-level problem."}, {"letter": "C", "text": "Add more technical detail so customers understand the underlying cause in full", "correct": false, "explanation": "Adding more technical detail moves the article further from, not closer to, the audience's reading level."}, {"letter": "D", "text": "Convert the article into a formal specification format used by the engineering team", "correct": false, "explanation": "A formal specification format is written for engineers and would make the article harder for typical customers to follow."}], "correct": "A", "select": 1}, {"id": "a2-086", "domain": 2, "task_id": "2.3", "objective": "Apply fact-checking and validation techniques", "situation": "A procurement officer will ask Claude to summarise an uploaded framework agreement into about forty points, and a colleague will have to spot-check that summary against the agreement in fifteen minutes.", "question": "What should she build into the request to make that check practical?", "options": [{"letter": "A", "text": "Ask for each point to name the clause of the agreement it came from, so the colleague can go straight to that clause and compare.", "correct": true, "explanation": "Tying each point to its clause turns verification into a direct comparison at source, which is the only step that shows whether the agreement was represented accurately, and it fits the time available."}, {"letter": "B", "text": "Ask for the points to be ordered by how commercially important each one is, so the colleague starts with the points that matter most.", "correct": false, "explanation": "Importance ordering helps the colleague choose what to look at, but it does not help him find the corresponding clause in a long agreement, which is where the fifteen minutes would actually be spent."}, {"letter": "C", "text": "Ask for the summary to be kept to short single sentences, so the colleague can read the whole of it well inside the fifteen minutes.", "correct": false, "explanation": "Brevity makes the summary quicker to read but no quicker to verify. The colleague can read every point and still be unable to locate the clause each one is supposed to rest on."}, {"letter": "D", "text": "Ask for a confidence rating beside each point, so the colleague can concentrate the checking effort on the points rated least certain.", "correct": false, "explanation": "A self-declared confidence rating is not a reliable guide to which points are wrong, and it can misdirect the check by giving high confidence to a point that was misread from the agreement."}], "correct": "A", "select": 1}, {"id": "a2-087", "domain": 2, "task_id": "2.2", "objective": "Identify hallucinations, inconsistencies, and biases in responses", "situation": "A team lead is explaining hallucinations to colleagues who are new to using Claude at work. Which two statements should she include because they are accurate?", "question": "Select two.", "options": [{"letter": "A", "text": "They occur mainly in long responses, so keeping answers short avoids the problem well.", "correct": false, "explanation": "Brevity does not confer accuracy. A one-line answer can contain an invented figure just as easily as a long one, and offers fewer clues to catch it."}, {"letter": "B", "text": "They only occur when Claude has no access to the web and must rely on what it knows.", "correct": false, "explanation": "Web access changes where material comes from but does not remove the possibility of a misstatement, including a source being summarised incorrectly."}, {"letter": "C", "text": "They can appear in answers that read fluently and hold together from start to finish.", "correct": true, "explanation": "Fluency and internal coherence are not evidence of truth, which is exactly why colleagues who judge by readability tend to miss fabricated content."}, {"letter": "D", "text": "They are a known limit of current frontier models, not a sign of a broken setup.", "correct": true, "explanation": "Framing this as a known model limitation is a separate and important point: colleagues should not assume a setting was configured wrongly and stop checking."}, {"letter": "E", "text": "They can be removed entirely by instructing Claude to answer only when it is certain.", "correct": false, "explanation": "No instruction eliminates the behaviour. Asking for certainty can suppress useful hedging while leaving confident errors in place."}], "correct": ["C", "D"], "select": 2}, {"id": "a2-088", "domain": 2, "task_id": "2.1", "objective": "Evaluate Claude-generated outputs for accuracy and completeness", "situation": "A budget summary written in prose gives departmental figures and then states a total. The finance officer reads it and finds nothing that looks out of place.", "question": "What is the greatest risk in accepting it?", "options": [{"letter": "A", "text": "The stated total may not be the sum of the figures given, and prose gives no visual cue that anything is wrong.", "correct": true, "explanation": "A total written as a sentence carries no arithmetic behind it, and prose hides the mismatch that a column of figures would make obvious. It has to be re-added by hand before the summary is trusted."}, {"letter": "B", "text": "The summary may express figures in thousands where the finance team's own reports express them in millions.", "correct": false, "explanation": "A units mismatch is a real hazard, but it usually shows up as figures that are wildly out of scale and so is caught on reading."}, {"letter": "C", "text": "The departments may be listed in a slightly different order from the one that the finance team's reports normally use.", "correct": false, "explanation": "Ordering affects how quickly a reader finds a department and does not change any of the values reported."}, {"letter": "D", "text": "The prose format makes the summary harder to move into the reporting spreadsheet the finance team uses.", "correct": false, "explanation": "Transferring prose into a spreadsheet is inconvenient, but it is handling effort rather than a threat to the numbers themselves."}], "correct": "A", "select": 1}, {"id": "a2-089", "domain": 2, "task_id": "2.5", "objective": "Edit, adapt, refine, and compare outputs for the intended audience", "situation": "A UX writer is unhappy that a Claude-drafted in-app error message describes the internal exception name and stack trace to end users.", "question": "What is the most appropriate refinement for this audience?", "options": [{"letter": "A", "text": "Add a short apology sentence before the existing exception details in the message", "correct": false, "explanation": "Adding an apology does not address the underlying problem that the message still exposes internal exception details."}, {"letter": "B", "text": "Rewrite the message in plain terms describing what happened and what to do next", "correct": true, "explanation": "End users need a clear description of the problem and next steps in plain language; internal exception details are not meaningful to this audience and should be removed."}, {"letter": "C", "text": "Keep the exception name but reformat the stack trace as a shorter numbered list", "correct": false, "explanation": "Reformatting the stack trace as a list still exposes internal implementation detail that is not useful to end users."}, {"letter": "D", "text": "Move the stack trace into a pop-up that appears only when the user hovers over it", "correct": false, "explanation": "Hiding the stack trace behind a hover interaction still surfaces internal detail instead of replacing it with a plain-language message."}], "correct": "B", "select": 1}, {"id": "a2-090", "domain": 2, "task_id": "2.4", "objective": "Determine when human review or additional verification is required", "situation": "A team applies the same single sign-off step to every Claude-assisted output, from internal meeting notes to a statement that will be filed with a regulator.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "Irreversible, externally visible work gets no more scrutiny than notes nobody will ever act on.", "correct": true, "explanation": "Review effort should scale with consequence and reversibility. A uniform step spends the same attention everywhere, which leaves the rare high-blast-radius output under-examined."}, {"letter": "B", "text": "Reviewers will refuse the extra workload, so the sign-off step will be abandoned within weeks.", "correct": false, "explanation": "Workload may become a problem, but the immediate exposure is that the process cannot tell important work from routine work."}, {"letter": "C", "text": "Reviewers will start to rely on Claude's wording, and their comments will grow shorter over time.", "correct": false, "explanation": "Review fatigue is real, but it is a consequence of the flat process rather than the central danger it creates."}, {"letter": "D", "text": "Every output will take equally long to clear, which slows the team's routine work considerably.", "correct": false, "explanation": "Slower routine work is an efficiency cost. The serious risk sits at the other end of the range, on the filings."}], "correct": "A", "select": 1}, {"id": "a2-091", "domain": 2, "task_id": "2.6", "objective": "Organize and curate information and select appropriate output formats (artifacts, inline, structured data)", "situation": "A user in Claude.ai asks Claude to write a 200-line Python script that parses log files and generates a summary report. The user says they plan to save the script and run it repeatedly.", "question": "How should Claude present the script?", "options": [{"letter": "A", "text": "Split across several separate chat messages, because breaking long scripts into chunks makes each message easier to read and manage.", "correct": false, "explanation": "Splitting the script across multiple messages may make individual chunks easier to read, but it fragments the code and complicates copying, editing, and reusing it as a single functional unit."}, {"letter": "B", "text": "The script should be inline in the chat response, since code blocks provide helpful syntax highlighting and an easy copy mechanism.", "correct": false, "explanation": "While inline code blocks offer syntax highlighting and a convenient copy mechanism, a script of this length and purpose is better suited as an artifact, which allows the user to save, edit, and run it as a standalone file without cluttering the conversation."}, {"letter": "C", "text": "Claude should present the script as an artifact, since it's a substantial, self-contained work for repeated reuse and iteration.", "correct": true, "explanation": "An artifact is designed for substantial, self-contained content that the user can edit, save, and reuse independently of the chat, making it ideal for a 200-line script intended for repeated iteration."}, {"letter": "D", "text": "Claude should provide the script as a downloadable PDF, since the user's primary need is a formatted report, not the source code.", "correct": false, "explanation": "The user's primary need is the runnable Python script itself, not a formatted report; providing a downloadable PDF would prevent the script from being executed or modified, failing to meet the explicit request for a reusable code file."}], "correct": "C", "select": 1}, {"id": "a2-092", "domain": 2, "task_id": "2.3", "objective": "Apply fact-checking and validation techniques", "situation": "A marketer asks Claude to add citations to a draft that Claude had already written from general knowledge.", "question": "What is the main concern?", "options": [{"letter": "A", "text": "Claude is unable to add citations to text it has already produced earlier in a conversation.", "correct": false, "explanation": "It will readily do so, which is precisely what makes the practice risky rather than impossible."}, {"letter": "B", "text": "Citations added afterwards cannot be formatted in the style that the publication requires.", "correct": false, "explanation": "Formatting is adjustable and is not what makes retrofitted citations unreliable."}, {"letter": "C", "text": "Adding citations later makes the draft longer, which the publication's word limit forbids.", "correct": false, "explanation": "Length is a practical constraint and is unrelated to whether the citations support the text."}, {"letter": "D", "text": "Sources found to fit finished text may not actually support the claims attached to them.", "correct": true, "explanation": "When the text comes first, the sources are selected to match wording that was never derived from them, so the citation can look right while supporting nothing."}], "correct": "D", "select": 1}, {"id": "a2-093", "domain": 2, "task_id": "2.5", "objective": "Edit, adapt, refine, and compare outputs for the intended audience", "situation": "A nonprofit needs a grant report for both a technical evaluator and a general public donor newsletter, but the first draft uses terms like 'confidence interval' and 'effect size' throughout.", "question": "What is the best way to adapt the newsletter version?", "options": [{"letter": "A", "text": "Remove all numbers from the newsletter so no explanation of findings is needed", "correct": false, "explanation": "Removing all numbers strips out meaningful impact information donors would want, rather than simply making it understandable."}, {"letter": "B", "text": "Restate the statistical findings in plain language describing what the results mean practically", "correct": true, "explanation": "Translating statistical jargon into plain, practical statements is exactly what adapting content for a general donor audience requires while preserving the actual findings."}, {"letter": "C", "text": "Keep the report identical for both audiences since the underlying data does not change", "correct": false, "explanation": "Using the identical technical report for donors ignores that the intended audience has a different background and needs different phrasing."}, {"letter": "D", "text": "Keep the statistical terms but define each one in a footnote at the end of the newsletter", "correct": false, "explanation": "Footnoted definitions still leave jargon embedded in the main text, requiring donors to cross-reference terms instead of reading naturally."}], "correct": "B", "select": 1}, {"id": "a2-094", "domain": 2, "task_id": "2.3", "objective": "Apply fact-checking and validation techniques", "situation": "A team wants its fact-checking of Claude outputs to be reviewable by someone else months later.", "question": "Which practice best achieves that?", "options": [{"letter": "A", "text": "Keep the full chat transcript, since it shows every single question the team put to Claude.", "correct": false, "explanation": "A transcript shows how the draft was produced but not which claims were subsequently checked or against which sources."}, {"letter": "B", "text": "Record which claims were checked and against what, and keep that alongside the document.", "correct": true, "explanation": "A record of coverage lets a later reviewer see what was verified and what was not, instead of having to redo the work or assume it was done."}, {"letter": "C", "text": "Have a second person read the finished document and confirm that it reads correctly.", "correct": false, "explanation": "A readability pass does not establish that any statement was traced to a source."}, {"letter": "D", "text": "Add a standard line to the document stating that all claims in it have been verified.", "correct": false, "explanation": "An unsupported assurance is not reviewable, and it obscures which claims were actually examined."}], "correct": "B", "select": 1}, {"id": "a2-095", "domain": 2, "task_id": "2.3", "objective": "Apply fact-checking and validation techniques", "situation": "An analyst has a Research-mode output on competitor pricing, with citations against most claims, that will inform a pricing decision. Which two validation steps are most useful?", "question": "Select two.", "options": [{"letter": "A", "text": "Confirm the headline figure against the company's own records rather than one source.", "correct": true, "explanation": "Bringing in an independent record is a different technique from checking sources: it avoids treating Claude as a singular source of truth on a decision that matters."}, {"letter": "B", "text": "Run the same research request again and accept the result if the two broadly agree.", "correct": false, "explanation": "Two runs can be wrong in the same way, so agreement between them is a consistency signal rather than any evidence about accuracy."}, {"letter": "C", "text": "Open a sample of cited pages and confirm each says what the summary claims.", "correct": true, "explanation": "A citation shows where a claim came from, not that it was read correctly. Opening the page is what catches a figure taken out of context or from the wrong year."}, {"letter": "D", "text": "Use the number of citations shown as a rough measure of how reliable the output is.", "correct": false, "explanation": "Citation count measures volume, not quality. A heavily cited output can rest on weak sources, and a lightly cited one can be entirely sound."}, {"letter": "E", "text": "Accept anything carrying a citation and spend the time only on uncited statements.", "correct": false, "explanation": "A citation is not a guarantee, and cited claims are precisely the ones that feel safe to skip, which is what lets a misread source pass through unchallenged."}], "correct": ["A", "C"], "select": 2}, {"id": "a2-096", "domain": 2, "task_id": "2.1", "objective": "Evaluate Claude-generated outputs for accuracy and completeness", "situation": "A strategy lead is drafting a board paper and Claude supplies a precise market-size figure for a sector she does not follow. The figure would strengthen her case considerably.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Compare it against the market sizes of neighbouring sectors to see whether its scale looks about right.", "correct": false, "explanation": "A plausibility check may catch a figure that is wildly wrong, but it comfortably passes a number that is merely incorrect."}, {"letter": "B", "text": "Establish where the figure comes from and check that published source before the figure goes anywhere.", "correct": true, "explanation": "A precise figure for an unfamiliar sector is exactly the kind of claim that has to be traced to a published source before it is used. Provenance comes before presentation."}, {"letter": "C", "text": "Include it while describing it as an estimate, which signals to the board that the precise value may move.", "correct": false, "explanation": "Calling a figure an estimate lowers its apparent precision but still places an unsourced number in front of the board as evidence."}, {"letter": "D", "text": "Ask for the figure again in a new conversation, and use it if the second answer gives the same value.", "correct": false, "explanation": "A repeated answer shows only that the same figure is produced twice, which is consistency rather than any external support."}], "correct": "B", "select": 1}, {"id": "a2-097", "domain": 2, "task_id": "2.3", "objective": "Apply fact-checking and validation techniques", "situation": "A university course administrator must confirm that a Claude-drafted section of the student handbook matches the university's published regulations.", "question": "In which order should these steps be taken?", "options": [{"letter": "A", "text": "Find each of those statements in the published regulations; identify the statements in the section that assert a regulation; correct any statement the regulations do not support; record which version of the regulations was used.", "correct": false, "explanation": "She would be looking statements up before deciding which statements matter, so the search has no defined scope and ordinary descriptive sentences get treated as regulatory claims."}, {"letter": "B", "text": "Identify the statements in the section that assert a regulation; find each of those statements in the published regulations; correct any statement the regulations do not support; record which version of the regulations was used.", "correct": true, "explanation": "Singling out the sentences that make a regulatory claim scopes the work; each of those is then traced to the published regulations, which is the authoritative source rather than the draft itself; corrections follow from what that comparison found; and recording the version used makes the check repeatable when the regulations change."}, {"letter": "C", "text": "Correct any statement the regulations do not support; identify the statements in the section that assert a regulation; find each of those statements in the published regulations; record which version of the regulations was used.", "correct": false, "explanation": "Editing comes first of all, before either the claims have been identified or the regulations opened, so the handbook is changed on no evidence at all."}, {"letter": "D", "text": "Identify the statements in the section that assert a regulation; correct any statement the regulations do not support; find each of those statements in the published regulations; record which version of the regulations was used.", "correct": false, "explanation": "Corrections are made before the regulations have been consulted, which means they rest on what she assumes the rules say. The lookup that should have driven the edits arrives after them."}], "correct": "B", "select": 1}, {"id": "a2-098", "domain": 2, "task_id": "2.1", "objective": "Evaluate Claude-generated outputs for accuracy and completeness", "situation": "A manager has fifteen minutes to check a six-page recommendation before a decision meeting. She cannot verify everything in it.", "question": "What should she concentrate on?", "options": [{"letter": "A", "text": "The passages that are written most confidently, because strongly worded claims carry the most weight in a meeting.", "correct": false, "explanation": "Confidence of expression is unrelated to whether a claim is load-bearing, and a quietly stated figure can be the one the recommendation turns on."}, {"letter": "B", "text": "An even sample drawn from each of the six pages, so that no part of the document goes entirely unexamined.", "correct": false, "explanation": "Even sampling spreads scarce attention across material of very unequal importance, so a decisive figure gets the same scrutiny as background."}, {"letter": "C", "text": "The opening and closing sections, since these are what most attendees will actually read before the meeting.", "correct": false, "explanation": "Checking what will be read protects the meeting's impression of the document but leaves the reasoning behind the recommendation unexamined."}, {"letter": "D", "text": "The few facts the recommendation depends on, since the decision changes if any one of them turns out to be wrong.", "correct": true, "explanation": "With limited time, verification should follow consequence. Identifying the handful of facts that would reverse the recommendation concentrates the check where an error would actually change the outcome."}], "correct": "D", "select": 1}, {"id": "a2-099", "domain": 2, "task_id": "2.6", "objective": "Organize and curate information and select appropriate output formats (artifacts, inline, structured data)", "situation": "A communications officer needs a one-page summary to hand out at an event, with the supporting detail available for anyone who asks afterwards.", "question": "How should she organise the material?", "options": [{"letter": "A", "text": "Produce one long document containing both the summary and every piece of supporting detail.", "correct": false, "explanation": "A single document that tries to do both is too long to hand out and too summary-led to serve as a reference."}, {"letter": "B", "text": "Produce the handout only, and answer any requests for supporting detail verbally on the day.", "correct": false, "explanation": "Relying on recall leaves the detail unavailable afterwards and inconsistent between conversations."}, {"letter": "C", "text": "Produce a short handout as the deliverable and keep the detail as a separate reference document.", "correct": true, "explanation": "Two audiences with different needs are served by two artefacts, so neither the handout nor the reference is compromised to accommodate the other."}, {"letter": "D", "text": "Produce the supporting detail only, since anyone wanting a summary can read its opening section.", "correct": false, "explanation": "An opening section written for a detailed document is not a handout, and event audiences will not read the rest."}], "correct": "C", "select": 1}, {"id": "a2-100", "domain": 2, "task_id": "2.4", "objective": "Determine when human review or additional verification is required", "situation": "A financial services firm asks Claude to generate personalized investment recommendations for retail clients based on their stated goals.", "question": "Before any recommendation is sent to a client, what should happen?", "options": [{"letter": "A", "text": "A licensed financial advisor reviews the recommendation for suitability and regulatory compliance", "correct": true, "explanation": "Investment advice is a regulated, high-stakes domain, so a licensed advisor must confirm suitability and compliance before a recommendation reaches a client."}, {"letter": "B", "text": "The recommendation is sent as-is since it was generated from the client's own stated goals", "correct": false, "explanation": "Basing the recommendation on client-stated goals does not eliminate the risk of an unsuitable or non-compliant suggestion reaching the client unreviewed."}, {"letter": "C", "text": "Claude is asked to double-check its own math and confirm the recommendation is internally consistent", "correct": false, "explanation": "Self-checking arithmetic does not address whether a recommendation is suitable for the client or compliant with financial regulations."}, {"letter": "D", "text": "A marketing team member reviews the wording only for tone and readability before sending", "correct": false, "explanation": "Reviewing only tone and readability ignores the substantive suitability and compliance risks that matter most for financial advice."}], "correct": "A", "select": 1}, {"id": "a2-101", "domain": 2, "task_id": "2.2", "objective": "Identify hallucinations, inconsistencies, and biases in responses", "situation": "A workforce planning lead uses Claude to condense several years of performance notes into short profiles, then ranks the profiles to build a promotion shortlist. Reviewing them, she sees that people from one team are consistently described in warmer language than people from another.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The profiles will read inconsistently to the panel, which may find the varying tone across the shortlist distracting during the meeting.", "correct": false, "explanation": "Inconsistent tone is a real presentational flaw, but it is a cosmetic one. It understates what is happening: the tone difference is tracking group membership, and that is what makes it serious rather than untidy."}, {"letter": "B", "text": "The condensing step will have shortened some profiles more than others, so the panel spends unequal reading time on each individual.", "correct": false, "explanation": "Uneven length affects how much attention each profile receives, which is a genuine fairness concern, but it is a second-order effect next to systematically warmer wording being applied to one group of people."}, {"letter": "C", "text": "A skew in the language becomes a skew in the ranking, so a decision about people's careers is shaped by a pattern nobody examined.", "correct": true, "explanation": "Bias in a summary becomes bias in the outcome once the summaries are ranked, and converting it into a shortlist hides the original wording, so an unexamined pattern silently drives a high-stakes decision about people."}, {"letter": "D", "text": "The notes may be old enough that some achievements described in the profiles have since been superseded by more recent work.", "correct": false, "explanation": "Stale source notes would undermine accuracy and are worth checking separately, but they do not explain a pattern that follows team membership, and they are not the greatest risk in a promotion decision."}], "correct": "C", "select": 1}, {"id": "a2-102", "domain": 2, "task_id": "2.4", "objective": "Determine when human review or additional verification is required", "situation": "A team is adding a new Claude-assisted output into an existing process and has to work out how much checking it will get.", "question": "In which order should these steps be taken?", "options": [{"letter": "A", "text": "Set the level of review that degree of risk requires; judge how visible the output is and how easily a mistake in it could be undone; have the named reviewer carry out that review; release the output to its audience.", "correct": false, "explanation": "The review level is being fixed before anyone has worked out the risk, so the assessment that should have driven the decision arrives too late to inform it."}, {"letter": "B", "text": "Release the output to its audience; judge how visible the output is and how easily a mistake in it could be undone; set the level of review that degree of risk requires; have the named reviewer carry out that review.", "correct": false, "explanation": "Releasing first defeats the whole exercise. Any risk assessment and review that follow publication cannot prevent the harm they were meant to catch."}, {"letter": "C", "text": "Judge how visible the output is and how easily a mistake in it could be undone; set the level of review that degree of risk requires; have the named reviewer carry out that review; release the output to its audience.", "correct": true, "explanation": "Visibility and reversibility determine how much scrutiny an output deserves, so they are assessed first; that assessment sets the required level of review, the named reviewer then performs it, and only then does the output reach its audience."}, {"letter": "D", "text": "Judge how visible the output is and how easily a mistake in it could be undone; have the named reviewer carry out that review; release the output to its audience; set the level of review that degree of risk requires.", "correct": false, "explanation": "The review happens with no agreed standard behind it, and the level of review is only settled after the output has already gone out, which is the wrong way round."}], "correct": "C", "select": 1}, {"id": "a2-103", "domain": 2, "task_id": "2.6", "objective": "Organize and curate information and select appropriate output formats (artifacts, inline, structured data)", "situation": "A team lead wants a reusable onboarding checklist that colleagues will revise over time.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Share it with the team; agree the steps it must cover; ask Claude for the checklist as a document; revise it in place.", "correct": false, "explanation": "Nothing exists to share at the first step, which makes the whole ordering unworkable."}, {"letter": "B", "text": "Ask Claude for the checklist as a document; agree the steps it must cover; revise it in place; share it with the team.", "correct": false, "explanation": "The checklist is drafted before anyone has agreed what belongs in it, so the revision step is doing work the first step should have done."}, {"letter": "C", "text": "Agree the steps it must cover; share it with the team; ask Claude for the checklist as a document; revise it in place.", "correct": false, "explanation": "The checklist is shared before it has been written, and the revision happens after colleagues have already received it."}, {"letter": "D", "text": "Agree the steps it must cover; ask Claude for the checklist as a document; revise it in place; share it with the team.", "correct": true, "explanation": "Content is settled first, a self-contained editable document is the right home for something reused and revised, and it is shared once it is ready."}], "correct": "D", "select": 1}, {"id": "a2-104", "domain": 2, "task_id": "2.4", "objective": "Determine when human review or additional verification is required", "situation": "An HR team has a Claude-drafted consultation letter that will be sent to forty staff whose roles are at risk, and the wording is governed by employment law.", "question": "Which sequence is correct?", "options": [{"letter": "A", "text": "Judge the stakes of the letter, then name the reviewer it requires, then have that reviewer check it against the consultation policy, then send it once approval is recorded.", "correct": true, "explanation": "The stakes are high, regulated and irreversible, so the level of scrutiny is decided first, a named person is made accountable for the review, the check is made against the governing policy, and despatch happens only after recorded approval."}, {"letter": "B", "text": "Record who approved the wording, then send the letter to the forty staff, then check it against the consultation policy, then have the HR lead read it.", "correct": false, "explanation": "Recording an approval before anyone has reviewed the letter documents a sign-off that never happened, and the substantive checks still fall after the letter has already gone out."}, {"letter": "C", "text": "Send the letter to the forty staff, then have the HR lead read it, then check it against the consultation policy, then record who approved the wording.", "correct": false, "explanation": "Sending first makes every later step a post-mortem. A consultation letter cannot be recalled from forty recipients, so review carried out after despatch cannot prevent the harm it is meant to prevent."}, {"letter": "D", "text": "Check the letter against the consultation policy, then send it to the forty staff, then judge the stakes of what was sent, then name a reviewer to look over it carefully afterwards.", "correct": false, "explanation": "Checking against the policy first is sensible, but sending before any named person has reviewed the letter or the stakes have been weighed removes the human review gate exactly where irreversibility demands it."}], "correct": "A", "select": 1}, {"id": "a2-105", "domain": 2, "task_id": "2.5", "objective": "Edit, adapt, refine, and compare outputs for the intended audience", "situation": "A manager has asked Claude for two different versions of a strategy update and must choose one for a board pack.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Ask Claude which of its two versions is the stronger and take that recommendation forward.", "correct": false, "explanation": "Claude has not been told what this board needs, so its preference is about writing quality rather than fit."}, {"letter": "B", "text": "Read both versions through carefully and pick whichever one reads more persuasively to her.", "correct": false, "explanation": "Persuasiveness to the writer is not the test; a polished draft can still fail to give the board what it needs to decide."}, {"letter": "C", "text": "Merge the two versions into one document so that nothing useful from either is left behind.", "correct": false, "explanation": "Merging before criteria exist produces a longer document that serves the board less well than either original."}, {"letter": "D", "text": "Write down what the board needs to take from the update before she reads either version.", "correct": true, "explanation": "Without stated criteria a comparison collapses into which draft reads more nicely. Setting the audience's needs down first makes the choice a judgment rather than a preference."}], "correct": "D", "select": 1}, {"id": "a2-106", "domain": 2, "task_id": "2.4", "objective": "Determine when human review or additional verification is required", "situation": "A programme office is setting a policy for how much scrutiny Claude-generated work should receive. Which two factors raise the level of human review required?", "question": "Select two.", "options": [{"letter": "A", "text": "The prompt that produced the output was long and included pasted background.", "correct": false, "explanation": "Prompt length says nothing about consequence. A well-supplied prompt may even reduce risk, and it certainly does not raise the review bar by itself."}, {"letter": "B", "text": "A large number of people would be affected if the output turned out to be wrong.", "correct": true, "explanation": "Blast radius is independent of reversibility: a mistake that is easy to correct can still reach many people before anyone notices it."}, {"letter": "C", "text": "The action that follows from the output would be hard to undo once taken.", "correct": true, "explanation": "Low reversibility means an error cannot be quietly fixed later, so the check has to happen before the action rather than after it."}, {"letter": "D", "text": "The output arrived in a document format the team has not used very often before.", "correct": false, "explanation": "An unfamiliar file format is a practical inconvenience for the team, not a reason to apply more scrutiny to the content inside it."}, {"letter": "E", "text": "The output was produced faster than the same work would have taken by hand.", "correct": false, "explanation": "Speed of production is not a risk factor. How long the work took has no bearing on what happens if the content is wrong."}], "correct": ["B", "C"], "select": 2}, {"id": "a2-107", "domain": 2, "task_id": "2.2", "objective": "Identify hallucinations, inconsistencies, and biases in responses", "situation": "A marketing lead asks Claude for a competitor briefing, and the draft states that the competitor won a named industry award last year. She can find no record of that award anywhere.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Establish whether the award claim came from material she supplied or from Claude's general knowledge.", "correct": true, "explanation": "Knowing where a claim originated is what separates a fabrication from a real detail she simply has not located, and it determines every step that follows."}, {"letter": "B", "text": "Ask Claude whether it is confident about the award before she spends any more time on it.", "correct": false, "explanation": "A statement of confidence is generated the same way the claim was, so it adds no independent information about whether the award exists."}, {"letter": "C", "text": "Ask Claude to restate the award claim with more supporting detail so that it becomes easier to check.", "correct": false, "explanation": "Asking for more detail around an unsupported claim usually produces more unsupported detail, which makes the claim harder to dismiss rather than easier to check."}, {"letter": "D", "text": "Delete the sentence about the award and circulate the remainder of the draft to her wider team.", "correct": false, "explanation": "Removing one sentence treats the symptom. If the claim was invented, other statements in the same draft may be too, and she has not yet found out."}], "correct": "A", "select": 1}, {"id": "a2-108", "domain": 2, "task_id": "2.6", "objective": "Organize and curate information and select appropriate output formats (artifacts, inline, structured data)", "situation": "A project manager asks Claude to map out a five-stage approval workflow with branching decision points so the team can reference it during a planning meeting.", "question": "What is the best way for Claude to present this?", "options": [{"letter": "A", "text": "A JSON object listing each stage as a key-value pair, since structured data is the standard way to represent any multi-step process", "correct": false, "explanation": "Structured JSON is intended for programmatic data consumption, not for a human-facing reference the team will look at during a meeting."}, {"letter": "B", "text": "An inline bulleted list describing each stage in one sentence, since bullet lists are always clearer than diagrams for workflow steps", "correct": false, "explanation": "A text list cannot easily convey branching decision points the way a flowchart can, so it is not the best fit here even though it could work for a simple linear list."}, {"letter": "C", "text": "An artifact containing a Mermaid flowchart, since diagrams are a supported artifact type well suited to visualizing branching workflows", "correct": true, "explanation": "Diagrams and flowcharts are a supported artifact content type, and a branching workflow with decision points is naturally suited to visual representation the team can reference later."}, {"letter": "D", "text": "A strict tool call that returns the workflow stages as validated tool parameters, since tool calls guarantee the stages are correctly ordered", "correct": false, "explanation": "Strict tool use validates parameters sent to an application tool; it is not a presentation format for sharing a workflow with a team."}], "correct": "C", "select": 1}, {"id": "a2-109", "domain": 2, "task_id": "2.2", "objective": "Identify hallucinations, inconsistencies, and biases in responses", "situation": "A grants officer reads a Claude-written explainer of a funding scheme and notices that it gives one closing date in the opening paragraph and a different closing date three paragraphs later.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Ask Claude which of the two dates it meant, and use whichever one it settles on when it is asked to resolve the conflict.", "correct": false, "explanation": "Asking the same conversation to arbitrate treats Claude as a singular source of truth on exactly the point where it has already demonstrated it is unreliable, and it will produce a confident answer either way."}, {"letter": "B", "text": "Delete the earlier paragraph so the explainer no longer contradicts itself, and circulate the corrected version today.", "correct": false, "explanation": "Removing one of the two dates makes the internal contradiction disappear without resolving it, which is worse than leaving it visible: the surviving date may well be the wrong one, and the warning sign has been erased."}, {"letter": "C", "text": "Go back to the funder's own published scheme rules and establish which closing date, if either of them, is the real one.", "correct": true, "explanation": "An internal contradiction shows at least one statement is unreliable, and it cannot be settled inside the response. Checking the funder's own published rules resolves it against an authoritative source rather than relying on Claude alone."}, {"letter": "D", "text": "Keep the later of the two dates on the basis that it was written after Claude had processed more of the scheme material.", "correct": false, "explanation": "Position in a response carries no evidential weight; a later sentence is not better sourced than an earlier one. Picking a date by where it appeared leaves the officer publishing a deadline she has never verified."}], "correct": "C", "select": 1}, {"id": "a2-110", "domain": 2, "task_id": "2.6", "objective": "Organize and curate information and select appropriate output formats (artifacts, inline, structured data)", "situation": "During a troubleshooting chat, a user asks Claude, \"Is port 443 the default HTTPS port?\" Claude's answer is a single confirming sentence.", "question": "What is the appropriate output treatment for this response?", "options": [{"letter": "A", "text": "Ask a strict tool to validate the port number against a schema before responding, since numeric facts require schema validation", "correct": false, "explanation": "Claude can directly answer common factual questions like default port numbers from its training data without needing external tool validation. Using a tool for such a basic fact would introduce unnecessary overhead and complexity."}, {"letter": "B", "text": "Create a Markdown document artifact titled \"HTTPS Port Reference,\" since any factual networking answer should be saved for later use", "correct": false, "explanation": "Not every factual answer demands an artifact. Artifacts are intended for content that is substantial and reusable, such as comprehensive guides or documents. A single port confirmation is too trivial to warrant an artifact."}, {"letter": "C", "text": "Package the answer as a JSON object with a \"confirmed\" boolean field, since factual yes/no answers should always use structured data", "correct": false, "explanation": "While structured data can be useful for programmatic interactions, a casual chat does not require packaging every fact into a JSON object. Anthropic’s documentation does not mandate that all yes/no answers be structured; inline natural language is perfectly acceptable."}, {"letter": "D", "text": "Answer inline within the conversation, since a single confirming sentence has no reuse value that would justify an artifact", "correct": true, "explanation": "Inline answers are appropriate for short, simple responses that do not constitute a substantial, reusable piece of content. Artifacts are reserved for content that is more complex and self-contained, such as code, documents, or long-form text. This aligns with Anthropic’s guidance on using artifacts for significant content."}], "correct": "D", "select": 1}, {"id": "a2-111", "domain": 2, "task_id": "2.4", "objective": "Determine when human review or additional verification is required", "situation": "An analyst has forty Claude-drafted product descriptions. About half will be posted on the public website and half will sit in an internal catalogue, and she has time to check roughly twenty of them properly before the deadline.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Check the longest descriptions first, because a longer description contains more claims to get wrong.", "correct": false, "explanation": "Length is not a proxy for stakes. A short line of public-facing copy can be far more damaging than a long internal entry, so sorting by length would leave the riskiest items unchecked."}, {"letter": "B", "text": "Check the descriptions in the order they were generated, until the time she has set aside runs out.", "correct": false, "explanation": "Working in generation order spends her limited review time at random. The website copy and the internal catalogue do not carry the same consequences, so order of arrival is the wrong basis for choosing."}, {"letter": "C", "text": "Sort the batch by where each description will end up and how hard an error there would be to undo.", "correct": true, "explanation": "With review capacity smaller than the batch, the first step is to triage by exposure and reversibility. Externally published copy that is hard to retract earns the scarce review time; an internal entry anyone can correct does not."}, {"letter": "D", "text": "Ask Claude to mark the descriptions it is least sure of, and check only the ones that it flags.", "correct": false, "explanation": "A model's own stated uncertainty is a weak and unvalidated signal, and it says nothing about consequences. It can be a useful supplement, but it cannot decide which outputs need human eyes."}], "correct": "C", "select": 1}, {"id": "a2-112", "domain": 2, "task_id": "2.5", "objective": "Edit, adapt, refine, and compare outputs for the intended audience", "situation": "A learning and development specialist has a Claude-drafted explanation of a new expenses process written for finance staff. She now needs a version for field engineers who never open the finance system.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Cut the draft down to roughly half its present length so that the field engineers will read it.", "correct": false, "explanation": "Halving the length is an arbitrary cut with no principle behind it, and it is as likely to remove the steps the engineers need as the ones they do not."}, {"letter": "B", "text": "Ask Claude to simplify the wording of the existing text until it reads a good deal more easily.", "correct": false, "explanation": "Simplifying the wording keeps the finance staff's version of the process and only changes its surface. The engineers would get an easier read of material that is still organised around someone else's job."}, {"letter": "C", "text": "Ask Claude to add a glossary explaining the finance terms that the draft already uses.", "correct": false, "explanation": "A glossary makes unfamiliar terminology survivable but leaves the engineers reading a finance-system walkthrough that mostly does not apply to them."}, {"letter": "D", "text": "Work out what the field engineers actually have to do differently, and write to that.", "correct": true, "explanation": "Adapting for an audience starts from what that audience needs to do, not from the existing text. Once she knows which parts of the process the engineers actually touch, the register, the length and the level of detail all follow from it."}], "correct": "D", "select": 1}, {"id": "a2-113", "domain": 2, "task_id": "2.4", "objective": "Determine when human review or additional verification is required", "situation": "A grants officer facing a deadline has used Claude to draft eligibility determinations for ninety applicants, and each determination will be sent directly to the applicant. Under Anthropic's Usage Policy, which two things must happen before the determinations are issued?", "question": "Select two.", "options": [{"letter": "A", "text": "Claude is asked to review its own determinations and flag any it is unsure about.", "correct": false, "explanation": "Self-review is not verification. A determination that misread a rule will often be reported as confident, so the cases most in need of attention stay hidden."}, {"letter": "B", "text": "A qualified officer checks each determination against the published eligibility rules.", "correct": true, "explanation": "The Usage Policy requires that a qualified professional review a decision directly affecting individuals before it is finalized. Eligibility for funding is such a decision, so each one needs checking against the actual rules rather than sampling or spot checks."}, {"letter": "C", "text": "Each determination that goes out tells the applicant that Claude helped produce it.", "correct": true, "explanation": "When model outputs are presented directly to individuals, the Usage Policy requires disclosing to them that AI helped produce the decision, so each determination sent to an applicant must say that Claude helped produce it. An internal file note alone would not tell the applicant."}, {"letter": "D", "text": "Only the determinations that look borderline are pulled out for a person to read.", "correct": false, "explanation": "Borderline cases are identified by the same process being verified, so anything wrongly classed as clear-cut is exactly what this approach fails to catch."}, {"letter": "E", "text": "Claude is recorded as the decision-maker, so responsibility for each one rests with the tool.", "correct": false, "explanation": "Responsibility cannot be handed to the tool. The Usage Policy states that the user or their organization is responsible for the accuracy and appropriateness of the information, so recording Claude as the decision-maker leaves no accountable person and does nothing to make the determinations correct."}], "correct": ["B", "C"], "select": 2}, {"id": "a2-114", "domain": 2, "task_id": "2.2", "objective": "Identify hallucinations, inconsistencies, and biases in responses", "situation": "A sales operations lead asked Claude to summarise two quarters of win-loss notes. Early in the summary it states that deals are lost mainly on price, and two paragraphs later it attributes most losses to slow implementation timelines.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Go back to the win-loss notes themselves and see what they actually say about the losses.", "correct": true, "explanation": "An internal contradiction means at least one of the two statements is unsupported, and the notes she uploaded are the ground truth. Reading what the losses were actually recorded as resolves the inconsistency and shows which half of the summary can be trusted."}, {"letter": "B", "text": "Keep the price explanation, since it appears first and is the more familiar of the two.", "correct": false, "explanation": "Position in the text is not evidence, and preferring the more familiar explanation is exactly the bias the review is supposed to catch. It could entrench a wrong conclusion about why the company is losing deals."}, {"letter": "C", "text": "Ask Claude to produce the summary again and go with whichever explanation comes back next.", "correct": false, "explanation": "Regenerating gives a fresh answer, not a checked one, and picking whichever version arrives next chooses between two unverified claims at random."}, {"letter": "D", "text": "Ask Claude which of the two explanations it considers the more likely to be correct.", "correct": false, "explanation": "Asking the same conversation to arbitrate between its own two claims produces another confident statement with nothing new behind it. The contradiction is evidence that the summary has drifted from the source, and only the source settles it."}], "correct": "A", "select": 1}, {"id": "a2-115", "domain": 2, "task_id": "2.4", "objective": "Determine when human review or additional verification is required", "situation": "A customer support chatbot built on Claude detects that a customer's message expresses distress and references potential legal action.", "question": "What should the system do with this conversation?", "options": [{"letter": "A", "text": "Let the automated conversation continue but have it apologize more frequently in each generated response", "correct": false, "explanation": "Adding more apologies does not address the substantive risk of mishandling a distressed customer or a legal threat."}, {"letter": "B", "text": "Close the support ticket automatically since the chatbot already provided a complete response", "correct": false, "explanation": "Closing the ticket automatically ignores the escalation risk and could leave a serious concern unresolved."}, {"letter": "C", "text": "Escalate the conversation to a trained human agent instead of letting the automated response continue", "correct": true, "explanation": "Messages signaling distress or potential legal action are high-risk situations where a trained human agent should take over rather than continuing fully automated handling."}, {"letter": "D", "text": "Ask the model to generate a longer response covering more possible resolutions for the customer", "correct": false, "explanation": "A longer automated response does not provide the human judgment needed for a sensitive, potentially legal situation."}], "correct": "C", "select": 1}, {"id": "a2-116", "domain": 2, "task_id": "2.5", "objective": "Edit, adapt, refine, and compare outputs for the intended audience", "situation": "An HR team drafts a benefits explanation for employees across countries whose languages and cultural norms around directness differ from the home office.", "question": "Beyond translation, what should the team check when adapting the content for each region?", "options": [{"letter": "A", "text": "Whether every regional version ends up with the exact same word count once translated", "correct": false, "explanation": "Matching word count across languages is not a meaningful measure of whether content is culturally and tonally appropriate."}, {"letter": "B", "text": "Whether the same idioms from the home office draft are preserved in each translated version", "correct": false, "explanation": "Preserving home-office idioms in translation often produces confusing phrasing rather than adapting content to local readers."}, {"letter": "C", "text": "Whether the tone, examples, and directness fit local workplace norms, not just the translated words", "correct": true, "explanation": "Adapting for an audience across regions means adjusting tone, examples, and directness to match local norms, which goes beyond literal translation of words."}, {"letter": "D", "text": "Whether the document uses the identical formatting template regardless of region or reader", "correct": false, "explanation": "Formatting template consistency does not address whether tone and content are appropriate for each region's norms."}], "correct": "C", "select": 1}, {"id": "a2-117", "domain": 2, "task_id": "2.1", "objective": "Evaluate Claude-generated outputs for accuracy and completeness", "situation": "A team is evaluating a customer support assistant and only measures whether responses are factually correct, ignoring tone, relevance, and consistency.", "question": "Following recommended evaluation practice, what is the main risk of this approach?", "options": [{"letter": "A", "text": "Adding more evaluation dimensions reduces overall assessment reliability, so the team should keep the evaluation limited to factual correctness.", "correct": false, "explanation": "Adding more evaluation dimensions does not reduce reliability; rather, it provides a more comprehensive and accurate assessment. Multidimensional evaluation is encouraged for complex tasks to capture quality aspects beyond factual accuracy."}, {"letter": "B", "text": "Single-dimension evaluation misses important quality issues, so complex tasks should be scored across multiple criteria.", "correct": true, "explanation": "Single-dimension evaluation misses important quality issues like tone, relevance, and consistency. Recommended practice calls for multidimensional scoring to ensure comprehensive assessment of complex tasks."}, {"letter": "C", "text": "Multidimensional evaluation is only needed for safety-critical systems, so factual correctness alone is appropriate for customer support.", "correct": false, "explanation": "Multidimensional evaluation is recommended broadly for complex tasks, not only safety-critical systems. A customer support assistant benefits from evaluating tone and relevance, as these influence user satisfaction."}, {"letter": "D", "text": "Factual correctness is the single most important factor for customer support, making this evaluation approach fully adequate.", "correct": false, "explanation": "Factual correctness alone is insufficient for complex tasks like customer support; ignoring tone, relevance, and consistency can lead to poor user experiences. Recommended practice emphasizes multidimensional evaluation to fully assess quality."}], "correct": "B", "select": 1}, {"id": "a2-118", "domain": 2, "task_id": "2.1", "objective": "Evaluate Claude-generated outputs for accuracy and completeness", "situation": "A small charity has no legal adviser. Its manager asks Claude what the new reporting rules require, receives a clear and specific answer, and files the return on that basis alone.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The explanation may be pitched at a level of detail that the charity's own trustees find difficult to follow later.", "correct": false, "explanation": "Detail pitched wrongly for trustees is a communication issue that a rewrite fixes, and it does not affect whether the filing is correct."}, {"letter": "B", "text": "A single unverified answer is being treated as authoritative on a filing with real consequences for the charity.", "correct": true, "explanation": "Regulatory filings carry consequences that are hard to reverse, and a confident answer is not a verified one. High-stakes guidance should be confirmed against the regulator's own material or a professional adviser."}, {"letter": "C", "text": "The explanation may cover the reporting rules while leaving out related rules the charity has not asked about.", "correct": false, "explanation": "Adjacent obligations being missed is a genuine gap, but it is one instance of the broader problem of relying on a single unchecked answer."}, {"letter": "D", "text": "The explanation may use technical vocabulary the manager has to look up before he can act on any of it.", "correct": false, "explanation": "Unfamiliar vocabulary slows the manager down and is easily resolved by asking for a plainer explanation."}], "correct": "B", "select": 1}, {"id": "a2-119", "domain": 2, "task_id": "2.5", "objective": "Edit, adapt, refine, and compare outputs for the intended audience", "situation": "A trainer keeps asking Claude to make a technical explanation simpler, and after four rounds it reads very easily.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The explanation will become too short for the length the training materials require overall.", "correct": false, "explanation": "Length is easy to see and easy to fix, and a shorter explanation is often the point of the exercise."}, {"letter": "B", "text": "Each round can quietly drop a qualifier, leaving a version that reads well but is no longer true.", "correct": true, "explanation": "Simplification removes conditions and caveats, and the wording that goes first is often what made the statement accurate rather than merely clear."}, {"letter": "C", "text": "Repeated rounds of rewriting will make the tone of the explanation gradually more informal.", "correct": false, "explanation": "Drifting tone is noticeable on a read-through and does not by itself make the content wrong."}, {"letter": "D", "text": "The final version will use vocabulary that differs from the earlier training material on the topic.", "correct": false, "explanation": "Inconsistent vocabulary is a presentation issue and is easily reconciled across materials."}], "correct": "B", "select": 1}, {"id": "a2-120", "domain": 2, "task_id": "2.3", "objective": "Apply fact-checking and validation techniques", "situation": "A chatbot answers a user's question about a niche historical event with a detailed, confident, but entirely fabricated account, since no similar event exists in its training data. The team wants to reduce this kind of failure through prompt design alone, without adding any tools.", "question": "Which instruction is most effective for reducing this type of fabrication?", "options": [{"letter": "A", "text": "Instruct Claude to always answer within a strict two-sentence limit regardless of how complex the topic is", "correct": false, "explanation": "Limiting response length does not address the underlying cause of fabrication and may even worsen it by forcing an answer without room for caveats."}, {"letter": "B", "text": "Instruct Claude to answer every question using only information gathered during its most recent training run", "correct": false, "explanation": "Claude cannot selectively restrict itself to only the most recent training run, and this instruction does nothing to prevent fabrication on topics with sparse training data."}, {"letter": "C", "text": "Tell Claude it is acceptable, and preferred, to say it cannot verify an answer rather than guessing when unsure", "correct": true, "explanation": "Explicitly permitting Claude to express uncertainty or decline to answer when it lacks grounded information is a well-established prompting technique for reducing confident fabrication, since models often default to producing a plausible-sounding answer when not told otherwise."}, {"letter": "D", "text": "Instruct Claude to always adopt a highly confident, authoritative tone in every reply so users trust the response", "correct": false, "explanation": "Encouraging a confident tone works against fact-checking goals, since it makes fabricated answers sound more credible rather than flagging uncertainty."}], "correct": "C", "select": 1}, {"id": "a2-121", "domain": 2, "task_id": "2.6", "objective": "Organize and curate information and select appropriate output formats (artifacts, inline, structured data)", "situation": "A user asks Claude to build an interactive to-do list where items can be checked off and reordered directly in the browser.", "question": "Which artifact type fits this request?", "options": [{"letter": "A", "text": "A CSV file listing each to-do item as a row, since CSV is the appropriate format whenever the user needs to track list items", "correct": false, "explanation": "CSV is a flat data format meant for tabular data exchange, not for delivering an interactive, clickable interface."}, {"letter": "B", "text": "An inline chat message with each item as a separate bullet, since inline lists let the user check items off directly in the conversation", "correct": false, "explanation": "Inline chat text is static; a user cannot click to check off or reorder items directly within the conversation text."}, {"letter": "C", "text": "A plain Markdown document artifact, since Markdown documents support checkboxes and are the standard format for to-do lists", "correct": false, "explanation": "A static Markdown document can show checkbox syntax but cannot support live reordering or interactive click behavior the way a component can."}, {"letter": "D", "text": "An interactive React component artifact, since it supports the stateful, clickable interactions the to-do list requires", "correct": true, "explanation": "React components are a supported artifact type built for interactive, stateful UI such as checking off and reordering items in the browser."}], "correct": "D", "select": 1}, {"id": "a2-122", "domain": 2, "task_id": "2.2", "objective": "Identify hallucinations, inconsistencies, and biases in responses", "situation": "A consultant is briefing colleagues on the guidance that Claude should not be relied on as a singular source of truth.", "question": "What does that principle mean in day-to-day practice for a claim she intends to act on?", "options": [{"letter": "A", "text": "That she should avoid Claude for factual work altogether and confine her use of it to drafting, wording and brainstorming.", "correct": false, "explanation": "Abandoning factual work overstates the guidance, which is about verifying rather than avoiding. It also gives up genuinely useful help with research and synthesis that a person can then check."}, {"letter": "B", "text": "That she should raise the same question in several fresh conversations and treat consistent answers as adequate confirmation.", "correct": false, "explanation": "Repeated conversations are still the same source, so agreement between them is not independent corroboration. A confidently wrong claim can be reproduced consistently across as many chats as she cares to open."}, {"letter": "C", "text": "That she should keep her questions narrow, because short and tightly scoped requests are inherently less prone to inaccuracy.", "correct": false, "explanation": "Narrower questions can improve focus, but scope is not a guarantee of accuracy and a short answer can be just as wrong. This substitutes a prompting habit for the verification the principle actually calls for."}, {"letter": "D", "text": "That a claim she is going to act on needs corroboration from an independent source before she treats it as established.", "correct": true, "explanation": "The principle is about where confirmation comes from: anything consequential should be corroborated outside the conversation, so Claude informs the decision rather than being the only thing standing behind it."}], "correct": "D", "select": 1}, {"id": "a2-123", "domain": 2, "task_id": "2.4", "objective": "Determine when human review or additional verification is required", "situation": "While researching a niche technical topic, an analyst receives a confident, well-written answer from Claude that includes a specific statistic with no source cited and no search tool used.", "question": "What should the analyst do before citing this statistic in a report?", "options": [{"letter": "A", "text": "Reword the statistic slightly so it reads as an approximation rather than an exact figure", "correct": false, "explanation": "Softening the wording does not verify whether the underlying number is correct; it only obscures precision."}, {"letter": "B", "text": "Ask Claude for the statistic a second time and use it if the second answer matches the first", "correct": false, "explanation": "A model can consistently reproduce the same fabricated figure on repeated asks, so matching answers do not confirm accuracy."}, {"letter": "C", "text": "Include the statistic as stated in the report because the surrounding explanation is detailed and well-written", "correct": false, "explanation": "Detailed, confident writing is not evidence that a specific unsourced statistic is factually accurate."}, {"letter": "D", "text": "Locate an independent, authoritative source that confirms the statistic before including it in the report", "correct": true, "explanation": "An uncited, ungrounded statistic may be a hallucination, so the analyst should confirm it against an independent authoritative source before citing it."}], "correct": "D", "select": 1}, {"id": "a2-124", "domain": 2, "task_id": "2.3", "objective": "Apply fact-checking and validation techniques", "situation": "A communications lead inherits a Claude-written fact sheet from a colleague who has left, with no record of what it was based on.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Publish the fact sheet with a note explaining that its original sources could not be traced by the team.", "correct": false, "explanation": "A disclaimer does not make unverified statements safe to publish; it just moves the risk onto the reader."}, {"letter": "B", "text": "Compare the fact sheet against a second Claude draft written from the same brief and topic.", "correct": false, "explanation": "Agreement between two generated drafts is not evidence, since both can be wrong in the same way."}, {"letter": "C", "text": "Paste the fact sheet into Claude and ask it to identify which of the statements were made up.", "correct": false, "explanation": "Claude cannot inspect how an earlier response was produced, and it has no access to the sources that were or were not used."}, {"letter": "D", "text": "Re-establish which sources the fact sheet rests on, since checking needs something to check against.", "correct": true, "explanation": "Verification is a comparison, and without an agreed set of authoritative sources there is nothing to compare the statements with."}], "correct": "D", "select": 1}, {"id": "a2-125", "domain": 2, "task_id": "2.2", "objective": "Identify hallucinations, inconsistencies, and biases in responses", "situation": "A developer wants to grade thousands of Claude outputs for quality and plans to have the same model that generated the outputs also score them. A colleague flags this as a risky design.", "question": "What is the main concern with using the same model to both generate and grade its own outputs?", "options": [{"letter": "A", "text": "The model cannot process its own prior outputs as input, so grading requests will fail", "correct": false, "explanation": "A model can process its own prior text as input to a new request; there is no technical restriction preventing this."}, {"letter": "B", "text": "The model will exceed its context window whenever it grades more than a few responses", "correct": false, "explanation": "Context window limits depend on total input size, not on whether the grading model is the same one that generated the text."}, {"letter": "C", "text": "The model's pricing doubles whenever it is reused for both generation and grading tasks", "correct": false, "explanation": "Pricing is based on token usage per request, not on whether the same model is reused across generation and grading steps."}, {"letter": "D", "text": "The model may favor its own outputs, so a separate model or human should grade instead", "correct": true, "explanation": "A recommended best practice is to use a separate model or human reviewers for grading, since a model grading its own generations can be biased toward judging them favorably."}], "correct": "D", "select": 1}, {"id": "a2-126", "domain": 2, "task_id": "2.3", "objective": "Apply fact-checking and validation techniques", "situation": "A newsletter editor needs a Claude-drafted piece to quote the organisation's own published figures accurately.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Ask Claude to draft the piece; gather the published figures; check the draft's numbers against them; send the piece out to readers.", "correct": false, "explanation": "Drafting before the figures are supplied invites Claude to fill the gaps, which creates the errors the later check then has to catch."}, {"letter": "B", "text": "Gather the published figures; check the draft's numbers against them; give them to Claude as the material; send the piece out to readers.", "correct": false, "explanation": "The numbers are checked before a draft exists, and the source material reaches Claude only after the piece has been written."}, {"letter": "C", "text": "Gather the published figures; give them to Claude as the material; check the draft's numbers against them; send the piece out to readers.", "correct": true, "explanation": "Supplying the authoritative figures first keeps the draft grounded, and the check afterwards confirms nothing drifted during writing."}, {"letter": "D", "text": "Give them to Claude as the material; send the piece out to readers; gather the published figures; check the draft's numbers against them.", "correct": false, "explanation": "The piece goes to readers before the figures have been gathered or any check has taken place."}], "correct": "C", "select": 1}, {"id": "a2-127", "domain": 2, "task_id": "2.4", "objective": "Determine when human review or additional verification is required", "situation": "An HR team uses Claude to generate a shortlist recommendation from a stack of job applications.", "question": "Before any candidate is rejected based on this shortlist, what should the team do?", "options": [{"letter": "A", "text": "Have a recruiter review the underlying applications to check for bias or overlooked qualified candidates", "correct": true, "explanation": "Employment decisions carry legal and fairness risk, so a recruiter should review the underlying applications to catch bias or qualified candidates the model may have overlooked."}, {"letter": "B", "text": "Shorten the job description so future applicant pools are easier for the model to rank consistently", "correct": false, "explanation": "Changing the job description does not address whether the current shortlist decision requires human review before candidates are rejected."}, {"letter": "C", "text": "Reject every applicant left off the shortlist since the model already considered all submitted application details", "correct": false, "explanation": "Rejecting candidates purely based on an unreviewed AI shortlist risks embedding bias or errors into a consequential employment decision."}, {"letter": "D", "text": "Ask Claude to regenerate the shortlist a second time and merge both lists into the final decision", "correct": false, "explanation": "Merging two outputs from the same model does not introduce the independent human judgment needed to catch bias in a hiring decision."}], "correct": "A", "select": 1}, {"id": "a2-128", "domain": 2, "task_id": "2.2", "objective": "Identify hallucinations, inconsistencies, and biases in responses", "situation": "A support engineer asks Claude to summarize a technical paper and include supporting references. The response includes a citation to a journal article whose title and authors do not appear in the source document or in any database the engineer can find.", "question": "What does this response demonstrate?", "options": [{"letter": "A", "text": "A hallucination, since Claude fabricated a citation with no basis in any verifiable source", "correct": true, "explanation": "A hallucination is content presented as fact that is not grounded in the source material or any verifiable reference; an invented citation is a classic example."}, {"letter": "B", "text": "A context limitation, since the source document was too long for Claude to fully process", "correct": false, "explanation": "Nothing in the scenario indicates the document exceeded the context window; the issue is a fabricated reference, not truncation."}, {"letter": "C", "text": "A bias, since Claude systematically favored one publication over equally valid alternatives", "correct": false, "explanation": "Bias involves skewed treatment of topics, groups, or viewpoints, not the invention of a nonexistent source."}, {"letter": "D", "text": "An inconsistency, since the citation format breaks from the rest of the response's style", "correct": false, "explanation": "Inconsistency refers to conflicting statements within or across responses, not a fabricated fact that never existed."}], "correct": "A", "select": 1}, {"id": "a2-129", "domain": 2, "task_id": "2.6", "objective": "Organize and curate information and select appropriate output formats (artifacts, inline, structured data)", "situation": "A coordinator needs a list of thirty training bookings from Claude in a form she can sort and filter in her spreadsheet.", "question": "Which output should she ask for?", "options": [{"letter": "A", "text": "A set of paragraphs, one for each booking, describing its details in full and readable prose.", "correct": false, "explanation": "Prose per booking preserves the records but leaves her converting thirty paragraphs into columns by hand."}, {"letter": "B", "text": "A diagram showing how the thirty bookings relate to one another across the training calendar.", "correct": false, "explanation": "A diagram shows relationships but produces nothing she can load into a spreadsheet and manipulate."}, {"letter": "C", "text": "A short narrative summary describing the overall pattern across the thirty bookings she has.", "correct": false, "explanation": "A narrative cannot be sorted or filtered, and it compresses away the individual records she needs."}, {"letter": "D", "text": "Rows and columns with one booking per row and a consistent field in each of the columns.", "correct": true, "explanation": "Structured, tabular output maps directly onto a spreadsheet, so each booking stays a separate record with fields she can sort and filter."}], "correct": "D", "select": 1}, {"id": "a2-130", "domain": 2, "task_id": "2.2", "objective": "Identify hallucinations, inconsistencies, and biases in responses", "situation": "A communications assistant receives an answer containing a clearly incorrect statement about her organisation. She corrects her own draft in a follow-up message, and also wants the problem itself recorded rather than just worked around.", "question": "What should she do?", "options": [{"letter": "A", "text": "Add a line to the conversation noting that the earlier statement was wrong, so the correction sits in the chat history.", "correct": false, "explanation": "A note in the chat helps her own record of what happened, but it stays inside her conversation. It does nothing to report the response as problematic beyond her own session."}, {"letter": "B", "text": "Start the same task again in a new conversation, on the basis that a clean chat is less likely to repeat the mistake.", "correct": false, "explanation": "A fresh conversation may or may not produce a better answer, and starting over is a workaround for her immediate task rather than a way of recording that the response was wrong."}, {"letter": "C", "text": "Rephrase her original prompt in more forceful language so that the incorrect statement is less likely to recur next time.", "correct": false, "explanation": "Emphatic rewording is a weak remedy and addresses only her future prompts. It neither reports the faulty response nor makes the problem visible to anyone else."}, {"letter": "D", "text": "Use the thumbs-down control on the response that contained the incorrect statement, in addition to fixing her own draft.", "correct": true, "explanation": "The thumbs-down feedback control on the response is the mechanism provided for flagging an incorrect or misleading answer, and it complements, rather than replaces, correcting the draft she is working on."}], "correct": "D", "select": 1}, {"id": "a2-131", "domain": 2, "task_id": "2.6", "objective": "Organize and curate information and select appropriate output formats (artifacts, inline, structured data)", "situation": "A manager asks Claude to pull figures from several uploaded reports into one table and plans to paste it straight into a board deck.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "A tidy table hides which report each number came from and which ones were inferred rather than read.", "correct": true, "explanation": "Putting figures in a grid strips away their provenance and any caveats, so numbers of very different reliability end up looking equally authoritative to the board."}, {"letter": "B", "text": "The table will be formatted in a way the presentation software cannot display neatly on a slide.", "correct": false, "explanation": "Layout problems are visible immediately and take minutes to correct, unlike a wrong figure nobody notices."}, {"letter": "C", "text": "The table will contain more rows than a board audience can reasonably absorb during a meeting.", "correct": false, "explanation": "Too much detail is an editing problem, and it is easily fixed once someone reads the table."}, {"letter": "D", "text": "Building the table will use up rather more of her allowance than reading the reports herself would have.", "correct": false, "explanation": "The cost of producing the table is minor compared with a board decision resting on untraceable figures."}], "correct": "A", "select": 1}, {"id": "a2-132", "domain": 2, "task_id": "2.6", "objective": "Organize and curate information and select appropriate output formats (artifacts, inline, structured data)", "situation": "A user has iterated on a code artifact through four rounds of edits and now decides the second version was actually the best one.", "question": "How can the user return to that earlier version?", "options": [{"letter": "A", "text": "Start an entirely new conversation, since artifact versions cannot be recovered once a newer version has been created", "correct": false, "explanation": "Version history persists within the same conversation, so starting a new conversation is not required to view an earlier version."}, {"letter": "B", "text": "Use the artifact's version selector to switch back to the second version, since each iteration is preserved and can be revisited", "correct": true, "explanation": "The artifact interface includes a version selector that lets users switch between prior iterations of the same artifact without losing any of them."}, {"letter": "C", "text": "Ask Claude to recall the exact wording of the second version from memory, since earlier artifact versions are not stored anywhere", "correct": false, "explanation": "Prior versions are stored and browsable through the version selector, not something Claude needs to reconstruct from memory."}, {"letter": "D", "text": "Manually retype the second version from the chat history, since artifacts only ever display the most recent iteration", "correct": false, "explanation": "Artifacts retain their edit history, so manually retyping an earlier version is unnecessary."}], "correct": "B", "select": 1}, {"id": "a2-133", "domain": 2, "task_id": "2.1", "objective": "Evaluate Claude-generated outputs for accuracy and completeness", "situation": "A consultant has an hour to review a long research brief before sending it to a client.", "question": "Which sequence should she follow?", "options": [{"letter": "A", "text": "Read the conclusion; identify the claims the conclusion rests on; check those claims; verify the supporting detail.", "correct": true, "explanation": "Reading the conclusion first reveals what the brief is asking the client to accept, which identifies the load-bearing claims. Those are checked while time remains, and remaining detail is verified with whatever is left."}, {"letter": "B", "text": "Check those claims; verify the supporting detail; identify the claims the conclusion rests on; read the conclusion.", "correct": false, "explanation": "Checking claims that have not yet been identified is not possible, and reading the conclusion last leaves the review with no organising purpose."}, {"letter": "C", "text": "Identify the claims the conclusion rests on; verify the supporting detail; read the conclusion; check those claims.", "correct": false, "explanation": "The claims a conclusion rests on cannot be identified before the conclusion has been read, so this ordering starts from an unknown."}, {"letter": "D", "text": "Verify the supporting detail; read the conclusion; identify the claims the conclusion rests on; check those claims.", "correct": false, "explanation": "Starting on supporting detail spends the scarce hour on material that may not matter, and the conclusion is read only after that time is gone."}], "correct": "A", "select": 1}, {"id": "a2-134", "domain": 2, "task_id": "2.2", "objective": "Identify hallucinations, inconsistencies, and biases in responses", "situation": "A company's customer service chatbot tells a customer, 'Our support team works only from our London office.' Later in the same conversation, it tells the same customer, 'Our support team works only from our New York office.' Someone reviewing this transcript cannot check where the company's offices really are.", "question": "Which issue does the transcript itself clearly show?", "options": [{"letter": "A", "text": "The chatbot gave two answers that contradict each other.", "correct": true, "explanation": "The chatbot first says its support team works only from the London office and later says it works only from the New York office. Both statements cannot be true at once, so the transcript alone shows the chatbot contradicting itself within one conversation, and no outside fact-checking is needed to see it."}, {"letter": "B", "text": "The chatbot favors customers in one city over another.", "correct": false, "explanation": "Both replies describe where the support team works and say nothing that treats one group of customers better than another, so the transcript shows no preferential or prejudiced treatment."}, {"letter": "C", "text": "The chatbot's statements are confirmed to be factually wrong.", "correct": false, "explanation": "Without the company's real office details the reviewer cannot confirm that either statement is wrong, and nothing in the transcript settles it. What the transcript does establish is that at most one of the two statements can be true, which is a contradiction rather than a confirmed factual error."}, {"letter": "D", "text": "The chatbot made up a source for its claim.", "correct": false, "explanation": "Neither reply cites or quotes any source, so there is no invented citation anywhere in the transcript. What is visible is that the two replies disagree with each other, not that supporting evidence was fabricated."}], "correct": "A", "select": 1}, {"id": "a2-135", "domain": 2, "task_id": "2.1", "objective": "Evaluate Claude-generated outputs for accuracy and completeness", "situation": "While checking a twelve-page report, an analyst finds that one section has used last year's headcount figures.", "question": "Which sequence should she follow next?", "options": [{"letter": "A", "text": "Correct that section; re-read the conclusions in light of the correction; look for the same error elsewhere; approve the report.", "correct": false, "explanation": "Fixing only the section where the error was noticed leaves identical errors elsewhere, and the conclusions are re-read before those are found."}, {"letter": "B", "text": "Look for the same error elsewhere; correct every affected section; re-read the conclusions in light of it; approve the report.", "correct": true, "explanation": "One instance of stale data suggests a source problem rather than a typing slip, so the scope of the error is established first, every affected section is then corrected, and the conclusions are re-read against the corrected figures before approval."}, {"letter": "C", "text": "Re-read the conclusions in light of it; correct that section; approve the report; look for the same error elsewhere.", "correct": false, "explanation": "Re-reading conclusions before the figures are corrected judges them against numbers already known to be wrong, and approval precedes the wider search."}, {"letter": "D", "text": "Approve the report; look for the same error elsewhere; correct every affected section; re-read the conclusions in light of it.", "correct": false, "explanation": "Approval comes before any of the corrective work, so the report is signed off in the state that was already known to be wrong."}], "correct": "B", "select": 1}, {"id": "a2-136", "domain": 2, "task_id": "2.1", "objective": "Evaluate Claude-generated outputs for accuracy and completeness", "situation": "A team keeps arguing about whether Claude's drafts of their monthly service report are good enough, and each reviewer reaches a different verdict.", "question": "What would most improve how they evaluate the drafts?", "options": [{"letter": "A", "text": "Asking for two drafts each month and adopting whichever of the two the reviewers prefer once they compare them.", "correct": false, "explanation": "Comparing two drafts identifies the better of the two, which may still fall short of what the report actually needs to contain."}, {"letter": "B", "text": "Writing down what a complete report must contain before the draft is produced, and reviewing against that list.", "correct": true, "explanation": "Disagreement of this kind comes from reviewers applying different unstated standards. A written statement of what the report must contain turns the review into a check against something specific instead of a matter of taste."}, {"letter": "C", "text": "Agreeing that a draft passes when a clear majority of the reviewers involved consider it good enough to publish.", "correct": false, "explanation": "Voting aggregates opinions formed on different criteria, so it produces a decision without producing a shared standard."}, {"letter": "D", "text": "Appointing the most experienced reviewer to make the final call whenever the reviewers disagree about a draft.", "correct": false, "explanation": "Deferring to seniority settles arguments without making the standard explicit, so the same disagreement returns every month."}], "correct": "B", "select": 1}, {"id": "a2-137", "domain": 2, "task_id": "2.4", "objective": "Determine when human review or additional verification is required", "situation": "An analyst asks Claude, without web search enabled, to describe the outcome of an event that occurred last week. Claude responds with a confident, detailed description.", "question": "What should the analyst do with this response?", "options": [{"letter": "A", "text": "Accept the description as accurate because Claude answered in a confident and detailed tone", "correct": false, "explanation": "Confidence and detail in a response are not reliable indicators of factual accuracy, especially for events beyond the model's knowledge cutoff."}, {"letter": "B", "text": "Treat the description as unverified and confirm the details against a current, independent source", "correct": true, "explanation": "Without web search, a response about a recent event falls outside the model's training data, so a confident-sounding answer may be fabricated and must be checked against a current source."}, {"letter": "C", "text": "Publish the description immediately since Claude is known for broad general knowledge", "correct": false, "explanation": "Publishing unverified claims about a recent event risks spreading inaccurate information that an independent source check would have caught."}, {"letter": "D", "text": "Ask Claude the same question again and accept the answer if the second response matches the first", "correct": false, "explanation": "Repeating a question to the same model without new grounding data can simply reproduce the same unverified fabrication."}], "correct": "B", "select": 1}, {"id": "a2-138", "domain": 2, "task_id": "2.4", "objective": "Determine when human review or additional verification is required", "situation": "A communications officer must publish a Claude-drafted market briefing on the company blog within the hour. Claude used web search to produce it, and every figure in the draft carries a citation.", "question": "What is the appropriate first step before she publishes?", "options": [{"letter": "A", "text": "Add a line at the foot of the briefing noting that it was prepared with the help of an AI assistant.", "correct": false, "explanation": "A disclosure line may well be required, but it does nothing about accuracy. Adding it and publishing would put unverified figures in front of an external audience."}, {"letter": "B", "text": "Open the cited sources and confirm that each figure in the draft does appear in the page it points to.", "correct": true, "explanation": "Where web search was used, the help centre's recommendation is to review Claude's cited sources, because reading the original content reveals anything that was misread or taken out of context. With citations present, checking them is faster and stronger than any other available check."}, {"letter": "C", "text": "Ask Claude how confident it is in each figure, and revise the ones it rates as least confident.", "correct": false, "explanation": "A self-reported confidence rating is not evidence. It can sound reassuring about a figure that the cited page does not actually support."}, {"letter": "D", "text": "Regenerate the briefing in a fresh conversation, and publish if the two drafts report the same figures.", "correct": false, "explanation": "Two runs can repeat the same error, so agreement between them is not confirmation. The sources are already cited and can be checked directly."}], "correct": "B", "select": 1}, {"id": "a2-139", "domain": 2, "task_id": "2.4", "objective": "Determine when human review or additional verification is required", "situation": "A manager has two Claude outputs in front of her: notes from an internal brainstorm, and a statement that will be sent to a regulator.", "question": "What should determine how much verification each one receives?", "options": [{"letter": "A", "text": "How long each took to produce, since a quickly generated output has had less care put into it.", "correct": false, "explanation": "Generation time carries no information about correctness and none about what an error would cost."}, {"letter": "B", "text": "The length and detail of each output, since a longer output contains more claims to get wrong.", "correct": false, "explanation": "Length correlates with the number of claims but not with the cost of an error, and a single wrong sentence in the filing outweighs pages of brainstorm notes."}, {"letter": "C", "text": "What acting on each would cost, and how easily a mistake in it could afterwards be undone.", "correct": true, "explanation": "Consequence and reversibility set the bar. Brainstorm notes are cheap to correct internally; a regulatory statement is externally visible and hard to retract."}, {"letter": "D", "text": "How confident each output sounds, since hedged writing signals that Claude was less sure of it.", "correct": false, "explanation": "Tone is not a reliable indicator of accuracy, and the regulator's statement would need checking however confidently it reads."}], "correct": "C", "select": 1}, {"id": "a2-140", "domain": 2, "task_id": "2.2", "objective": "Identify hallucinations, inconsistencies, and biases in responses", "situation": "A hiring-assistant application uses Claude to write short candidate summaries. During review, evaluators notice that summaries for candidates from one demographic group consistently include more hedging language and fewer positive descriptors than summaries for equally qualified candidates from another group.", "question": "What issue does this pattern represent?", "options": [{"letter": "A", "text": "An inconsistency, since the summaries use noticeably different sentence structures overall", "correct": false, "explanation": "Varying sentence structure alone is not an inconsistency; the issue here is the systematic skew tied to group membership."}, {"letter": "B", "text": "A bias, since responses are systematically skewed by a group attribute, not qualifications", "correct": true, "explanation": "A systematic difference in tone or content correlated with a demographic group, rather than with actual qualifications, is the definition of bias in model output."}, {"letter": "C", "text": "A hallucination, since Claude invented qualifications that the candidates do not actually have", "correct": false, "explanation": "The scenario describes a skew in tone and word choice, not fabricated facts or invented credentials."}, {"letter": "D", "text": "A context failure, since Claude appears to have ignored the resumes provided in the prompt", "correct": false, "explanation": "The scenario does not indicate that resume content was ignored, only that similarly qualified candidates were described differently based on group."}], "correct": "B", "select": 1}, {"id": "a2-141", "domain": 2, "task_id": "2.1", "objective": "Evaluate Claude-generated outputs for accuracy and completeness", "situation": "An operations analyst asked six specific questions about a supplier's performance and received one flowing summary in reply. She is about to forward it to her manager.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Forward it with a note asking her manager to flag anything that looks incomplete or unclear to him.", "correct": false, "explanation": "Her manager does not have her list of six questions, so he cannot tell what is missing, and the check she owns is passed to someone less able to make it."}, {"letter": "B", "text": "Read the summary through for anything that strikes her as odd or that contradicts what she already knows.", "correct": false, "explanation": "Reading for oddities catches things she already suspects. A question that was never answered at all leaves no trace in the text, so it is exactly what this kind of reading misses."}, {"letter": "C", "text": "Ask for the summary again in a shorter form, so her manager can take it in without a long explanation.", "correct": false, "explanation": "Shortening compresses a summary whose coverage has not yet been established, and it can remove the very answers she has not verified are present."}, {"letter": "D", "text": "Go back to her six questions and confirm that each one is actually answered somewhere in the summary.", "correct": true, "explanation": "Completeness is judged against what was asked, not against how the answer reads. A fluent summary can cover four of six questions and still feel finished, so checking question by question is what exposes the gap."}], "correct": "D", "select": 1}, {"id": "a2-142", "domain": 2, "task_id": "2.4", "objective": "Determine when human review or additional verification is required", "situation": "An operations manager is preparing to let her team use Claude-drafted replies across a busy customer queue. She has not yet decided who checks what.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Send the replies out and collect complaints, so the team learns which cases were handled badly.", "correct": false, "explanation": "Learning from customer complaints puts the cost of the error onto customers, which is the outcome a review step exists to prevent."}, {"letter": "B", "text": "Ask Claude to mark the replies it is least certain about so those can be routed for checking.", "correct": false, "explanation": "A self-reported certainty score is produced by the same process as the reply, so it is not a dependable basis for routing work to a reviewer."}, {"letter": "C", "text": "Have every reply read by two colleagues, so that nothing can reach a customer unchecked at all.", "correct": false, "explanation": "Double-reading everything is expensive and unfocused, and it is a decision about process taken before anyone has defined what reviewers are looking for."}, {"letter": "D", "text": "Agree what a good reply looks like and which mistakes would actually be costly to the business.", "correct": true, "explanation": "Review only works against a standard. Naming what a correct reply contains and which failures would hurt is what tells the team where scrutiny is needed and where it is not."}], "correct": "D", "select": 1}, {"id": "a2-143", "domain": 2, "task_id": "2.2", "objective": "Identify hallucinations, inconsistencies, and biases in responses", "situation": "A project lead notices that Claude keeps giving the same wrong answer about the travel policy, always matching last year's rules. Both last year's and this year's policy documents sit in the project's knowledge.", "question": "What best explains this?", "options": [{"letter": "A", "text": "Claude is drawing on the superseded document, so the fault sits in the knowledge, not the model.", "correct": true, "explanation": "A repeatable answer that matches an old document is being read from that document. No amount of reprompting fixes an out-of-date source that is still available to be used."}, {"letter": "B", "text": "Claude has invented the policy, since fabrication is the usual cause of a confident error.", "correct": false, "explanation": "Fabrication produces varied invented detail, not the same answer every time that happens to match a real superseded document."}, {"letter": "C", "text": "Claude is biased towards older material and will always prefer whichever document is oldest.", "correct": false, "explanation": "There is no such preference for age. The problem is that two conflicting versions are both present and nothing marks which is current."}, {"letter": "D", "text": "Claude has contradicted itself within a single answer, which is the definition of an inconsistency.", "correct": false, "explanation": "Nothing here describes two conflicting statements inside one response; the answers are consistent and consistently wrong."}], "correct": "A", "select": 1}, {"id": "a2-144", "domain": 2, "task_id": "2.6", "objective": "Organize and curate information and select appropriate output formats (artifacts, inline, structured data)", "situation": "A project lead has twelve loosely related Claude answers gathered over a month and wants one reference document for her team.", "question": "What is the most effective approach?", "options": [{"letter": "A", "text": "Decide the structure the reference needs, then have Claude assemble the answers and resolve conflicts.", "correct": true, "explanation": "Curation is more than collection: choosing the structure first and then reconciling what the answers say is what turns scattered material into something a team can rely on."}, {"letter": "B", "text": "Keep the twelve answers as they are and give the team a list of which conversation holds each one.", "correct": false, "explanation": "An index still leaves colleagues reading twelve separate answers and reconciling them individually."}, {"letter": "C", "text": "Paste all twelve of the answers into one document in the order in which they were originally produced.", "correct": false, "explanation": "Chronological order reflects how the questions arose, not how a reader needs the material, and it leaves conflicts in place."}, {"letter": "D", "text": "Ask Claude to shorten each of the twelve answers so that the collection becomes quicker to read.", "correct": false, "explanation": "Shortening reduces reading time but leaves twelve disconnected pieces and can remove the detail that resolves a conflict."}], "correct": "A", "select": 1}, {"id": "a2-145", "domain": 2, "task_id": "2.4", "objective": "Determine when human review or additional verification is required", "situation": "An events team wants to put a Claude-drafted safety briefing for a public open day on the venue website tomorrow. Nobody has yet been named to check it.", "question": "What should the team lead do first?", "options": [{"letter": "A", "text": "Ask Claude to review its own briefing and confirm that the safety guidance in it is accurate.", "correct": false, "explanation": "Asking the same conversation to certify its own accuracy provides reassurance rather than assurance, and it does not put a person's name against the content."}, {"letter": "B", "text": "Decide who has to sign the briefing off, given that it is public and safety-related content.", "correct": true, "explanation": "Public, safety-related content sits at the top of the range where a named human must review and own the output before it goes out. Establishing who that person is comes before anything else, because it determines whether tomorrow's deadline is even achievable."}, {"letter": "C", "text": "Publish the briefing and ask staff on the day to correct anything that turns out to be wrong.", "correct": false, "explanation": "Correcting a published safety briefing after members of the public have read it does not undo the harm. High visibility and low reversibility are precisely the conditions that call for review before release, not after."}, {"letter": "D", "text": "Circulate the briefing to the whole team so that anyone who wishes to can comment on it.", "correct": false, "explanation": "General circulation invites comment without assigning responsibility. Everyone may assume someone else has checked the safety guidance, which is how an unreviewed briefing reaches the public."}], "correct": "B", "select": 1}, {"id": "a2-146", "domain": 2, "task_id": "2.3", "objective": "Apply fact-checking and validation techniques", "situation": "An analyst appends a reference list that Claude produced to her report without opening any of the items on it.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The list will be longer than the report itself requires, which readers are likely to find distracting.", "correct": false, "explanation": "Length is a presentation issue and says nothing about whether the sources back the claims."}, {"letter": "B", "text": "The references will be formatted in a style that her organisation's template does not permit.", "correct": false, "explanation": "Formatting is easily corrected and does not affect whether the report's claims are supported."}, {"letter": "C", "text": "A reference may not exist at all, or may exist but not support the claim attached to it.", "correct": true, "explanation": "An unopened reference list gives the report the appearance of grounding while providing none, and readers will treat it as though the checking was done."}, {"letter": "D", "text": "Readers will follow the references instead of reading the analysis that she has written herself.", "correct": false, "explanation": "Readers consulting sources is a good outcome; the danger is that the sources will not hold up."}], "correct": "C", "select": 1}, {"id": "a2-147", "domain": 2, "task_id": "2.6", "objective": "Organize and curate information and select appropriate output formats (artifacts, inline, structured data)", "situation": "A user asks Claude.ai to build a working single-page HTML landing page with embedded CSS that they can preview, tweak, and eventually export.", "question": "Which output format best serves this request?", "options": [{"letter": "A", "text": "Choose an inline HTML code block because code blocks render identically to artifacts and provide the same live preview and export options.", "correct": false, "explanation": "Inline code blocks only display the code as text; they do not offer the live preview or one-click export features that artifacts provide, so they are not equivalent."}, {"letter": "B", "text": "Use a Mermaid diagram because diagrams are the standard way to represent the structure of a web page before writing any code.", "correct": false, "explanation": "Mermaid diagrams are meant for flowcharts and diagrams, not for creating interactive web pages; they cannot fulfill the request for a working, exportable landing page."}, {"letter": "C", "text": "Use a plain-text description of the page layout because writing prose about the design avoids the risk of embedding malformed markup.", "correct": false, "explanation": "A plain-text description of the page layout fails to deliver the functional, previewable HTML page the user explicitly requested, and writing prose does not mitigate markup risks better than actual code."}, {"letter": "D", "text": "Use an artifact because a single-page HTML site is self-contained, substantial, and the user previews and refines it outside the chat.", "correct": true, "explanation": "Artifacts are designed for self-contained, substantial content like a single-page HTML site, allowing the user to preview and refine the work in a dedicated panel and later export it for use outside the chat."}], "correct": "D", "select": 1}, {"id": "a2-148", "domain": 2, "task_id": "2.6", "objective": "Organize and curate information and select appropriate output formats (artifacts, inline, structured data)", "situation": "An operations analyst is comparing eight vendors and must decide whether to ask Claude for prose or for a structured table with fixed columns. Which two are sound reasons to choose the structured form?", "question": "Select two.", "options": [{"letter": "A", "text": "The results will be moved into a tracker where every field has to line up exactly.", "correct": true, "explanation": "When output feeds a downstream system, a predictable field structure removes the manual re-keying that prose would force the analyst to do by hand."}, {"letter": "B", "text": "Tables remove the need to check the content, because the figures are already aligned.", "correct": false, "explanation": "Alignment is presentational. A neat table still needs the same accuracy and completeness checks as any other output before it is relied on."}, {"letter": "C", "text": "Structured output is inherently more accurate than the same content written as prose.", "correct": false, "explanation": "Format does not change correctness. A table can hold exactly the same wrong figure as a paragraph, presented rather more convincingly."}, {"letter": "D", "text": "A fixed set of columns makes it obvious when a vendor has no value for a field.", "correct": true, "explanation": "Visible gaps are a curation benefit distinct from downstream use: prose can pass over a missing vendor detail without the omission ever showing."}, {"letter": "E", "text": "Producing a table takes Claude noticeably less effort than writing flowing paragraphs.", "correct": false, "explanation": "Relative effort is not a reason a knowledge worker should choose a format. The decision should follow from how the information will be used."}], "correct": ["A", "D"], "select": 2}, {"id": "a2-149", "domain": 2, "task_id": "2.1", "objective": "Evaluate Claude-generated outputs for accuracy and completeness", "situation": "A programme manager has a Claude-drafted quarterly performance report and one hour to review it before it goes to her director.", "question": "In which order should these steps be taken?", "options": [{"letter": "A", "text": "List what the original request asked the report to cover; check the draft against that list for anything missing; verify the claims and figures that carry the most weight; note what was checked and pass the draft to the director.", "correct": true, "explanation": "Restating what was asked for gives her a standard; comparing the draft against it exposes missing sections cheaply and tells her where the substance actually is; verification effort then goes to the claims that matter most; recording what was checked lets the director see the limits of the review."}, {"letter": "B", "text": "List what the original request asked the report to cover; verify the claims and figures that carry the most weight; note what was checked and pass the draft to the director; check the draft against that list for anything missing.", "correct": false, "explanation": "The draft reaches the director before the completeness check, so a missing section is discovered only after the report has already been handed over."}, {"letter": "C", "text": "Verify the claims and figures that carry the most weight; check the draft against that list for anything missing; list what the original request asked the report to cover; note what was checked and pass the draft to the director.", "correct": false, "explanation": "Verification begins before anyone has established what the report was meant to contain, so the hour is spent on whatever happens to be on the page and the coverage list arrives too late to direct it."}, {"letter": "D", "text": "Check the draft against that list for anything missing; list what the original request asked the report to cover; verify the claims and figures that carry the most weight; note what was checked and pass the draft to the director.", "correct": false, "explanation": "The draft is compared against a list that has not been written yet, so the comparison has no standard behind it and reduces to reading the report and hoping something looks absent."}], "correct": "A", "select": 1}, {"id": "a2-150", "domain": 2, "task_id": "2.4", "objective": "Determine when human review or additional verification is required", "situation": "A coordinator uses Claude to draft an internal wiki page describing how to book a meeting room. Any colleague can edit the page, and a mistake would cost someone a few minutes.", "question": "What is the most appropriate level of checking?", "options": [{"letter": "A", "text": "A read-through by someone who knows the booking process, then publish it to the wiki.", "correct": true, "explanation": "Verification effort should match consequence. The page is low-stakes and easily corrected, so a quick check by someone who knows the process is proportionate."}, {"letter": "B", "text": "A formal review board should approve the page before it is added to the internal wiki at all.", "correct": false, "explanation": "A standing approval body is disproportionate for a page that is trivially reversible and carries almost no consequence."}, {"letter": "C", "text": "Every step should be performed once by a colleague who then confirms the page in writing.", "correct": false, "explanation": "Physically re-performing each step and documenting it spends far more effort than the risk of a wrong booking instruction justifies."}, {"letter": "D", "text": "No checking is needed, because internal pages are not seen by customers or outside parties.", "correct": false, "explanation": "Some check is still warranted; being internal reduces the stakes but does not remove the value of a quick look."}], "correct": "A", "select": 1}, {"id": "a2-151", "domain": 2, "task_id": "2.1", "objective": "Evaluate Claude-generated outputs for accuracy and completeness", "situation": "A bid coordinator has a Claude-assembled tender response built from the company's own past submissions, and has time to review it properly before the deadline.", "question": "Which sequence is correct?", "options": [{"letter": "A", "text": "Verify the figures against the source submissions, then polish the wording, then check coverage against the tender questions, then check the response for internal contradictions.", "correct": false, "explanation": "Verifying figures before knowing whether every tender question is answered means checking numbers in sections that may still need rewriting, and polishing precedes both remaining content checks."}, {"letter": "B", "text": "Polish the wording, then check the response for internal contradictions, then check coverage against the tender questions, then verify the figures against the source submissions.", "correct": false, "explanation": "Polishing wording first invests effort in sentences that may be deleted, and leaves the coverage check, the most decisive step for a tender, until after two rounds of editing have already been done."}, {"letter": "C", "text": "Check coverage against the tender questions, then verify the figures against the source submissions, then check the response for internal contradictions, then polish the wording.", "correct": true, "explanation": "Coverage comes first because a missing answer is the costliest defect and changes what else needs checking; accuracy of figures and internal consistency come next; wording is polished last, once the content is settled."}, {"letter": "D", "text": "Check the response for internal contradictions, then polish the wording, then verify the figures against the source submissions, then check coverage against the tender questions.", "correct": false, "explanation": "Internal consistency is a useful check but cannot be conclusive while sections are still missing, and coverage, the step that determines whether the bid is even compliant, is left to the very end."}], "correct": "C", "select": 1}, {"id": "a2-152", "domain": 2, "task_id": "2.1", "objective": "Evaluate Claude-generated outputs for accuracy and completeness", "situation": "A researcher has checked most of a twenty-claim brief but two claims could not be confirmed against any source before her deadline.", "question": "What should she do with those two?", "options": [{"letter": "A", "text": "Delete them quietly, on the basis that a brief containing only verified claims is the safest thing to circulate.", "correct": false, "explanation": "Silent deletion hides that a question was raised and not resolved, and a later reader may reintroduce the same claim having no idea it was doubted."}, {"letter": "B", "text": "Keep them but mark them clearly as unconfirmed, so readers know which parts rest on nothing she could check.", "correct": true, "explanation": "Marking them preserves information the reader needs and shows exactly where the evidence stops, which lets anyone relying on the brief decide whether those two points matter to their decision."}, {"letter": "C", "text": "Leave them in the brief unchanged, since they were not shown to be wrong and removing them would weaken it.", "correct": false, "explanation": "Presented alongside verified claims and in the same voice, unconfirmed claims inherit the credibility of the ones that were checked."}, {"letter": "D", "text": "Soften their wording so they read as more tentative, which lowers the risk without drawing attention to them.", "correct": false, "explanation": "Hedged wording signals mild uncertainty rather than absent verification, and it deliberately avoids making the gap visible."}], "correct": "B", "select": 1}, {"id": "a2-153", "domain": 2, "task_id": "2.2", "objective": "Identify hallucinations, inconsistencies, and biases in responses", "situation": "A researcher uploaded a single industry report and asked for a summary of it. Web search was switched off throughout.", "question": "Which observation is the strongest sign that a detail in the summary was fabricated rather than drawn from the report?", "options": [{"letter": "A", "text": "The summary opens with general background about the industry before it turns to the specific findings the report contains.", "correct": false, "explanation": "General framing before the specifics is a stylistic choice and is easy to remove. It becomes a concern only if the background contains checkable claims, which is not what this observation describes."}, {"letter": "B", "text": "The summary is noticeably longer than the researcher expected for a report of this size, and repeats one point in two places.", "correct": false, "explanation": "Length and repetition point to a shape-and-quality problem rather than an invention problem. A verbose or redundant summary can still be entirely faithful to the document it was built from."}, {"letter": "C", "text": "The summary states a precise figure, with a named source and a year, that appears nowhere in the uploaded report itself.", "correct": true, "explanation": "With no web search and a single uploaded document, everything factual should trace back to that document. A precise, confidently attributed figure that is absent from the source is the classic signature of fabricated detail."}, {"letter": "D", "text": "The summary uses phrasing of its own rather than reusing the exact sentences that appear in the uploaded industry report.", "correct": false, "explanation": "Paraphrasing is normal and often desirable in a summary; different wording says nothing about whether the underlying content came from the report. This would flag almost every competent summary as suspect."}], "correct": "C", "select": 1}, {"id": "a2-154", "domain": 2, "task_id": "2.1", "objective": "Evaluate Claude-generated outputs for accuracy and completeness", "situation": "A finance officer pasted eighty rows of expense data into a chat and asked Claude to total them by cost centre. Which two verifications should she perform on the returned figures?", "question": "Select two.", "options": [{"letter": "A", "text": "Look over the totals to see whether they seem about right for a month of this kind.", "correct": false, "explanation": "A plausibility glance catches only large errors. A total that is wrong by one misfiled row looks entirely normal, which is what makes this check weak on its own."}, {"letter": "B", "text": "Ask Claude to check its own arithmetic and proceed once it reports that all is correct.", "correct": false, "explanation": "Self-checking produces another generated answer, not an independent verification, and it can confidently confirm a figure that was wrong to begin with."}, {"letter": "C", "text": "Confirm that all eighty of the rows are counted in exactly one of the totals.", "correct": true, "explanation": "This is a coverage test rather than an arithmetic one: totals can each be added correctly while several rows were skipped or double-counted."}, {"letter": "D", "text": "Recalculate a sample of the cost centre totals herself and compare them with the output.", "correct": true, "explanation": "An independent recalculation is direct evidence about accuracy, and it catches a total that is internally consistent but simply computed wrongly."}, {"letter": "E", "text": "Check that currency symbols and decimal places are applied consistently across the table.", "correct": false, "explanation": "Consistent formatting says nothing about whether the numbers are right. Tidy presentation is often what makes an incorrect total feel trustworthy."}], "correct": ["C", "D"], "select": 2}, {"id": "a2-155", "domain": 2, "task_id": "2.5", "objective": "Edit, adapt, refine, and compare outputs for the intended audience", "situation": "A city communications team drafts a public road-construction notice, but the first draft uses formal legal phrasing like 'aforementioned thoroughfare' that is likely to confuse residents.", "question": "What is the best refinement for this audience?", "options": [{"letter": "A", "text": "Keep the legal phrasing exactly as written but enlarge the font of the notice.", "correct": false, "explanation": "Enlarging the font does not make the phrasing any easier to understand; the notice would still contain confusing legal terms. The core issue is word choice, not visibility."}, {"letter": "B", "text": "Translate the phrase 'aforementioned thoroughfare' into another commonly spoken local language.", "correct": false, "explanation": "Translating jargon into another language does not guarantee it becomes plain or accessible; it may remain just as formal and confusing. The goal is to replace complex terms with clear, everyday language, not merely translate them."}, {"letter": "C", "text": "Replace legal jargon with plain language, and ensure all facts remain accurate and complete.", "correct": true, "explanation": "Plain language makes the notice understandable to all residents, not just those familiar with legal jargon. Accuracy must be maintained so the public receives correct and complete information about the road construction."}, {"letter": "D", "text": "Add a formal cover letter from city officials before the notice to provide necessary context.", "correct": false, "explanation": "A formal cover letter would introduce even more official language, failing to simplify the confusing terms in the notice itself. Residents would still encounter the same unclear phrasing."}], "correct": "C", "select": 1}, {"id": "a2-156", "domain": 2, "task_id": "2.3", "objective": "Apply fact-checking and validation techniques", "situation": "A Claude draft attributes a sentence to a named industry report as a direct quotation.", "question": "What is the appropriate validation step?", "options": [{"letter": "A", "text": "Open the report and find that exact sentence, since a close paraphrase may not appear in it.", "correct": true, "explanation": "A quotation asserts that specific words appear in a specific document, so only locating those words in the report validates it."}, {"letter": "B", "text": "Ask Claude for the page number, and accept the quotation once a number has been supplied to her.", "correct": false, "explanation": "A page number can be generated as readily as the quotation, so it adds no independent confirmation."}, {"letter": "C", "text": "Search for the sentence's key phrase and accept it if similar wording appears anywhere.", "correct": false, "explanation": "Similar wording elsewhere does not establish that this report contains this sentence."}, {"letter": "D", "text": "Confirm that the report exists and was published, which establishes the quotation's origin.", "correct": false, "explanation": "The report existing says nothing about whether the quoted sentence is in it."}], "correct": "A", "select": 1}, {"id": "a2-157", "domain": 2, "task_id": "2.2", "objective": "Identify hallucinations, inconsistencies, and biases in responses", "situation": "A communications officer asks Claude for balanced talking points on a contested planning decision. The draft states one side's arguments plainly and introduces the other side's with repeated hedging.", "question": "How should this be classified?", "options": [{"letter": "A", "text": "A hallucination, because the arguments on one side must therefore have been invented.", "correct": false, "explanation": "Nothing suggests the arguments are fabricated. The issue is how comparable arguments are framed, not whether they exist."}, {"letter": "B", "text": "A bias, because the response treats comparable arguments unevenly through its choice of framing.", "correct": true, "explanation": "Systematically softer language for one side skews the reader's impression even where every individual statement is accurate, which is what bias in an output looks like."}, {"letter": "C", "text": "A formatting problem, because the hedging simply reflects a different paragraph style in use.", "correct": false, "explanation": "Hedging changes how strongly a claim is asserted, which is a matter of substance rather than presentation."}, {"letter": "D", "text": "An inconsistency, because the two halves of the response use rather different sentence constructions.", "correct": false, "explanation": "An inconsistency is a conflict between statements about the same fact, not a difference in phrasing between sections."}], "correct": "B", "select": 1}, {"id": "a2-158", "domain": 2, "task_id": "2.5", "objective": "Edit, adapt, refine, and compare outputs for the intended audience", "situation": "A product team compares two Claude-drafted release notes for the same feature, one for a public changelog and one for an internal engineering wiki.", "question": "What is a key factor to check for each draft's fit?", "options": [{"letter": "A", "text": "Whether the public draft includes internal ticket numbers so support staff can reference them", "correct": false, "explanation": "Internal ticket numbers are meaningful to engineers but are not something a general public audience needs or can act on."}, {"letter": "B", "text": "Whether the internal draft is written with the same marketing tone as the public changelog", "correct": false, "explanation": "A marketing tone suits the public draft but is not a meaningful fit for an internal wiki meant for precise technical reference."}, {"letter": "C", "text": "Whether both drafts are published to the same location so all readers see identical content", "correct": false, "explanation": "Publishing both drafts to the same location defeats the purpose of tailoring one for public users and one for internal engineers."}, {"letter": "D", "text": "Whether the public draft avoids internal codenames and detail that the engineering draft can include", "correct": true, "explanation": "A public changelog should be free of internal codenames and implementation detail, while the internal wiki draft can appropriately retain that detail for engineers."}], "correct": "D", "select": 1}, {"id": "a3-001", "domain": 3, "task_id": "3.3", "objective": "Align model selection with task requirements (cost, speed, quality)", "situation": "A team's answers are frequently wrong, and someone proposes moving everything to the deepest-reasoning tier to fix it.", "question": "What is the greatest risk in doing that?", "options": [{"letter": "A", "text": "The deepest tier will take longer to respond, which will slow the team's daily work down considerably.", "correct": false, "explanation": "Response speed is a real trade-off but a minor one next to continuing to act on wrong answers."}, {"letter": "B", "text": "If the cause is a vague prompt or a stale document, they buy a more confident version of the error.", "correct": true, "explanation": "Reasoning depth cannot supply missing context or override an out-of-date source, so the wrong answers persist while the allowance is consumed faster."}, {"letter": "C", "text": "The team will grow used to the deepest tier and find lighter tiers unsatisfactory afterwards.", "correct": false, "explanation": "Preference drift is speculative and is not the reason this move can fail."}, {"letter": "D", "text": "The deepest tier will refuse tasks it considers too simple, so routine work will stop being done.", "correct": false, "explanation": "No tier refuses work for being simple; that is not a behaviour a user encounters."}], "correct": "B", "select": 1}, {"id": "a3-002", "domain": 3, "task_id": "3.1", "objective": "Select appropriate Claude product features (Projects, research mode, chat, artifacts)", "situation": "A consultant needs Claude to both draft a client proposal referencing the consultant's past proposal templates and also look up the client company's most recent press releases before finalizing the draft.", "question": "Which approach best fits both needs?", "options": [{"letter": "A", "text": "Work inside a Project with the past templates uploaded to its knowledge base and additionally use research mode within that conversation to gather recent press releases.", "correct": true, "explanation": "A Project's knowledge base is designed to store the consultant's own reusable templates, while research mode can be used within that conversation to gather current external information like recent press releases. This approach combines both features to meet the two distinct needs efficiently."}, {"letter": "B", "text": "Use only research mode to search the web for similar proposal templates and the client's latest press releases, then reference the templates while incorporating the releases into the draft.", "correct": false, "explanation": "Research mode searches the web but does not have access to the consultant's private past proposal templates unless they are uploaded. It would only find publicly available templates, not the specific proprietary ones needed, so it cannot fully substitute for project knowledge."}, {"letter": "C", "text": "Use only a Project with uploaded templates, because its knowledge base will automatically have the client's most recent press releases alongside the consultant's past proposals.", "correct": false, "explanation": "Project knowledge only contains documents that the user has manually uploaded; it does not automatically pull in a company's current press releases. Without an uploaded source or research mode, the client's latest press releases would not be available."}, {"letter": "D", "text": "Avoid both features by manually pasting the templates and press releases into a single long chat message so the model processes everything at once and drafts the proposal.", "correct": false, "explanation": "Manually pasting all templates and press releases into a long message is more effort-intensive and does not provide the persistent reuse benefit of project knowledge. It also lacks the automated, cited gathering that research mode provides, making it less efficient for this task."}], "correct": "A", "select": 1}, {"id": "a3-003", "domain": 3, "task_id": "3.2", "objective": "Differentiate between Claude model types (Haiku, Sonnet, Opus)", "situation": "A department sets the deepest-reasoning model as the default for all of its work, including short routine replies.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "Short replies will be written at greater length than the person asking the question wanted.", "correct": false, "explanation": "Response length is controlled by the request, not by the tier, and it is easy to correct."}, {"letter": "B", "text": "Routine work consumes far more allowance, so it runs out during the work that actually needs it.", "correct": true, "explanation": "The deepest tier uses meaningfully more of the usage allowance per message, so spending it on trivial work leaves less available for the genuinely hard problems."}, {"letter": "C", "text": "Short replies will come back noticeably slower, which colleagues will find frustrating to work with.", "correct": false, "explanation": "Slower replies are an annoyance rather than the substantive cost of this policy."}, {"letter": "D", "text": "The department will be unable to switch to a lighter model once a default has been set for it.", "correct": false, "explanation": "The model can be changed at any time, so nothing about the choice is locked in."}], "correct": "B", "select": 1}, {"id": "a3-004", "domain": 3, "task_id": "3.1", "objective": "Select appropriate Claude product features (Projects, research mode, chat, artifacts)", "situation": "", "question": "Which of the following best describes the core difference between using Claude's research feature and using a standard chat conversation for a question?", "options": [{"letter": "A", "text": "Research mode conducts multiple agentic, interconnected searches to build a thorough, cited answer, while standard chat usually responds from existing knowledge or a single lookup.", "correct": true, "explanation": "Research mode performs multiple agentic searches that build upon each other to produce a comprehensive, cited answer. In contrast, standard chat typically draws from its pre-trained knowledge or a single lookup, lacking this multi-step investigation."}, {"letter": "B", "text": "Research mode and standard chat use the same underlying model and generation pipeline, differing only in interface color and the inclusion of a search summary, without functional change.", "correct": false, "explanation": "The two differ functionally: research mode’s agentic search process fundamentally changes how information is gathered and used before responding. It is not merely a cosmetic or superficial interface change."}, {"letter": "C", "text": "Research mode operates without generating any citations, producing concise answers solely from search summaries, while standard chat meticulously appends citation markers to verify every assertion.", "correct": false, "explanation": "Research mode does generate citations, producing thorough answers with easy-to-check references. Standard chat does not always include citation markers, contrary to the claim made here."}, {"letter": "D", "text": "Standard chat conversations can access and analyze uploaded project files and attachments, whereas research mode relies solely on the model's pre-existing training data with no access to external documents.", "correct": false, "explanation": "Research mode is specifically designed to perform external searches, not rely solely on pre-existing training data. Standard chat’s ability to analyze project files is a separate capability and not what distinguishes these modes."}], "correct": "A", "select": 1}, {"id": "a3-005", "domain": 3, "task_id": "3.1", "objective": "Select appropriate Claude product features (Projects, research mode, chat, artifacts)", "situation": "A market analyst must produce a competitor briefing drawn from many public sources, and the team will keep editing that briefing for weeks afterwards. Which two Claude features fit this job?", "question": "Select two.(Select 2)", "options": [{"letter": "A", "text": "Keep the briefing inline in the chat and re-paste the newest version whenever an edit lands.", "correct": false, "explanation": "Re-pasting an inline answer for every edit is precisely the friction artifacts exist to remove, and versions drift apart quickly."}, {"letter": "B", "text": "Turn on a mail connector first so that the competitor facts are drawn from the team inbox.", "correct": false, "explanation": "An internal mailbox is not the source of public competitor information, and connecting one does not answer the sourcing need here."}, {"letter": "C", "text": "Use Research mode, so the briefing is built by multi-step searching and returns citations.", "correct": true, "explanation": "Research mode runs an agentic, multi-step search and returns citations, which is exactly what a sourced competitor briefing needs."}, {"letter": "D", "text": "Ask for the briefing as an artifact, so the team can edit and reuse it outside the chat.", "correct": true, "explanation": "An artifact is the durable, editable deliverable format, so weeks of later editing happen on one object rather than in chat replies."}, {"letter": "E", "text": "Open a separate chat for each competitor so that no single conversation ever grows long.", "correct": false, "explanation": "Splitting by competitor fragments the briefing and does nothing about sourcing or about the editing the team has ahead of it."}], "correct": ["C", "D"], "select": 2}, {"id": "a3-006", "domain": 3, "task_id": "3.1", "objective": "Select appropriate Claude product features (Projects, research mode, chat, artifacts)", "situation": "A user wants Claude to plan a multi-day team offsite by checking the team's shared Google Calendar for open dates, reviewing relevant emails about budget constraints, and cross-referencing venue reviews found on the web.", "question": "Which combination of feature and prerequisite is needed?", "options": [{"letter": "A", "text": "Enable Research mode with web search and connect Google services as data sources.", "correct": true, "explanation": "Enabling Research mode with web search allows Claude to access and cross-reference venue reviews on the web. Connecting Google services such as Calendar and Gmail as data sources provides live access to open dates and budget emails, meeting all the user's needs."}, {"letter": "B", "text": "A basic chat conversation without web search or external services won't work.", "correct": false, "explanation": "Although a basic chat conversation without web search or external services indeed cannot handle the task, this option does not propose a working combination of feature and prerequisite. The question asks for the combination needed, not a statement of what won't work."}, {"letter": "C", "text": "Using an artifact that shows a static calendar for the offsite avoids web search.", "correct": false, "explanation": "An artifact displaying a static calendar for the offsite avoids web search, but the task explicitly requires web search for venue reviews. Artifacts also do not provide live calendar checks or email review for budget constraints."}, {"letter": "D", "text": "Using a Project containing the team's quarterly calendar PDF can find open dates.", "correct": false, "explanation": "A Project containing a quarterly calendar PDF only captures static snapshots that can quickly become outdated. It lacks the ability to check live calendar availability, review emails about budget constraints, or search the web for venue reviews."}], "correct": "A", "select": 1}, {"id": "a3-007", "domain": 3, "task_id": "3.3", "objective": "Align model selection with task requirements (cost, speed, quality)", "situation": "A daily customer-insight digest that used to be quick now takes far longer and uses much more of the team's allowance than it did six months ago, although the tier it runs on has never changed.", "question": "What should the team do first?", "options": [{"letter": "A", "text": "Move the digest up to the deepest-reasoning tier, on the basis that stronger answers should need fewer re-runs each morning.", "correct": false, "explanation": "Escalating adds cost to a task that is already consuming too much, and depth does not address inputs that have quietly grown. It is as likely to worsen the allowance problem as to fix the re-runs."}, {"letter": "B", "text": "Spread the digest across several colleagues' accounts, so that the daily allowance draw is shared out rather than falling on just one.", "correct": false, "explanation": "Spreading the draw across accounts hides the consumption rather than reducing it, leaves the underlying growth undiagnosed, and is likely to conflict with how the organisation expects allowances to be used."}, {"letter": "C", "text": "Move the digest down to the fastest and cheapest tier straight away and see whether the daily output is still good enough.", "correct": false, "explanation": "Changing tier is a plausible eventual answer, but doing it first treats a symptom. If the digest is now processing far more material and being re-run repeatedly, a lighter tier will simply do the bloated task less well."}, {"letter": "D", "text": "Look at how the digest itself has grown, in the length of what is fed in and the number of re-runs, before changing tier at all.", "correct": true, "explanation": "Nothing about the tier has changed, so the growth is in the task. Understanding how much longer the inputs are and how often the digest is being re-run identifies the real cause before any tier decision is made."}], "correct": "D", "select": 1}, {"id": "a3-008", "domain": 3, "task_id": "3.4", "objective": "Understand and manage context limitations and memory considerations (when to restart, summarize, or persist)", "situation": "A researcher's chat approaches its context limit, and Claude begins condensing the earlier messages in that conversation. Which two statements about this are accurate?", "question": "Select two.", "options": [{"letter": "A", "text": "Condensing raises the context window of that single chat right up to the next size that is available.", "correct": false, "explanation": "The window is a property of the model in use, and condensing does not enlarge it."}, {"letter": "B", "text": "Condensing permanently erases those earlier messages from the conversation history for good.", "correct": false, "explanation": "Nothing is erased, so treating condensing as deletion misreads what is happening in the conversation."}, {"letter": "C", "text": "Claude condenses the earlier messages so that the conversation is able to carry on past that point.", "correct": true, "explanation": "As a conversation nears the context limit, earlier messages are condensed so the chat can continue rather than stop."}, {"letter": "D", "text": "The full chat history remains accessible to her even after the earlier messages are condensed.", "correct": true, "explanation": "Condensing changes what is held in working memory; the full history of the conversation stays available to read."}, {"letter": "E", "text": "Condensing appears as a signal that her account has run out of usage allowance for the period.", "correct": false, "explanation": "This confuses length limits with usage limits; nearing a context limit says nothing about the period's allowance."}], "correct": ["C", "D"], "select": 2}, {"id": "a3-009", "domain": 3, "task_id": "3.3", "objective": "Align model selection with task requirements (cost, speed, quality)", "situation": "A manager lets each colleague pick a model tier per task according to how important the task feels, with nothing written down.", "question": "What is the greatest concern?", "options": [{"letter": "A", "text": "Colleagues will always choose the deepest tier, which will exhaust the team's allowance quickly.", "correct": false, "explanation": "Some will over-select and some will under-select, and assuming one direction misses the actual problem."}, {"letter": "B", "text": "The manager will be unable to see which tier each colleague chose for any given piece of work.", "correct": false, "explanation": "Visibility is a reporting question and is not what makes the arrangement unreliable."}, {"letter": "C", "text": "Colleagues will spend a large part of each working day deciding which tier they ought to use.", "correct": false, "explanation": "The decision itself is quick; the cost lies in its inconsistency rather than in the time it takes."}, {"letter": "D", "text": "Similar work gets different treatment, and nobody can tell whether the cheaper choice sufficed.", "correct": true, "explanation": "With no shared standard the same task is handled inconsistently, and because nothing defines adequacy there is no way to learn from any of the choices made."}], "correct": "D", "select": 1}, {"id": "a3-010", "domain": 3, "task_id": "3.1", "objective": "Select appropriate Claude product features (Projects, research mode, chat, artifacts)", "situation": "A colleague suggests running Research for a question with a single settled answer that is already well documented internally.", "question": "How should this suggestion be assessed?", "options": [{"letter": "A", "text": "It is a good fit, because a multi-step search always produces a more reliable answer overall.", "correct": false, "explanation": "More searching does not automatically mean a better answer when the question is already settled and documented."}, {"letter": "B", "text": "It is a good fit, because Research is the only way to obtain citations for any given factual claim.", "correct": false, "explanation": "Citations are useful, but Research is not the only route to a sourced answer and the value here does not justify the effort."}, {"letter": "C", "text": "It is a poor fit, because Research is built for multi-step investigation and uses limits faster.", "correct": true, "explanation": "Research runs several searches that build on one another, which is valuable for open questions and disproportionate for a settled fact; it also draws on usage limits more quickly."}, {"letter": "D", "text": "It is a poor fit, because Research cannot answer questions that have one settled answer.", "correct": false, "explanation": "It can answer such questions perfectly well. The objection is proportionality, not capability."}], "correct": "C", "select": 1}, {"id": "a3-011", "domain": 3, "task_id": "3.4", "objective": "Understand and manage context limitations and memory considerations (when to restart, summarize, or persist)", "situation": "A teacher's project instructions have grown to several pages, and she notices less room for the lesson material she pastes into chats.", "question": "What is the best adjustment?", "options": [{"letter": "A", "text": "Duplicate the instructions into project knowledge as well, so that Claude encounters them twice over.", "correct": false, "explanation": "Duplicating them consumes more space rather than less, and repetition does not make guidance more reliable."}, {"letter": "B", "text": "Leave the instructions exactly as they are, since project instructions never take up any conversation space.", "correct": false, "explanation": "They do take up space in every chat in the project, which is precisely why she is seeing less room."}, {"letter": "C", "text": "Trim the instructions to the essentials, since they occupy window space in every chat inside the project.", "correct": true, "explanation": "Project instructions are carried into each conversation in the project, so keeping them concise leaves more of the window for the material she is actually working on."}, {"letter": "D", "text": "Move the instructions into the first message of every chat so they no longer take up any window space.", "correct": false, "explanation": "The same words consume the same space wherever they sit, and she also loses the durability of project instructions."}], "correct": "C", "select": 1}, {"id": "a3-012", "domain": 3, "task_id": "3.1", "objective": "Select appropriate Claude product features (Projects, research mode, chat, artifacts)", "situation": "A user asks Claude to generate a 40-line Python script that parses log files, then five minutes later asks Claude to fix a bug in that same script.", "question": "Where does the updated script most likely appear so the user can compare it to the earlier version?", "options": [{"letter": "A", "text": "In the artifact panel, updated as a new version accessible through the version selector", "correct": true, "explanation": "A 40-line script is substantial, self-contained content suited to an artifact; when Claude revises it, the update appears in the same artifact panel, and the version selector lets the user switch between the original and fixed versions."}, {"letter": "B", "text": "In a newly created Project, since bug fixes always require project-level knowledge storage", "correct": false, "explanation": "Fixing a bug in an already-generated script does not require creating a new Project; the existing artifact already supports iterative revision without added project structure."}, {"letter": "C", "text": "Nowhere retrievable, since Claude cannot revise previously generated code in the same conversation", "correct": false, "explanation": "Claude can revise previously generated code within the same conversation, and with an artifact, the fixed version remains retrievable through version history."}, {"letter": "D", "text": "In research mode results, since fixing a bug counts as an investigative web search task", "correct": false, "explanation": "Research mode is for gathering external information via web search, not for revising code that Claude already generated in the conversation."}], "correct": "A", "select": 1}, {"id": "a3-013", "domain": 3, "task_id": "3.3", "objective": "Align model selection with task requirements (cost, speed, quality)", "situation": "A team is setting up a new recurring weekly task and has to decide how much reasoning depth it needs.", "question": "Which sequence is correct?", "options": [{"letter": "A", "text": "Pick a tier by how important the task feels, then run it for a month, then write down what good output means, then compare the month's results against that.", "correct": false, "explanation": "Choosing on instinct and defining quality a month later means there is no standard to judge the intervening results against, and a month of work has already been committed to an unexamined choice."}, {"letter": "B", "text": "Run the task on several tiers at once, then write down what good output means, then compare the results against that definition, then keep the tier that did best.", "correct": false, "explanation": "Committing several tiers to the task before any standard exists means the trial is designed without knowing what it is testing for, and it duplicates the work and the allowance draw on every run."}, {"letter": "C", "text": "Compare a sample of results with what colleagues produce by hand, then write down what good output means, then pick a tier on that basis, then run the task each week from then on.", "correct": false, "explanation": "Comparing results before the task has been run is out of order, and benchmarking against manual work is not the same as agreeing what this task's output has to achieve."}, {"letter": "D", "text": "Write down what good output means, then run a real sample on the default tier, then compare that sample against the definition, then change tier only where it falls short.", "correct": true, "explanation": "Define the quality bar, test the default tier on real work, check that sample against the bar, and change tier only where the evidence shows the default falls short or is more than the task needs."}], "correct": "D", "select": 1}, {"id": "a3-014", "domain": 3, "task_id": "3.1", "objective": "Select appropriate Claude product features (Projects, research mode, chat, artifacts)", "situation": "A grants officer on a paid plan needs a one-off overview of the funding landscape, drawn from the published criteria of roughly twenty funders she has not looked at before. Her director will ask where each statement came from.", "question": "Which approach best fits the job?", "options": [{"letter": "A", "text": "Ask a separate question about each of the twenty funders in ordinary chats, then merge the twenty replies into an overview herself.", "correct": false, "explanation": "Twenty separate questions is slow, and it puts the entire burden of deciding what to investigate and of stitching the findings together back on the officer, which is the work she wanted help with."}, {"letter": "B", "text": "Use Research, which runs the multi-step search across the funders itself and returns an answer with sources she can open and check.", "correct": true, "explanation": "Research is built for exactly this: an agentic multi-step search across many external sources that decides what to look at next and returns an answer with sources attached, which is what her director's question requires."}, {"letter": "C", "text": "Set up a project for the overview, put the twenty funder names into the project instructions and work through them chat by chat.", "correct": false, "explanation": "A project provides durable shared context for recurring work, but this is a one-off task and the funder criteria are external material she has not gathered. The project adds setup without doing any of the searching."}, {"letter": "D", "text": "Ask one broad question in an ordinary chat, then ask for a reference list to be added underneath the overview once it is written.", "correct": false, "explanation": "References requested after the overview is written are assembled to fit text that was produced without them, so they are the weakest form of provenance and still leave every claim to be traced by hand."}], "correct": "B", "select": 1}, {"id": "a3-015", "domain": 3, "task_id": "3.4", "objective": "Understand and manage context limitations and memory considerations (when to restart, summarize, or persist)", "situation": "A consultant on a paid plan is checking how much material she can work with inside one chat. Which two statements are correct?", "question": "Select two.", "options": [{"letter": "A", "text": "The context window grows automatically as soon as she upgrades to a higher paid plan.", "correct": false, "explanation": "The window is tied to the model in use, not to how high the plan tier is."}, {"letter": "B", "text": "The largest chat context window is offered on Claude Opus 5 and Claude Sonnet 5 only.", "correct": true, "explanation": "In chat, the largest context window is limited to Claude Opus 5 and Claude Sonnet 5; other models sit at smaller sizes."}, {"letter": "C", "text": "Every model on a paid plan offers exactly the same context window inside a single chat.", "correct": false, "explanation": "Context window size varies by model, so a paid plan on its own does not give every model the same window."}, {"letter": "D", "text": "A usage limit covers how much she interacts over a period, not what fits in one chat.", "correct": true, "explanation": "A usage limit is a budget for interaction over a time period, while the context window is the working memory of one conversation."}, {"letter": "E", "text": "Her remaining usage allowance is what decides how much one chat can hold at any time.", "correct": false, "explanation": "This is the classic confusion between usage limits and length limits; the allowance does not set a chat's capacity."}], "correct": ["B", "D"], "select": 2}, {"id": "a3-016", "domain": 3, "task_id": "3.1", "objective": "Select appropriate Claude product features (Projects, research mode, chat, artifacts)", "situation": "A user asks Claude to research mode-style investigate an emerging technical standard, but Claude answers immediately from general knowledge without doing multiple searches.", "question": "What is the most likely reason research mode did not activate?", "options": [{"letter": "A", "text": "Web search was not enabled, or the user did not explicitly prompt Claude for research, so it fell back to a normal response.", "correct": true, "explanation": "Research mode requires web search to be enabled, and if the user does not explicitly prompt Claude to use research capabilities, it may fall back to a normal response. This explains why Claude answered from general knowledge without performing multiple searches."}, {"letter": "B", "text": "Research mode only works when the query is phrased as a single yes-or-no question, so Claude defaults to general knowledge without initiating research.", "correct": false, "explanation": "Research mode is designed for open-ended, multi-faceted investigative queries, not limited to yes-or-no questions. Query phrasing does not determine whether research mode initiates."}, {"letter": "C", "text": "Research mode was permanently removed from Claude.ai and replaced entirely by Projects; all multi-step research now happens within project setups.", "correct": false, "explanation": "Research mode is still available on Claude.ai and has not been replaced by Projects. They are separate features designed for different use cases, so research mode remains a distinct capability."}, {"letter": "D", "text": "The user's account tier does not support any form of web-connected search, regardless of plan, so Claude cannot initiate research mode when requested.", "correct": false, "explanation": "Research mode is supported on certain paid account tiers, so it is not true that no tier supports web-connected search. The scenario does not provide account details, and the feature is not universally unavailable."}], "correct": "A", "select": 1}, {"id": "a3-017", "domain": 3, "task_id": "3.1", "objective": "Select appropriate Claude product features (Projects, research mode, chat, artifacts)", "situation": "A user asks Claude to write a two-sentence reply to a coworker's Slack message. Claude responds with the two sentences directly in the conversation rather than opening a side panel.", "question": "Why did this response not trigger an artifact?", "options": [{"letter": "A", "text": "Artifacts are disabled by default and must be manually enabled for every conversation, preventing side panel opening for any content.", "correct": false, "explanation": "Artifacts are enabled by default and appear based on content length and complexity, not a manual per-conversation setting. Short replies simply do not meet the threshold for triggering an artifact."}, {"letter": "B", "text": "The content is short and conversational, not substantial or self-contained text typically exceeding 15 lines.", "correct": true, "explanation": "Artifacts are designed for substantial, self-contained content, typically over 15 lines, that users may want to modify or reference later. A brief two-sentence reply is short and conversational, so it stays inline rather than opening a side panel."}, {"letter": "C", "text": "Research mode must be active for an artifact to appear, and short conversational replies do not automatically enable this mode.", "correct": false, "explanation": "Research mode and artifacts are separate features; activating research mode is not required for an artifact to appear. Short, conversational replies do not generate artifacts regardless of whether research mode is on."}, {"letter": "D", "text": "Two-sentence replies are only supported when composing within a Project's knowledge base, not in standard chat conversations.", "correct": false, "explanation": "Two-sentence replies are standard in any chat conversation and are not restricted to Projects' knowledge bases. The content's length and self-contained nature—not the conversation type—determine artifact creation."}], "correct": "B", "select": 1}, {"id": "a3-018", "domain": 3, "task_id": "3.2", "objective": "Differentiate between Claude model types (Haiku, Sonnet, Opus)", "situation": "A support operations lead must classify a large daily backlog of short inbound messages, and separately needs one deep analysis of a tangled contract dispute. Which two choices fit?", "question": "Select two.", "options": [{"letter": "A", "text": "Use the deepest-reasoning tier for both of the jobs, since the strongest option is never bad.", "correct": false, "explanation": "Using the deepest tier for routine classification spends quota on work that does not need the extra reasoning depth."}, {"letter": "B", "text": "Use the fastest, cheapest tier for both jobs, since the tiers differ mainly in reply speed.", "correct": false, "explanation": "The tiers differ in reasoning depth as well as speed, so the lightest tier is a poor fit for a tangled dispute."}, {"letter": "C", "text": "Use the deepest-reasoning tier for the contract dispute, accepting the larger quota it uses.", "correct": true, "explanation": "A tangled one-off analysis is the case where deeper reasoning is worth the meaningfully larger quota it consumes."}, {"letter": "D", "text": "Use the fastest, cheapest tier for the high-volume classification, where throughput matters.", "correct": true, "explanation": "High-volume, low-complexity classification is exactly where the fastest and cheapest tier earns its place."}, {"letter": "E", "text": "Use whichever tier the team already has open, since the tier rarely changes what comes back.", "correct": false, "explanation": "Tier choice changes depth, speed and quota use, so treating it as incidental is how both of these jobs go wrong."}], "correct": ["C", "D"], "select": 2}, {"id": "a3-019", "domain": 3, "task_id": "3.4", "objective": "Understand and manage context limitations and memory considerations (when to restart, summarize, or persist)", "situation": "A marketer says she has 'run out of context', yet she can immediately open a new chat and continue working without waiting.", "question": "Which statement fits what happened?", "options": [{"letter": "A", "text": "She reached a usage limit for the period; that one chat still had plenty of room left inside it.", "correct": false, "explanation": "A usage limit would block her in every chat, including the new one, until the period reset."}, {"letter": "B", "text": "She reached neither limit, because new chats are only blocked when a plan is downgraded or cancelled.", "correct": false, "explanation": "She clearly hit something; the evidence that a new chat works points specifically to a length limit."}, {"letter": "C", "text": "She reached a length limit inside that one chat; her overall usage allowance was not the constraint.", "correct": true, "explanation": "Length limits are confined to a single conversation's context window, which is why a new chat lets her continue straight away."}, {"letter": "D", "text": "She reached both limits at once, since a chat filling up always exhausts the period's allowance too.", "correct": false, "explanation": "The two limits are independent. Filling one conversation says nothing about the allowance left for the period."}], "correct": "C", "select": 1}, {"id": "a3-020", "domain": 3, "task_id": "3.4", "objective": "Understand and manage context limitations and memory considerations (when to restart, summarize, or persist)", "situation": "A project lead has let a project's knowledge base grow to hundreds of documents, including several superseded versions of the same policy.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "Retrieval will stop working altogether once the knowledge base grows past the model's context window.", "correct": false, "explanation": "It is precisely at that point that retrieval activates automatically, holding far more content than a window alone."}, {"letter": "B", "text": "Retrieval will hold the project to the plan's smallest context window until documents have been removed.", "correct": false, "explanation": "The window is determined by the model in use; retrieval extends reach over the knowledge base rather than shrinking it."}, {"letter": "C", "text": "Retrieval may surface an outdated version, giving confident answers that no amount of reprompting fixes.", "correct": true, "explanation": "When the wrong source is retrieved, the answer is wrong at the root. Rewording the prompt cannot correct a problem that lives in the knowledge base itself."}, {"letter": "D", "text": "Retrieval will require her to configure a search index before the project can answer any questions.", "correct": false, "explanation": "No setup or configuration is required; retrieval turns itself on when the knowledge base grows large enough."}], "correct": "C", "select": 1}, {"id": "a3-021", "domain": 3, "task_id": "3.1", "objective": "Select appropriate Claude product features (Projects, research mode, chat, artifacts)", "situation": "A support team lead wants several colleagues to be able to view and use the same set of onboarding documents and a shared system of custom instructions when chatting with Claude, without each person needing to configure their own settings.", "question": "Which feature addresses this on a Team or Enterprise plan?", "options": [{"letter": "A", "text": "An artifact exported as a file and emailed to each team member separately", "correct": false, "explanation": "Exporting and emailing an artifact distributes a static file rather than providing an ongoing shared workspace with consistent instructions."}, {"letter": "B", "text": "A single long chat conversation that is copied and sent to each colleague individually", "correct": false, "explanation": "Copying and distributing a single chat transcript does not give colleagues an ongoing shared, editable space with consistent instructions and knowledge."}, {"letter": "C", "text": "A shared Project that team members are added to with view or edit permissions", "correct": true, "explanation": "Team and Enterprise projects can be shared with specific members or the whole organization, with permission levels like 'Can use' and 'Can edit', giving everyone consistent access to the same knowledge base and instructions."}, {"letter": "D", "text": "Research mode configured to fetch the onboarding documents whenever asked", "correct": false, "explanation": "Research mode searches the web and connected sources on demand; it is not a mechanism for sharing a persistent internal knowledge base and instructions among team members."}], "correct": "C", "select": 1}, {"id": "a3-022", "domain": 3, "task_id": "3.1", "objective": "Select appropriate Claude product features (Projects, research mode, chat, artifacts)", "situation": "A user is having a quick, one-off conversation with Claude to brainstorm five possible names for a new podcast. The conversation does not need to be revisited later or reference any uploaded files.", "question": "Which feature is the most appropriate fit for this task?", "options": [{"letter": "A", "text": "A new Project created specifically to store the podcast name brainstorm", "correct": false, "explanation": "Creating a Project adds persistent knowledge-base structure that is unnecessary overhead for a single one-off brainstorm that won't be revisited."}, {"letter": "B", "text": "An artifact, so the list of names appears in a dedicated side panel", "correct": false, "explanation": "A short list of five names is not substantial, standalone content requiring iterative editing, so an artifact is not needed for this simple exchange."}, {"letter": "C", "text": "Research mode, so Claude searches the web for existing podcast names to avoid", "correct": false, "explanation": "Enabling research mode to search the web for existing podcast names is unnecessary for a quick five-name brainstorm and would add latency without meaningful benefit."}, {"letter": "D", "text": "A standard chat conversation, without creating a Project or enabling research mode", "correct": true, "explanation": "A short, self-contained brainstorming task with no need for persistent knowledge or external research fits naturally into a standard chat conversation without the added structure of a Project, research mode, or an artifact."}], "correct": "D", "select": 1}, {"id": "a3-023", "domain": 3, "task_id": "3.4", "objective": "Understand and manage context limitations and memory considerations (when to restart, summarize, or persist)", "situation": "A project lead is told she cannot send any more messages for several hours, and the block applies even in a brand-new empty chat.", "question": "Which limit has she reached?", "options": [{"letter": "A", "text": "A length limit, which governs how much information a single chat can work with at one time.", "correct": false, "explanation": "A length limit is confined to one conversation's context window; starting a fresh chat would clear it immediately."}, {"letter": "B", "text": "A knowledge limit, which governs how many documents a single project workspace is able to store.", "correct": false, "explanation": "A cap on stored documents would not stop her sending messages in an unrelated new chat."}, {"letter": "C", "text": "An artifact limit, which governs how many separate documents Claude can produce within one session.", "correct": false, "explanation": "There is no such block on messaging, and artifacts are an output format rather than a messaging quota."}, {"letter": "D", "text": "A usage limit, which governs how much she can interact with Claude over a given period of time.", "correct": true, "explanation": "Usage limits control how much you can interact with Claude across a time period, so they follow you into every chat, including a new one."}], "correct": "D", "select": 1}, {"id": "a3-024", "domain": 3, "task_id": "3.4", "objective": "Understand and manage context limitations and memory considerations (when to restart, summarize, or persist)", "situation": "A researcher notices that colleagues' very long chats keep running smoothly, while hers stops making room for new content once it fills up.", "question": "What is the most likely explanation?", "options": [{"letter": "A", "text": "Automatic context management is unavailable to her because her plan does not include the Artifacts feature.", "correct": false, "explanation": "Artifacts are an output format and are widely available; they have no bearing on context management."}, {"letter": "B", "text": "Automatic context management is unavailable to her because code execution is not switched on in settings.", "correct": true, "explanation": "Code execution must be enabled for automatic context management to work, so without it a conversation simply runs out of room."}, {"letter": "C", "text": "Automatic context management is unavailable to her because web search has not been enabled for the chat.", "correct": false, "explanation": "Web search affects what Claude can look up, not whether earlier messages can be condensed."}, {"letter": "D", "text": "Automatic context management is unavailable to her because she has too many documents inside a Project.", "correct": false, "explanation": "The size of a project's knowledge base does not govern whether a conversation can condense itself."}], "correct": "B", "select": 1}, {"id": "a3-025", "domain": 3, "task_id": "3.3", "objective": "Align model selection with task requirements (cost, speed, quality)", "situation": "A project manager drafts talking points live during meetings and wants each suggestion back immediately, knowing she will rewrite them afterwards.", "question": "Which tier fits best?", "options": [{"letter": "A", "text": "The default middle tier, since anything lighter cannot produce usable talking points at all.", "correct": false, "explanation": "The lightest tier handles short drafting perfectly well, particularly when a person rewrites the result."}, {"letter": "B", "text": "The deepest-reasoning tier, since talking points shape decisions and need the strongest reasoning.", "correct": false, "explanation": "The drafts are rewritten afterwards, so depth she discards is depth she paid for unnecessarily."}, {"letter": "C", "text": "The fastest tier, since immediate response is the binding constraint and the drafts are rough.", "correct": true, "explanation": "Where an immediate reply is what makes the tool usable and the output is a rough first pass, speed is worth more than additional reasoning depth."}, {"letter": "D", "text": "Whichever tier holds the largest context window, since meetings involve a lot of discussion.", "correct": false, "explanation": "Window size governs how much a conversation can hold, not how quickly a short suggestion arrives."}], "correct": "C", "select": 1}, {"id": "a3-026", "domain": 3, "task_id": "3.1", "objective": "Select appropriate Claude product features (Projects, research mode, chat, artifacts)", "situation": "A free-tier user wants to organize their conversations about a personal budgeting spreadsheet into a dedicated space, separate from unrelated chats about recipes and travel planning.", "question": "Is this possible without upgrading to a paid plan?", "options": [{"letter": "A", "text": "No, because a Pro subscription is required to create a Project, so free accounts cannot have separate spaces.", "correct": false, "explanation": "Anthropic's documentation explicitly states that Projects are available to all users, including free accounts. A Pro subscription is not required; free users can create a maximum of five Projects."}, {"letter": "B", "text": "Yes, but only through research mode, which is available on free accounts and organizes chats into separate threads.", "correct": false, "explanation": "The research does not identify a 'research mode' as the documented way to create separate conversation spaces on free accounts. The supported feature is Projects, which free users can create up to five of."}, {"letter": "C", "text": "Yes, because Projects are available to free accounts, and free users can create a maximum of five Projects.", "correct": true, "explanation": "Yes. According to Anthropic's official documentation, Projects are available to all users, including free Claude accounts, and free users can create up to five Projects. A Project is a self-contained workspace with its own chat history and knowledge base, so the user can isolate budgeting conversations without upgrading."}, {"letter": "D", "text": "No, because free accounts are limited to one continuous chat, preventing users from separating conversations by topic.", "correct": false, "explanation": "Free accounts are not limited to one continuous chat. Users on the free plan can create Projects, each with its own chat history, allowing separation by topic such as budgeting, recipes, or travel."}], "correct": "C", "select": 1}, {"id": "a3-027", "domain": 3, "task_id": "3.2", "objective": "Differentiate between Claude model types (Haiku, Sonnet, Opus)", "situation": "An operations team runs one large workload through Claude: several thousand short, repetitive record summaries, plus one difficult reconciliation that decides how the whole set is interpreted. Everything currently runs on the deepest-reasoning tier and the allowance is gone by Wednesday.", "question": "What should the team do first?", "options": [{"letter": "A", "text": "Ask for a larger allowance before making any change, since the current work is valuable and the shortfall is only a couple of days.", "correct": false, "explanation": "Asking for more allowance funds the existing pattern rather than examining it, and it leaves thousands of routine summaries consuming the most expensive capacity available for no additional benefit."}, {"letter": "B", "text": "Cut the length of every prompt in the workload, so that each individual request costs less and the allowance stretches further.", "correct": false, "explanation": "Shorter prompts trim the cost of each request but leave the real problem untouched: routine work is still being run on the tier reserved for difficult reasoning, so the same imbalance recurs next week."}, {"letter": "C", "text": "Move the whole workload to the fastest and cheapest tier, then see over the next few weeks whether anyone complains about quality.", "correct": false, "explanation": "Moving everything down applies one answer to two very different tasks, and the reconciliation is exactly the piece that would suffer. Waiting for complaints is a poor detection mechanism for degraded reasoning."}, {"letter": "D", "text": "Separate the bulk repetitive work from the one genuinely difficult piece, so each part can be matched to what it actually needs.", "correct": true, "explanation": "The workload is not one thing. Splitting the high-volume routine part from the small part that genuinely needs deep reasoning is what makes any sensible tier decision possible, and it is where nearly all the waste is."}], "correct": "D", "select": 1}, {"id": "a3-028", "domain": 3, "task_id": "3.1", "objective": "Select appropriate Claude product features (Projects, research mode, chat, artifacts)", "situation": "A student is using Claude to draft a single essay outline and does not plan to reuse any reference material or revisit the conversation after submitting the assignment. A classmate suggests creating a Project first.", "question": "What is the most reasonable evaluation of that suggestion?", "options": [{"letter": "A", "text": "Required, since Claude cannot generate an essay outline without an active Project, as the outlining process depends on referencing materials, which is only enabled within a Project.", "correct": false, "explanation": "Claude can generate an essay outline directly in a standard chat without needing a Project. While referencing materials can be helpful, it is not required for drafting, and documents can be uploaded in regular conversations if needed."}, {"letter": "B", "text": "Unnecessary, since Projects are exclusively for storing images and cannot contain text-based outlines, as their document handling is limited to visual media like diagrams.", "correct": false, "explanation": "The reasoning here is false: Projects are not exclusively for storing images; they can contain text documents, code, and other file types. Although the conclusion that creating a Project is unnecessary is correct, a reasonable evaluation must be based on accurate information."}, {"letter": "C", "text": "Required, since only Projects allow more than one exchange within a conversation, meaning a student cannot ask follow-up questions or iterate on the outline without a Project.", "correct": false, "explanation": "Standard chat conversations already support unlimited back-and-forth exchanges. A student can ask follow-up questions and iterate without a Project, so this requirement is false."}, {"letter": "D", "text": "Unnecessary, since a Project's main benefit—persistent knowledge and context across multiple conversations—is not needed for a single one-time outline.", "correct": true, "explanation": "A Project’s primary benefit is providing persistent knowledge and context that can be reused across multiple conversations. For a one-time outline task, this persistent structure adds no value and the work can be done in a standard chat."}], "correct": "D", "select": 1}, {"id": "a3-029", "domain": 3, "task_id": "3.3", "objective": "Align model selection with task requirements (cost, speed, quality)", "situation": "A strategy lead must reconcile three conflicting demand forecasts and explain the reasoning to an executive committee.", "question": "Which tier fits this work best?", "options": [{"letter": "A", "text": "Whichever tier her colleagues most recently used, so the team's outputs remain consistent.", "correct": false, "explanation": "Consistency of tier is not a goal in itself, and it ignores what this particular task demands."}, {"letter": "B", "text": "The deepest-reasoning tier, accepting the higher allowance cost, since the value is the reasoning.", "correct": true, "explanation": "When several conflicting inputs must be held together and the argument itself is the deliverable, the extra reasoning depth is what the higher allowance cost buys."}, {"letter": "C", "text": "The default middle tier, because reconciling forecasts is fairly ordinary analytical work for a team.", "correct": false, "explanation": "Reconciling contradictory sources and defending the reasoning is precisely the harder end of the range."}, {"letter": "D", "text": "The fastest tier, because the committee needs the analysis quickly and speed is what matters.", "correct": false, "explanation": "Speed is not the binding constraint when the whole value of the piece lies in the quality of the reasoning."}], "correct": "B", "select": 1}, {"id": "a3-030", "domain": 3, "task_id": "3.4", "objective": "Understand and manage context limitations and memory considerations (when to restart, summarize, or persist)", "situation": "An operations manager has kept one chat running for three weeks to plan a vendor migration. Claude's latest answers now contradict constraints the team agreed in the first week.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Send the same question again with stronger wording so that Claude gives the earlier decisions more weight.", "correct": false, "explanation": "Repeating a prompt more forcefully does not restore detail that no longer fits in the conversation's working memory."}, {"letter": "B", "text": "Identify the decisions and constraints that must survive, and record them as durable reference material.", "correct": true, "explanation": "The conversation has outgrown what a single chat can hold, so the working state has to be captured before anything is restarted or moved. Naming what must survive is the step every other option depends on."}, {"letter": "C", "text": "Switch to a model with a smaller window so the conversation is forced to stay short and focused.", "correct": false, "explanation": "A smaller context window reduces how much Claude can hold at once; it makes the loss of earlier detail worse rather than better."}, {"letter": "D", "text": "Ask Claude to repeat the whole conversation back to her in full before she continues working in it.", "correct": false, "explanation": "This spends window space and usage allowance restating material that is already the problem, and a recap produced from a strained conversation is not a reliable record."}], "correct": "B", "select": 1}, {"id": "a3-031", "domain": 3, "task_id": "3.4", "objective": "Understand and manage context limitations and memory considerations (when to restart, summarize, or persist)", "situation": "An operations manager is worried that letting a long negotiation chat condense itself will destroy the record of what was said.", "question": "Which statement is accurate?", "options": [{"letter": "A", "text": "Earlier messages are condensed so the chat can continue, and the full chat history stays readable to her.", "correct": true, "explanation": "Automatic context management compresses what is held in working memory to make room, while the complete history remains accessible in the conversation."}, {"letter": "B", "text": "Earlier messages are moved into project knowledge automatically, where colleagues can then also read them.", "correct": false, "explanation": "Condensed content is not exported into a project, and nothing is shared with colleagues as a side effect."}, {"letter": "C", "text": "Earlier messages are stripped out of the chat entirely, so only the condensed version can be reviewed.", "correct": false, "explanation": "Nothing is removed from the history. Compression affects working memory rather than the stored conversation."}, {"letter": "D", "text": "Earlier messages are exported to a file on her account, which she must download before the chat resumes.", "correct": false, "explanation": "There is no download step; the conversation continues without any manual action from her."}], "correct": "A", "select": 1}, {"id": "a3-032", "domain": 3, "task_id": "3.4", "objective": "Understand and manage context limitations and memory considerations (when to restart, summarize, or persist)", "situation": "A communications lead simply starts a new chat whenever the current one begins to feel unwieldy, without saving anything out of it.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "Her earlier conversations are permanently deleted from her account as soon as a new chat is opened.", "correct": false, "explanation": "Past chats remain in her history; the loss is that the new conversation does not inherit their content."}, {"letter": "B", "text": "Each new chat counts against a monthly cap on conversations, which she will exhaust before month end.", "correct": false, "explanation": "Allowance is measured by how much she interacts over a period, not by a count of conversations opened."}, {"letter": "C", "text": "Decisions and constraints agreed earlier are lost, so she covers the same ground over and over again.", "correct": true, "explanation": "Restarting clears the window but also discards the durable output. Without persisting it, each new chat rebuilds context that was already settled."}, {"letter": "D", "text": "Opening new chats reduces the context window available to her for the rest of that working day.", "correct": false, "explanation": "The window is per conversation and per model; opening a new chat gives her a full one rather than a reduced one."}], "correct": "C", "select": 1}, {"id": "a3-033", "domain": 3, "task_id": "3.3", "objective": "Align model selection with task requirements (cost, speed, quality)", "situation": "A marketing team wants to move its weekly performance report from the deepest-reasoning model to a lighter one to save allowance.", "question": "What should they do first?", "options": [{"letter": "A", "text": "Ask Claude whether a lighter model could produce the same report to the same standard.", "correct": false, "explanation": "That answer is a general claim about models rather than evidence from their own report."}, {"letter": "B", "text": "Switch the report over immediately, and change back if anyone complains about the quality.", "correct": false, "explanation": "Waiting for complaints makes readers the quality check, and subtle thinning of the analysis rarely produces a complaint."}, {"letter": "C", "text": "Shorten the report first, so that a lighter model would have less material to work through each week.", "correct": false, "explanation": "Cutting scope changes the deliverable rather than testing whether the lighter tier can produce it."}, {"letter": "D", "text": "Produce next week's report on both tiers and compare them against what the report must deliver.", "correct": true, "explanation": "A side-by-side run on their own work, judged against an agreed standard, is the only evidence that tells them whether the cheaper tier is adequate here."}], "correct": "D", "select": 1}, {"id": "a3-034", "domain": 3, "task_id": "3.2", "objective": "Differentiate between Claude model types (Haiku, Sonnet, Opus)", "situation": "A project lead is setting up a new recurring piece of work and is unsure which Claude model to use for it.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Pick the deepest-reasoning model, since starting with the strongest option is always safest.", "correct": false, "explanation": "The deepest tier consumes meaningfully more allowance, and reaching for it by default spends that budget on work that does not need it."}, {"letter": "B", "text": "Judge how much reasoning depth the work really needs, since that is what separates the tiers.", "correct": true, "explanation": "The tiers trade reasoning depth against speed and allowance, so the choice starts with an honest read of how hard the task actually is."}, {"letter": "C", "text": "Pick the fastest and cheapest model, since a lighter model can be upgraded at any later point.", "correct": false, "explanation": "Defaulting to the lightest tier is the same mistake in the other direction, and it can hide a quality gap she never looks for."}, {"letter": "D", "text": "Ask Claude which of the available models it would recommend for the work she has described.", "correct": false, "explanation": "That answer is generated from a general description rather than from her actual quality bar and constraints."}], "correct": "B", "select": 1}, {"id": "a3-035", "domain": 3, "task_id": "3.1", "objective": "Select appropriate Claude product features (Projects, research mode, chat, artifacts)", "situation": "A data analyst asks Claude to generate a single-page HTML dashboard with an embedded chart that visualizes quarterly revenue, and wants to be able to download the finished HTML file.", "question": "Which feature is designed to produce and export this kind of standalone content?", "options": [{"letter": "A", "text": "The chat history sidebar, which lists previous conversations by date", "correct": false, "explanation": "The chat history sidebar only organizes past conversations; it does not generate or export standalone downloadable content."}, {"letter": "B", "text": "A Project's custom instructions, which define Claude's tone for future chats", "correct": false, "explanation": "Custom instructions shape how Claude responds across a project's conversations; they do not render or export a standalone HTML dashboard file."}, {"letter": "C", "text": "Artifacts, which render self-contained content like HTML pages and support exporting or downloading", "correct": true, "explanation": "Artifacts are built for substantial, self-contained content such as single-page HTML websites, and they support exporting content by copying or downloading the file."}, {"letter": "D", "text": "Research mode, which searches external sites for existing revenue dashboards", "correct": false, "explanation": "Research mode gathers external web information, but generating and exporting a self-contained HTML dashboard is the artifact feature's role, not a search function."}], "correct": "C", "select": 1}, {"id": "a3-036", "domain": 3, "task_id": "3.2", "objective": "Differentiate between Claude model types (Haiku, Sonnet, Opus)", "situation": "A department shares one pooled allowance. Several colleagues have got into the habit of running everyday drafting and tidying-up work on the deepest-reasoning tier because it is available to them.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The department loses track of which tier produced which document, making it harder to reproduce a particular draft afterwards.", "correct": false, "explanation": "Not recording which tier produced a draft is an untidy practice, but it rarely affects the outcome of everyday drafting and does not explain why the allowance is being exhausted."}, {"letter": "B", "text": "The routine drafts come back more slowly than they need to, so each colleague spends a little longer waiting for everyday work.", "correct": false, "explanation": "Slower turnaround on routine drafting is a real irritation and worth avoiding, but a few extra seconds per draft is minor next to the shared pool running dry before the hard work gets done."}, {"letter": "C", "text": "Colleagues become accustomed to the deeper tier's style of answer and find the default tier's phrasing harder to work with later.", "correct": false, "explanation": "Getting used to one tier's phrasing is a mild adjustment, not a risk to the department's work. Everyday drafting is well within what the default tier handles, and preference is not the issue here."}, {"letter": "D", "text": "The shared allowance is consumed by low-value work, so it is not there when a colleague has genuinely hard analysis to run.", "correct": true, "explanation": "The deepest tier consumes meaningfully more of a shared allowance, so spending it on work the default tier handles well leaves nothing for the small number of tasks that genuinely need that depth."}], "correct": "D", "select": 1}, {"id": "a3-037", "domain": 3, "task_id": "3.2", "objective": "Differentiate between Claude model types (Haiku, Sonnet, Opus)", "situation": "A colleague new to Claude at work asks why anyone would use the default tier when a deeper-reasoning one is offered alongside it.", "question": "Which explanation is accurate?", "options": [{"letter": "A", "text": "The default tier exists to hold costs down and produces work that is noticeably weaker than the deeper tier on every kind of task attempted.", "correct": false, "explanation": "This understates the default badly. It is a capable tier that handles most everyday work well; the difference in reasoning depth shows on genuinely hard problems, not uniformly across every task."}, {"letter": "B", "text": "The default tier is the quickest and cheapest option available, and it is intended for very high volumes of short, repetitive requests.", "correct": false, "explanation": "This describes the lightest tier rather than the default. Being fastest and cheapest for very high-volume short requests is a different position in the range from the general-purpose default."}, {"letter": "C", "text": "The default tier is fast and capable enough for most everyday work, so the deeper tier's greater allowance cost is worth paying only when depth decides it.", "correct": true, "explanation": "The default is fast and capable and suits the majority of work, while the deepest tier consumes meaningfully more allowance, so it is reserved for tasks where reasoning depth changes the outcome."}, {"letter": "D", "text": "The default tier is intended for people who are still learning to use Claude, and experienced colleagues are generally expected to move up once they feel confident.", "correct": false, "explanation": "Tier choice tracks the demands of the task, not the experience of the user. An expert doing routine drafting is well served by the default, and a beginner facing a hard analysis may need more depth."}], "correct": "C", "select": 1}, {"id": "a3-038", "domain": 3, "task_id": "3.1", "objective": "Select appropriate Claude product features (Projects, research mode, chat, artifacts)", "situation": "A marketing analyst needs Claude to compile a report on a competitor's recent product launches by pulling in current news articles, press releases, and blog posts published in the last month.", "question": "Which feature should the analyst use?", "options": [{"letter": "A", "text": "Request an artifact so Claude generates a standalone report document with current news and press releases.", "correct": false, "explanation": "Artifacts are a way to display and interact with substantial content generated by Claude, but they do not have web-search capabilities. Requesting an artifact does not enable gathering of current news or press releases."}, {"letter": "B", "text": "Enable the research feature for Claude to run agentic web searches over current sources.", "correct": true, "explanation": "The research feature allows Claude to perform agentic web searches, automatically exploring multiple sources like news articles, press releases, and blogs to gather current information. It synthesizes findings and cites sources, making it ideal for compiling a report on recent competitor launches."}, {"letter": "C", "text": "Lower the model to a smaller size to generate the report from current news and press releases.", "correct": false, "explanation": "Lowering the model to a smaller size affects response speed and cost, not the ability to access external information. Web search functionality is tied to the research feature, not model size."}, {"letter": "D", "text": "Open a Project and rely on its stored knowledge to pull recent articles and press releases for the report.", "correct": false, "explanation": "A Project's stored knowledge consists only of documents manually uploaded by the user; it does not fetch live external content. Therefore, it cannot pull recent articles and press releases from the web."}, {"letter": "E", "text": "Continue in a standard chat conversation to compile the report using recent news and press releases.", "correct": false, "explanation": "Standard chat without research mode typically relies on Claude's internal training data or a single search, which is insufficient for compiling comprehensive, up-to-date information from multiple external sources like news and press releases."}], "correct": "B", "select": 1}, {"id": "a3-039", "domain": 3, "task_id": "3.1", "objective": "Select appropriate Claude product features (Projects, research mode, chat, artifacts)", "situation": "Five colleagues each keep re-uploading the same three reference documents into their own chats, and they want one shared place for that context.", "question": "What should the team do first?", "options": [{"letter": "A", "text": "Check which plan they are on, since sharing one project with colleagues needs Team or Enterprise.", "correct": true, "explanation": "On personal plans a project is private to its creator, so on Free, Pro or Max each colleague would still end up maintaining a separate copy of the same knowledge."}, {"letter": "B", "text": "Nominate one person to hold the documents and forward the answers on to the rest of the team each time.", "correct": false, "explanation": "Routing every question through one colleague creates a bottleneck and does not give anyone else the shared context they need."}, {"letter": "C", "text": "Ask Claude to remember the three documents so that everyone's chats can draw on them later.", "correct": false, "explanation": "Context from one person's conversations does not carry across to colleagues' chats; shared knowledge has to be held somewhere deliberately."}, {"letter": "D", "text": "Create a project each and upload the same three documents into every one of those projects.", "correct": false, "explanation": "Five parallel copies is the duplication they are trying to remove, and the copies will drift apart as documents are updated."}], "correct": "A", "select": 1}, {"id": "a3-040", "domain": 3, "task_id": "3.3", "objective": "Align model selection with task requirements (cost, speed, quality)", "situation": "A support team must triage several thousand incoming messages a week, of which a small proportion involve genuinely difficult judgment.", "question": "What is the most sensible arrangement?", "options": [{"letter": "A", "text": "Run the bulk on the fastest tier and send the cases it flags as difficult to the deepest tier.", "correct": true, "explanation": "A two-stage arrangement matches each message to the depth it needs, so volume is handled cheaply and the small hard tail still gets proper reasoning."}, {"letter": "B", "text": "Run everything on the deepest tier, so that the difficult cases are never handled by a lighter one.", "correct": false, "explanation": "Paying the deepest tier's allowance cost on thousands of routine messages to protect a handful of hard ones is a poor trade."}, {"letter": "C", "text": "Run everything on the fastest tier, since the difficult cases are only a small proportion.", "correct": false, "explanation": "The hard cases are usually where the cost of a mistake sits, so treating them as negligible is the expensive error."}, {"letter": "D", "text": "Alternate between tiers message by message, so that the allowance is spread evenly over time.", "correct": false, "explanation": "Alternating assigns depth at random, which means some hard cases get the lightest treatment."}], "correct": "A", "select": 1}, {"id": "a3-041", "domain": 3, "task_id": "3.1", "objective": "Select appropriate Claude product features (Projects, research mode, chat, artifacts)", "situation": "A market analyst must produce a competitor briefing drawn from many public sources, and the team will keep editing that briefing for weeks afterwards. Which two Claude features fit this job?", "question": "Select two.", "options": [{"letter": "A", "text": "Open a separate chat for each competitor so that no single conversation ever grows long.", "correct": false, "explanation": "Splitting by competitor fragments the briefing and does nothing about sourcing or about the editing the team has ahead of it."}, {"letter": "B", "text": "Ask for the briefing as an artifact, so the team can edit and reuse it outside the chat.", "correct": true, "explanation": "An artifact is the durable, editable deliverable format, so weeks of later editing happen on one object rather than in chat replies."}, {"letter": "C", "text": "Use Research mode, so the briefing is built by multi-step searching and returns citations.", "correct": true, "explanation": "Research mode runs an agentic, multi-step search and returns citations, which is exactly what a sourced competitor briefing needs."}, {"letter": "D", "text": "Turn on a mail connector first so that the competitor facts are drawn from the team inbox.", "correct": false, "explanation": "An internal mailbox is not the source of public competitor information, and connecting one does not answer the sourcing need here."}, {"letter": "E", "text": "Keep the briefing inline in the chat and re-paste the newest version whenever an edit lands.", "correct": false, "explanation": "Re-pasting an inline answer for every edit is precisely the friction artifacts exist to remove, and versions drift apart quickly."}], "correct": ["B", "C"], "select": 2}, {"id": "a3-042", "domain": 3, "task_id": "3.2", "objective": "Differentiate between Claude model types (Haiku, Sonnet, Opus)", "situation": "A logistics company processes a very high volume of straightforward shipment-status classifications per day and has tight per-request latency requirements, but still wants strong reasoning quality for edge cases.", "question": "Which model tier is the best fit?", "options": [{"letter": "A", "text": "Claude Sonnet, since high-volume straightforward tasks are always better served by a mid-tier model", "correct": false, "explanation": "While Claude Sonnet offers a balance of intelligence and speed, Anthropic specifically designates Claude Haiku for straightforward classifications. There is no universal rule that mid-tier models are always better for such tasks; the choice depends on the specific requirements for latency, cost, and capability."}, {"letter": "B", "text": "A model chosen at random each request, since reasoning quality does not depend on which tier handles the request", "correct": false, "explanation": "Reasoning quality varies significantly across model tiers, and randomly selecting a model would lead to inconsistent performance and unpredictable costs. Best practice is to select the most appropriate model based on the task's requirements."}, {"letter": "C", "text": "Claude Opus, since strong reasoning on edge cases requires the most capable and expensive tier regardless of volume", "correct": false, "explanation": "Claude Opus is the most capable model, but using it for every request in a high-volume scenario would be extremely costly and introduce higher latency, making it unsuitable for straightforward classification tasks. The documentation advises using the appropriate model tier based on task complexity and budget."}, {"letter": "D", "text": "Claude Haiku, since it offers near-frontier performance with the fastest speed at the most economical price point", "correct": true, "explanation": "Anthropic documentation recommends Claude Haiku for straightforward classification tasks like routing and intent detection, emphasizing its speed and cost-efficiency. It provides near-frontier performance at the most economical price, making it ideal for high-volume, low-latency requirements. For edge cases requiring deeper reasoning, a multi-model architecture can escalate requests to more capable models."}], "correct": "D", "select": 1}, {"id": "a3-043", "domain": 3, "task_id": "3.2", "objective": "Differentiate between Claude model types (Haiku, Sonnet, Opus)", "situation": "A records team reclassifies several thousand short archive descriptions overnight, and an archivist spot-checks a sample of the results each morning before anything is published.", "question": "Which consideration should carry most weight in the team's tier choice?", "options": [{"letter": "A", "text": "The preference of the archivist who runs the morning check, since she is the person most familiar with how the descriptions read.", "correct": false, "explanation": "The archivist's familiarity makes her a good reviewer, but personal preference is not a basis for a decision about throughput and allowance across thousands of items processed overnight."}, {"letter": "B", "text": "The overall importance of the archive to the organisation, since anything touching the permanent record deserves the best available.", "correct": false, "explanation": "General importance is too blunt to guide the choice. It would justify the deepest tier for every task the team ever runs, including thousands of short, formulaic descriptions that do not need that depth."}, {"letter": "C", "text": "Whichever tier the rest of the organisation has settled on, so that the records team's results stay comparable with everyone else's.", "correct": false, "explanation": "Matching what others use ignores the specifics that actually matter here: this is high-volume, repetitive work with a human check already in place, which is not the shape of most colleagues' tasks."}, {"letter": "D", "text": "The volume and routine nature of the descriptions, together with the fact that a person already reviews a sample before publication.", "correct": true, "explanation": "High volume, short and repetitive items, and an existing human spot-check together point to speed and allowance as the deciding factors rather than the deepest available reasoning."}], "correct": "D", "select": 1}, {"id": "a3-044", "domain": 3, "task_id": "3.3", "objective": "Align model selection with task requirements (cost, speed, quality)", "situation": "A team runs two workloads that are alike in volume and subject matter. One is edited line by line by a specialist before use; the other is published automatically with no human reading it.", "question": "Which consideration should weigh most heavily when choosing a tier for each?", "options": [{"letter": "A", "text": "The subject matter the two workloads share, since comparable material should be handled the same way for consistency of output.", "correct": false, "explanation": "Shared subject matter makes the two workloads look alike, but it is not what determines the risk. Consistency for its own sake ignores the fact that only one of them has anyone checking the result."}, {"letter": "B", "text": "The volume of the two workloads, since equal volumes imply equal throughput needs and therefore the same balance of speed and cost.", "correct": false, "explanation": "Equal volume does mean similar throughput demands, and that matters. It is still secondary to the fact that one workload reaches the outside world without any person having read it."}, {"letter": "C", "text": "How much a downstream human check would catch, since the unchecked workload has to carry its own quality with no safety net.", "correct": true, "explanation": "The strength of the downstream check changes how much the output has to get right on its own, so the unchecked workload justifies more reasoning depth while the edited one can lean towards speed and allowance."}, {"letter": "D", "text": "The seniority of the people who own each workload, since the more senior owner is accountable for the more visible of the two.", "correct": false, "explanation": "Ownership determines who answers for a problem, not how likely a problem is. Seniority is an accountability question and does not tell the team where errors would actually escape unnoticed."}], "correct": "C", "select": 1}, {"id": "a3-045", "domain": 3, "task_id": "3.4", "objective": "Understand and manage context limitations and memory considerations (when to restart, summarize, or persist)", "situation": "An operations manager has deliberately kept one chat alive for months, letting Claude condense earlier messages rather than ever starting fresh.", "question": "What is the biggest concern with this habit?", "options": [{"letter": "A", "text": "A conversation that has been condensed can no longer be exported or shared, so her team is unable to review the thread once it passes the length limit", "correct": false, "explanation": "Condensing does not restrict exporting or sharing a conversation; that is unrelated to how context is managed as a chat grows."}, {"letter": "B", "text": "Condensing earlier messages deletes them from her chat history for good, so nobody on the team is able to go back and reread what was originally agreed", "correct": false, "explanation": "Condensing affects what Claude carries forward as working context; the full chat history remains accessible for her to scroll back and read."}, {"letter": "C", "text": "Automatic condensing is offered only on free accounts, so a paid plan will refuse to shorten the conversation and will simply stop accepting further messages", "correct": false, "explanation": "Automatic context management is not a free-account-only behaviour, and a conversation reaching its length limit does not simply refuse further messages in the way described."}, {"letter": "D", "text": "The decisions her team relies on exist only inside one conversation, and Claude now works from a condensed version rather than what was originally agreed", "correct": true, "explanation": "Two things go wrong at once: the durable decisions are trapped in a single fragile conversation instead of being persisted somewhere reusable, and once earlier messages are condensed Claude is reasoning from a summary rather than the original wording. Moving what must survive into a Project fixes both."}], "correct": "D", "select": 1}, {"id": "a3-046", "domain": 3, "task_id": "3.2", "objective": "Differentiate between Claude model types (Haiku, Sonnet, Opus)", "situation": "A policy writer expects roughly nine rounds of ordinary wording and structure changes on a report, with one genuinely hard judgement in the middle about how to reconcile two conflicting evidence bases.", "question": "How should she spread the work across tiers?", "options": [{"letter": "A", "text": "Run every round on the fastest and cheapest tier, and accept whatever it produces for the reconciliation to keep the process simple.", "correct": false, "explanation": "Simplicity is bought at the cost of the one step that matters. The reconciliation of conflicting evidence is precisely the task where shallower reasoning is most likely to produce a weak result."}, {"letter": "B", "text": "Do the wording rounds on the default tier and reserve the deepest-reasoning tier for the one hard reconciliation in the middle.", "correct": true, "explanation": "The default tier is fast, capable and right for routine drafting and wording iterations, while the extra allowance cost of the deepest tier is worth paying only for the single step where reasoning depth genuinely decides the outcome."}, {"letter": "C", "text": "Run every round on the deepest-reasoning tier, so that the hard judgement benefits from all the context built up in earlier rounds.", "correct": false, "explanation": "Running nine wording rounds on the most expensive tier spends heavily on work the default tier handles well, and the accumulated context does not require every earlier round to have used the deeper tier."}, {"letter": "D", "text": "Alternate between the tiers round by round, so that the report gets a mix of perspectives as the drafting work progresses.", "correct": false, "explanation": "Alternating by round bears no relation to what each round demands, so the hard judgement may land on a lighter tier by chance while routine wording changes consume the deeper one."}], "correct": "B", "select": 1}, {"id": "a3-047", "domain": 3, "task_id": "3.1", "objective": "Select appropriate Claude product features (Projects, research mode, chat, artifacts)", "situation": "A recruiting team answers similar candidate questions every week and re-pastes the same rubric and tone rules into every new chat. Which two moves fit the available features?", "question": "Select two.", "options": [{"letter": "A", "text": "Ask Claude at the end of each chat to remember the rubric for whichever chat opens next.", "correct": false, "explanation": "A request made inside one chat does not durably carry team context into the next one; that is what a Project is for."}, {"letter": "B", "text": "Move the rubric and the tone rules into a Project, so each chat starts from that context.", "correct": true, "explanation": "A Project is the feature for recurring work that needs shared context, so the rubric and tone rules stop being re-pasted every week."}, {"letter": "C", "text": "Run every one of the candidate questions through Research mode so answers arrive with citations.", "correct": false, "explanation": "Research mode is for agentic multi-step investigation of open questions, not for applying a known internal rubric."}, {"letter": "D", "text": "Keep the reusable reply template as an artifact the team can open and edit between chats.", "correct": true, "explanation": "The reply template is self-contained content the team returns to and edits, which is what an artifact is for."}, {"letter": "E", "text": "Keep one very long running chat open so the rubric stays somewhere in the history above.", "correct": false, "explanation": "One ever-growing chat pushes against the context limit and still leaves the rubric buried instead of durably configured."}], "correct": ["B", "D"], "select": 2}, {"id": "a3-048", "domain": 3, "task_id": "3.4", "objective": "Understand and manage context limitations and memory considerations (when to restart, summarize, or persist)", "situation": "A teacher pastes a whole term of student feedback into a single chat and keeps asking Claude to summarise themes as the conversation grows very long.", "question": "What is the greatest risk to the quality of her summaries?", "options": [{"letter": "A", "text": "The chat will refuse to accept any further messages until she upgrades to a higher-priced plan tier.", "correct": false, "explanation": "Reaching a length limit does not force a plan upgrade; she can continue in a new chat or move the material into a Project."}, {"letter": "B", "text": "Every answer will be produced by a weaker model once the conversation has passed the context window limit.", "correct": false, "explanation": "The model in use does not change because a conversation is long. Approaching the limit affects available working memory, not model quality."}, {"letter": "C", "text": "Claude will delete the earlier messages from the chat history, leaving her nothing at all to check.", "correct": false, "explanation": "Condensing earlier messages does not erase them. The full chat history stays readable, so she can still go back and check."}, {"letter": "D", "text": "Detail from early in the chat may be condensed away, so her theme counts can quietly omit some feedback.", "correct": true, "explanation": "As a conversation approaches the context limit, earlier messages are condensed to make room. The summary still reads confidently, but it may no longer rest on every comment she pasted in."}], "correct": "D", "select": 1}, {"id": "a3-049", "domain": 3, "task_id": "3.2", "objective": "Differentiate between Claude model types (Haiku, Sonnet, Opus)", "situation": "A voice assistant application requires the lowest possible response latency among current Claude models so that spoken exchanges feel natural and immediate.", "question": "Which model tier has the fastest comparative latency?", "options": [{"letter": "A", "text": "Claude Haiku, since it is described as having the fastest comparative latency of the current model lineup", "correct": true, "explanation": "In Anthropic's model comparison, Haiku is listed with the fastest comparative latency among the current-generation models."}, {"letter": "B", "text": "Claude Opus, since higher intelligence models always respond with lower latency under normal load", "correct": false, "explanation": "Greater intelligence does not translate into lower latency; Opus is listed with moderate comparative latency, slower than both Sonnet and Haiku."}, {"letter": "C", "text": "Claude Sonnet, since its balance of speed and intelligence makes it faster than any other current tier", "correct": false, "explanation": "Sonnet is described as fast, but the comparison table ranks Haiku's comparative latency ahead of Sonnet's."}, {"letter": "D", "text": "All current Claude models share the same comparative latency, differing only in output quality", "correct": false, "explanation": "The models differ noticeably in comparative latency, not just in output quality, which is precisely why latency is one of the criteria for choosing among them."}], "correct": "A", "select": 1}, {"id": "a3-050", "domain": 3, "task_id": "3.2", "objective": "Differentiate between Claude model types (Haiku, Sonnet, Opus)", "situation": "A team is designing an agent that must autonomously produce very long outputs, such as an extensive generated report or a large batch of generated code, and wants the model tier best associated with handling the most demanding, large-scale generation tasks.", "question": "Which tier should they choose?", "options": [{"letter": "A", "text": "Any tier works equally well, since maximum output length does not differ meaningfully across current models", "correct": false, "explanation": "Model tiers differ in intelligence, latency, and cost for demanding generation tasks, so the choice of tier is not interchangeable for the most complex, large-scale work described here."}, {"letter": "B", "text": "Claude Sonnet, since it is the only tier that supports generating long-form output at all", "correct": false, "explanation": "Long-form output generation is supported across the current model lineup, not exclusively by Sonnet, so this option misstates the family's capabilities."}, {"letter": "C", "text": "Claude Haiku, since its lower cost per token always outweighs output-length considerations for any task", "correct": false, "explanation": "Lower cost per token is a real advantage of Haiku, but for the most demanding, large-scale generation tasks Anthropic points to the most capable tier rather than defaulting to the cheapest one."}, {"letter": "D", "text": "Claude Opus, since it is positioned for the largest-scale, most complex generation and engineering tasks", "correct": true, "explanation": "Opus is recommended for the largest-scale, most complex engineering and generation work, matching a scenario centered on demanding, large-scale autonomous output generation."}], "correct": "D", "select": 1}, {"id": "a3-051", "domain": 3, "task_id": "3.4", "objective": "Understand and manage context limitations and memory considerations (when to restart, summarize, or persist)", "situation": "A long strategy chat has drifted off track, and it keeps re-deriving background the team will still need next week. Which two actions are appropriate here?", "question": "Select two.", "options": [{"letter": "A", "text": "Start a fresh chat that restates the current goal and carries only the material still needed.", "correct": true, "explanation": "Restarting with a tight restatement clears accumulated drift and keeps only the context that is still relevant."}, {"letter": "B", "text": "Switch over to the lightest and fastest tier so that continuing this long conversation costs less.", "correct": false, "explanation": "A cheaper tier does not fix drift, and it does not persist the background the team will need again."}, {"letter": "C", "text": "Move the durable background into a Project, so later chats begin from it without redoing it.", "correct": true, "explanation": "Background needed again next week is durable context, and a Project is where durable context belongs."}, {"letter": "D", "text": "Carry on in the same chat and restate the goal more forcefully every time answers drift off.", "correct": false, "explanation": "Repeating the goal more forcefully in a drifting conversation rarely recovers it and adds to an already crowded context."}, {"letter": "E", "text": "Delete earlier messages from the chat so that older material stops shaping the newer replies.", "correct": false, "explanation": "Pruning messages is not how a drifting chat is repaired, and it does not make the background durable for next week."}], "correct": ["A", "C"], "select": 2}, {"id": "a3-052", "domain": 3, "task_id": "3.2", "objective": "Differentiate between Claude model types (Haiku, Sonnet, Opus)", "situation": "A team is deciding between Claude Haiku and Claude Opus for a new internal tool and wants a simple rule of thumb for when accuracy should be prioritized over cost and speed.", "question": "According to Anthropic's guidance, which situation should push the team toward starting with Opus instead of Haiku?", "options": [{"letter": "A", "text": "When rapid iteration is needed for prototyping new features without real user testing, start with Opus to ensure high accuracy from the start.", "correct": false, "explanation": "Anthropic actually recommends Haiku over Opus for rapid prototyping. Haiku's faster speed and lower cost allow teams to iterate quickly and test ideas without incurring high expenses or latency, making Opus an unnecessarily costly and slow choice at this stage."}, {"letter": "B", "text": "For high-volume repetitive tasks with tight budgets, Opus is preferred over Haiku for its lower cost per task and faster speed.", "correct": false, "explanation": "This statement inaccurately describes Opus. Anthropic guidance states that Haiku is the fastest and most cost-effective model, ideal for high-volume, budget-sensitive tasks. Opus, while more capable, is significantly more expensive and slower, making it unsuitable for such use cases."}, {"letter": "C", "text": "Complex reasoning where accuracy outweighs cost, such as scientific or advanced coding, calls for Opus.", "correct": true, "explanation": "According to Anthropic's official guidance, Opus models are the most capable and are recommended when accuracy and advanced reasoning are top priorities, even at higher cost. Use cases include complex scientific research, advanced coding, and high-stakes enterprise tasks where output must be reliable on the first pass. This rule of thumb directly aligns with documentation emphasizing Opus for 'complex reasoning where accuracy outweighs cost.'"}, {"letter": "D", "text": "To ensure fast response times for low-latency chatbots during peak traffic, choose Opus over Haiku due to its optimized inference.", "correct": false, "explanation": "Anthropic recommends Haiku for low-latency applications like peak-traffic chatbots because Haiku provides the fastest response times and lowest cost per request. Opus is designed for depth rather than speed, so using it here would lead to higher latency and cost without proportional benefit."}], "correct": "C", "select": 1}, {"id": "a3-053", "domain": 3, "task_id": "3.1", "objective": "Select appropriate Claude product features (Projects, research mode, chat, artifacts)", "situation": "An operations manager on a Pro plan is working within a single Project whose uploaded reference files are approaching the context limit. The manager wants Claude to keep giving high-quality answers grounded in that large body of documents without manually trimming the file set.", "question": "What underlying capability does the Project rely on to handle this?", "options": [{"letter": "A", "text": "Research mode replaces the uploaded files with live web search results, automatically expanding project context without manual trimming.", "correct": false, "explanation": "Research mode is a separate web-search capability; it does not tap into the project's uploaded document base or automatically replace files to expand context."}, {"letter": "B", "text": "An artifact compresses all uploaded files into a summarized attachment, expanding project capacity while preserving response quality.", "correct": false, "explanation": "Artifacts are for displaying generated content like code or documents; there is no artifact-based compression or summarization mechanism for uploaded reference files. Capacity expansion is handled by RAG."}, {"letter": "C", "text": "Retrieval Augmented Generation automatically expands project capacity by retrieving relevant chunks, preserving quality.", "correct": true, "explanation": "Projects use Retrieval Augmented Generation to automatically retrieve relevant chunks from uploaded files, expanding effective context capacity while maintaining answer quality without manual trimming."}, {"letter": "D", "text": "A downgrade to a smaller model uses a reduced context window per document, allowing more reference files within the same overall limit.", "correct": false, "explanation": "Downgrading to a smaller model changes generation cost and speed, not the way a project handles its uploaded knowledge base; it does not reduce per-document context usage to fit more files."}], "correct": "C", "select": 1}, {"id": "a3-054", "domain": 3, "task_id": "3.4", "objective": "Understand and manage context limitations and memory considerations (when to restart, summarize, or persist)", "situation": "A project lead needs Claude to work across a very large set of meeting notes inside one conversation on her paid plan.", "question": "Which choice best supports that?", "options": [{"letter": "A", "text": "Choose the fastest and cheapest model available, because lighter models are given more room for long inputs.", "correct": false, "explanation": "The fastest, cheapest tier is not the one with the largest window; speed and window size are separate trade-offs."}, {"letter": "B", "text": "Choose whichever model she used last, because the context window is identical across every available model.", "correct": false, "explanation": "Window size varies by model, so the choice of model directly changes how much a single chat can hold."}, {"letter": "C", "text": "Choose a previous-generation model, because older models were given more room before the newer releases.", "correct": false, "explanation": "The newest flagship models offer the largest windows; earlier models are more limited, not less."}, {"letter": "D", "text": "Choose one of the newest flagship models, which offer the largest context window for chatting on paid plans.", "correct": true, "explanation": "The newest flagship models provide a one-million-token window when chatting on any paid plan, which is the most room available for a single conversation."}], "correct": "D", "select": 1}, {"id": "a3-055", "domain": 3, "task_id": "3.3", "objective": "Align model selection with task requirements (cost, speed, quality)", "situation": "Two tasks draw almost identically on a team's allowance: notes from an internal brainstorm that nobody will act on directly, and a figure that will appear in a published statement. A colleague argues the two should therefore be treated the same way.", "question": "How should the team decide?", "options": [{"letter": "A", "text": "Treat them the same, since equal allowance draw means equal cost to the team and there is no clear financial reason to separate them.", "correct": false, "explanation": "Equal allowance draw only tells the team what each request costs to run. It says nothing about what a mistake costs, which is where the two tasks differ enormously."}, {"letter": "B", "text": "Let what an error would cost drive the decision, so the published figure earns more depth and checking than the internal notes do.", "correct": true, "explanation": "Consequence, not price, should drive the choice: a wrong figure in a published statement is hard to retract, while a weak brainstorm note is discarded, so the published work justifies more depth and scrutiny."}, {"letter": "C", "text": "Treat them the same, since a single rule applied to both tasks is easier for colleagues to remember and to follow consistently.", "correct": false, "explanation": "A single rule is simpler to follow but forces one of two errors: over-spending on throwaway notes, or under-serving a published figure. Simplicity is not worth either outcome."}, {"letter": "D", "text": "Let the person who happens to run each task decide, since they know their own deadline and how much allowance they have left.", "correct": false, "explanation": "Leaving it to whoever runs the task makes an important external output depend on one person's remaining allowance and time pressure, which is exactly when the published figure would be short-changed."}], "correct": "B", "select": 1}, {"id": "a3-056", "domain": 3, "task_id": "3.4", "objective": "Understand and manage context limitations and memory considerations (when to restart, summarize, or persist)", "situation": "A project lead pastes the same 40-page requirements document into every new chat instead of storing it anywhere.", "question": "What is the greatest concern with this approach?", "options": [{"letter": "A", "text": "Each re-paste spends both window space and usage allowance that storing the document once would save.", "correct": true, "explanation": "The document consumes a large share of the conversation's working memory every time, and the repeated processing draws on her usage budget for no added benefit."}, {"letter": "B", "text": "The document will be copied into other people's projects across her organisation without her consent.", "correct": false, "explanation": "Pasting into her own chat does not propagate the document to colleagues' workspaces."}, {"letter": "C", "text": "Claude will answer only from the closing pages of the document and silently ignore everything before.", "correct": false, "explanation": "Claude does not restrict itself to the final pages; the real exposure is cost and space, not truncation by position."}, {"letter": "D", "text": "Claude will refuse to accept a document of that size and will ask her to shorten it before continuing.", "correct": false, "explanation": "A document of that size is readable; the cost is in the space and allowance it consumes, not a refusal."}], "correct": "A", "select": 1}, {"id": "a3-057", "domain": 3, "task_id": "3.1", "objective": "Select appropriate Claude product features (Projects, research mode, chat, artifacts)", "situation": "A user needs to generate a report and then make direct edits to its content.", "question": "Which Claude feature is most appropriate?", "options": [{"letter": "A", "text": "Enable Research mode to refine the report with external sources.", "correct": false, "explanation": "Research mode is designed to gather and incorporate external sources, not to directly edit or refine report text. It does not include a built-in editor for modifying the report's content."}, {"letter": "B", "text": "Generate the report as an Artifact, then edit it in the artifact editor.", "correct": true, "explanation": "Artifacts provide a dedicated editing workspace for longer-form content like reports. Generating the report as an Artifact allows direct, structured modifications in the artifact editor, aligning perfectly with the user's need."}, {"letter": "C", "text": "Use Chat to output the report and then edit the raw text inline.", "correct": false, "explanation": "Chat can generate the report, but editing raw text inline in the chat interface is inefficient and lacks dedicated editing tools. This method is not suited for making structured, direct edits to a report."}, {"letter": "D", "text": "Use Projects to store the report and make edits in a separate file.", "correct": false, "explanation": "Projects are intended for organizing conversations and files, not for direct editing of reports. Storing a report and then editing it in a separate file is a manual workaround that lacks the integrated editing experience needed here."}], "correct": "B", "select": 1}, {"id": "a3-058", "domain": 3, "task_id": "3.3", "objective": "Align model selection with task requirements (cost, speed, quality)", "situation": "A team wants to decide whether a lighter tier is good enough for its weekly summary work.", "question": "What is the most reliable way to make that call?", "options": [{"letter": "A", "text": "Run real examples of their own summaries on both tiers and judge them against an agreed standard.", "correct": true, "explanation": "Adequacy is specific to the task and the bar, so only their own work compared against a stated standard answers the question."}, {"letter": "B", "text": "Read published comparisons of the tiers and adopt whichever one they describe as sufficient.", "correct": false, "explanation": "General descriptions cannot tell the team whether a lighter tier meets their particular standard on their own material."}, {"letter": "C", "text": "Adopt the lighter tier and revisit that decision if the team's allowance stops running out so early.", "correct": false, "explanation": "Allowance consumption measures cost, not whether the summaries still meet the required standard."}, {"letter": "D", "text": "Ask Claude to compare its own output on both tiers and report which version is the stronger.", "correct": false, "explanation": "A self-assessment is not independent evidence, and it has not been told what this team's summary must achieve."}], "correct": "A", "select": 1}, {"id": "a3-059", "domain": 3, "task_id": "3.3", "objective": "Align model selection with task requirements (cost, speed, quality)", "situation": "A team drafts a long monthly report in Claude and keeps exhausting its usage allowance before the month ends. Some sections are routine - recurring summaries and figures pulled from the same sources each time - while other sections carry the analysis the report is read for. The team already signs the report off against an agreed quality checklist, and it wants to cut usage by moving some of the drafting to a lighter Claude model tier without weakening the sections that matter.", "question": "Which sequence of steps should the team follow?", "options": [{"letter": "A", "text": "Run the routine sections on a lighter tier; decide which sections need human judgment; check the drafts against the agreed checklist; send back to the deeper tier the sections that fell short.", "correct": false, "explanation": "The routine sections are moved to the lighter tier before anyone has worked out which sections carry the judgment, so the split is guesswork about the content rather than a decision based on it, and the checklist pass that follows can only repair a division that was arbitrary to start with."}, {"letter": "B", "text": "Decide which sections need human judgment; run the routine sections on a lighter tier; check the drafts against the agreed checklist; send back to the deeper tier the sections that fell short.", "correct": true, "explanation": "Deciding which sections need human judgment comes first because it is the only step that reasons about the content rather than the tooling; the lighter tier is then applied only where quality is not at stake; and the checklist the team already signs the report off against is what tells them which moves to keep and which to send back to the deeper tier. Claude Academy's model-selection tutorial applies the same test - run the work on the lighter model, then judge whether the output met your bar."}, {"letter": "C", "text": "Decide which sections need human judgment; check the drafts against the agreed checklist; send back to the deeper tier the sections that fell short; run the routine sections on a lighter tier.", "correct": false, "explanation": "The drafts are checked and sections are sent back to the deeper tier before the lighter tier has been used at all, so the move to the lighter tier is the last thing done and nothing verifies its output - the one step the team cannot afford to leave unchecked given that quality is the stated constraint."}, {"letter": "D", "text": "Check the drafts against the agreed checklist; decide which sections need human judgment; run the routine sections on a lighter tier; send back to the deeper tier the sections that fell short.", "correct": false, "explanation": "The checklist pass happens before anything has been moved, so it measures the drafts the team already produces and says nothing about how the lighter tier performs; the final step then sends back sections that were never re-checked after the change."}], "correct": "B", "select": 1}, {"id": "a3-060", "domain": 3, "task_id": "3.2", "objective": "Differentiate between Claude model types (Haiku, Sonnet, Opus)", "situation": "A support operations lead must classify a large daily backlog of short inbound messages, and separately needs one deep analysis of a tangled contract dispute. Which two choices fit?", "question": "Select two.(Select 2)", "options": [{"letter": "A", "text": "Use whichever tier the team already has open, since the tier rarely changes what comes back.", "correct": false, "explanation": "Tier choice changes depth, speed and quota use, so treating it as incidental is how both of these jobs go wrong."}, {"letter": "B", "text": "Use the deepest-reasoning tier for the contract dispute, accepting the larger quota it uses.", "correct": true, "explanation": "A tangled one-off analysis is the case where deeper reasoning is worth the meaningfully larger quota it consumes."}, {"letter": "C", "text": "Use the fastest, cheapest tier for the high-volume classification, where throughput matters.", "correct": true, "explanation": "High-volume, low-complexity classification is exactly where the fastest and cheapest tier earns its place."}, {"letter": "D", "text": "Use the fastest, cheapest tier for both jobs, since the tiers differ mainly in reply speed.", "correct": false, "explanation": "The tiers differ in reasoning depth as well as speed, so the lightest tier is a poor fit for a tangled dispute."}, {"letter": "E", "text": "Use the deepest-reasoning tier for both of the jobs, since the strongest option is never bad.", "correct": false, "explanation": "Using the deepest tier for routine classification spends quota on work that does not need the extra reasoning depth."}], "correct": ["B", "C"], "select": 2}, {"id": "a3-061", "domain": 3, "task_id": "3.4", "objective": "Understand and manage context limitations and memory considerations (when to restart, summarize, or persist)", "situation": "A marketer's project knowledge has grown far past what a single conversation could hold, and she is unsure whether Claude can still draw on all of it.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Delete older documents until the knowledge base is small enough to fit inside a single conversation window.", "correct": false, "explanation": "Deleting useful material to fit an old constraint throws away knowledge that retrieval could still reach."}, {"letter": "B", "text": "Move the material into individual chats so that each single conversation carries only the documents it needs.", "correct": false, "explanation": "This reintroduces the repeated pasting that a project exists to avoid, and it drains usage allowance."}, {"letter": "C", "text": "Split the material across several smaller projects before she asks any further questions in that workspace.", "correct": false, "explanation": "Splitting fragments the knowledge base and creates maintenance work to solve a problem retrieval already handles."}, {"letter": "D", "text": "Ask questions in the project and check the answers, since retrieval over a large knowledge base is automatic.", "correct": true, "explanation": "When project knowledge approaches or exceeds the context limit, retrieval activates on its own with no setup, holding far more content. Testing it is the cheapest way to find out where she stands."}], "correct": "D", "select": 1}, {"id": "a3-062", "domain": 3, "task_id": "3.4", "objective": "Understand and manage context limitations and memory considerations (when to restart, summarize, or persist)", "situation": "A marketer's long-running chat has begun producing copy that ignores the campaign brief agreed weeks earlier.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Open a fresh chat; check whether the old chat had grown too long; paste the brief in; capture the decisions afterwards.", "correct": false, "explanation": "Restarting before capturing the decisions risks losing them, and diagnosing after the fact wastes the step."}, {"letter": "B", "text": "Capture the decisions; open a fresh chat; check whether the old chat had grown too long; paste the brief in afterwards.", "correct": false, "explanation": "Working in the new chat before the brief is in place produces exactly the drift she is trying to fix."}, {"letter": "C", "text": "Check whether the chat has simply grown too long; capture the brief and decisions; open a fresh chat; paste them in there.", "correct": true, "explanation": "Diagnosing the cause comes before acting, capturing the durable material protects it, and only then is restarting safe rather than destructive."}, {"letter": "D", "text": "Paste the brief in again; capture the decisions; check whether the chat has grown too long; open a fresh chat right at the end.", "correct": false, "explanation": "Re-pasting into an already overloaded conversation adds pressure to the window before anything has been diagnosed."}], "correct": "C", "select": 1}, {"id": "a3-063", "domain": 3, "task_id": "3.4", "objective": "Understand and manage context limitations and memory considerations (when to restart, summarize, or persist)", "situation": "A marketer is midway through a one-off brainstorm that will not be revisited, and the conversation is approaching its limit.", "question": "What is the most appropriate action?", "options": [{"letter": "A", "text": "Let the conversation continue and condense itself, since nothing in it needs to outlive this session.", "correct": true, "explanation": "When the output is disposable, allowing automatic context management to make room is the lowest-effort choice; there is no durable knowledge worth persisting."}, {"letter": "B", "text": "Build a Project and move the whole brainstorm into project knowledge before writing anything more.", "correct": false, "explanation": "A project is for context that recurs. Setting one up for work that will never be revisited is unnecessary overhead."}, {"letter": "C", "text": "Stop the brainstorm now and restart it in a new chat so the ideas come from a completely clean window.", "correct": false, "explanation": "Restarting discards the ideas generated so far and forces her to regenerate them at further cost."}, {"letter": "D", "text": "Split the brainstorm across four separate chats so that no single conversation ever nears its limit.", "correct": false, "explanation": "Fragmenting the session means no chat holds the full set of ideas, which undermines the brainstorm itself."}], "correct": "A", "select": 1}, {"id": "a3-064", "domain": 3, "task_id": "3.2", "objective": "Differentiate between Claude model types (Haiku, Sonnet, Opus)", "situation": "A company is scaling a content-creation and data-analysis product used by thousands of business customers. They need strong general capability across code generation, analysis, and visual understanding, without committing to the most expensive tier.", "question": "Which model tier matches this description in Anthropic's model selection matrix?", "options": [{"letter": "A", "text": "Use a mix of unrelated third-party models because Claude tiers fail to include data analysis and visual understanding.", "correct": false, "explanation": "Claude models, especially Sonnet, include robust data analysis and visual understanding features, so the described product's needs are well covered without resorting to external models."}, {"letter": "B", "text": "Choose Claude Sonnet, as it offers frontier intelligence for coding, analysis, and visual understanding at scale.", "correct": true, "explanation": "Claude Sonnet offers frontier intelligence for coding, analysis, and visual understanding at scale, making it the right tier for a product that needs strong, general capabilities without the highest cost."}, {"letter": "C", "text": "Select Claude Haiku because code generation and analysis tasks should default to the cheapest model tier available.", "correct": false, "explanation": "Defaulting to the cheapest tier ignores that this product requires robust performance across demanding tasks; Haiku is optimized for cost-efficient, simpler workloads, not the breadth of capabilities needed here."}, {"letter": "D", "text": "Opt for Claude Opus because content platforms need peak reasoning and visual analysis, requiring the top tier.", "correct": false, "explanation": "Opus is designed for the most complex, high-autonomy, or accuracy-critical tasks; a scaling content and analysis product fits Sonnet's balanced profile better than the top, most expensive tier."}], "correct": "B", "select": 1}, {"id": "a3-065", "domain": 3, "task_id": "3.3", "objective": "Align model selection with task requirements (cost, speed, quality)", "situation": "An operations lead has two hundred short, standard customer replies to prepare today and is worried about her weekly allowance.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Ask Claude to estimate how much of her allowance the batch is likely to consume in total.", "correct": false, "explanation": "An estimate of consumption does not change what the batch costs or tell her which parts need care."}, {"letter": "B", "text": "Run the whole batch on the deepest-reasoning model, so no reply needs to be redone afterwards.", "correct": false, "explanation": "Applying the most expensive tier to two hundred routine items is the fastest way to exhaust the allowance she is worried about."}, {"letter": "C", "text": "Separate the replies that genuinely need judgment from the routine bulk of the batch.", "correct": true, "explanation": "Once the small number of awkward cases is identified, the routine bulk can go to the fastest and cheapest tier and the allowance is reserved for the cases that need it."}, {"letter": "D", "text": "Split the batch across several days, so that no single day uses too much of her allowance.", "correct": false, "explanation": "Spreading the work delays the replies without reducing the total allowance the batch consumes."}], "correct": "C", "select": 1}, {"id": "a3-066", "domain": 3, "task_id": "3.4", "objective": "Understand and manage context limitations and memory considerations (when to restart, summarize, or persist)", "situation": "A teacher wants Claude to answer student questions from a large and steadily growing set of course materials, across many separate chats.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Add the course materials to each chat; create a Project; ask questions in that project; keep the instructions brief.", "correct": false, "explanation": "Pasting the materials into individual chats is the very repetition a project is meant to remove."}, {"letter": "B", "text": "Ask questions in a plain chat; keep the instructions brief; create a Project; add the course materials as project knowledge.", "correct": false, "explanation": "The questions are asked before any of the course materials exist as project knowledge, so those first answers have nothing to draw on."}, {"letter": "C", "text": "Keep the instructions brief; create a Project; ask questions in that project; add the course materials as project knowledge.", "correct": false, "explanation": "Asking before the materials are uploaded means the early answers have nothing to draw on."}, {"letter": "D", "text": "Create a Project; add the course materials as project knowledge; keep the instructions brief; ask questions in that project.", "correct": true, "explanation": "The project has to exist before materials can be attached, brief instructions leave more room for the materials themselves, and only chats started inside the project can draw on them."}], "correct": "D", "select": 1}, {"id": "a3-067", "domain": 3, "task_id": "3.2", "objective": "Differentiate between Claude model types (Haiku, Sonnet, Opus)", "situation": "A facilitator wants Claude to produce short prompts and summaries live during workshops, with participants waiting in the room. She standardises on the deepest-reasoning tier so that the material is always as good as it can be.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The facilitator becomes reliant on having Claude available, so a session held somewhere without connectivity is harder to run.", "correct": false, "explanation": "Dependence on connectivity is a sensible contingency to plan for, but it applies equally whichever tier she picks, so it is not the risk created by standardising on the deepest one."}, {"letter": "B", "text": "The material becomes more detailed than a live workshop can absorb, so participants receive more text than they can read aloud.", "correct": false, "explanation": "Over-long output is a genuine annoyance, but it is fixed by asking for something shorter. It is not caused by the tier choice, and the facilitator can control length in the request itself."}, {"letter": "C", "text": "The workshop material varies in tone between sessions, so participants attending more than one workshop notice the difference.", "correct": false, "explanation": "Some variation between sessions is expected with any tier and matters little for workshop prompts. It is a consistency preference rather than something that stops the session working."}, {"letter": "D", "text": "The answers take longer to come back than a room of waiting participants can tolerate, so the help is unusable in the moment.", "correct": true, "explanation": "Depth is bought with time, and in a live setting responsiveness is the binding constraint. Choosing the deepest tier here sacrifices the very quality, immediacy, that made Claude useful in the room."}], "correct": "D", "select": 1}, {"id": "a3-068", "domain": 3, "task_id": "3.1", "objective": "Select appropriate Claude product features (Projects, research mode, chat, artifacts)", "situation": "A developer asks Claude to build a small interactive React component that visualizes sales data with adjustable filters, and expects to iterate on the code across several follow-up messages.", "question": "Which feature will display this output in a way that supports iterative editing and version history?", "options": [{"letter": "A", "text": "A Project's knowledge base, which stores the component as reference material for future chats", "correct": false, "explanation": "Project knowledge is meant for background reference documents the user uploads, not for displaying and iteratively editing generated code with version tracking."}, {"letter": "B", "text": "Research mode, which searches the web for similar existing React components", "correct": false, "explanation": "Research mode is for gathering external information through web searches, not for rendering or iterating on generated code."}, {"letter": "C", "text": "The main chat transcript, where the code is printed inline with each response", "correct": false, "explanation": "Printing code inline in the main transcript does not provide the dedicated side-panel view or version history that iterative component development benefits from."}, {"letter": "D", "text": "The artifact panel, which shows the component separately and tracks versions as it is revised", "correct": true, "explanation": "Artifacts open substantial, self-contained content like a React component in a dedicated side panel and provide a version selector, letting the developer iterate and switch between revisions."}], "correct": "D", "select": 1}, {"id": "a3-069", "domain": 3, "task_id": "3.4", "objective": "Understand and manage context limitations and memory considerations (when to restart, summarize, or persist)", "situation": "A marketer runs a new campaign chat each week and re-pastes the same brand guidelines and tone-of-voice document into every one. She wants to stop repeating herself.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Add project instructions; run each campaign chat inside it; create a Project; upload the guidelines as project knowledge.", "correct": false, "explanation": "Instructions and chats are both scoped to a project, so neither can come before the project is created."}, {"letter": "B", "text": "Upload the guidelines as project knowledge; run each campaign chat inside it; create a Project; add project instructions.", "correct": false, "explanation": "Knowledge is uploaded into a project, so the upload cannot precede creating one, and the chats would run without the instructions."}, {"letter": "C", "text": "Create a Project; upload the guidelines as project knowledge; add project instructions; run each campaign chat inside it.", "correct": true, "explanation": "The workspace has to exist before knowledge and instructions can be attached to it, and the chats only inherit that context once they are started inside the finished project."}, {"letter": "D", "text": "Run each campaign chat first; add project instructions; create a Project; upload the guidelines as project knowledge.", "correct": false, "explanation": "The campaign chats would run before the guidelines and instructions are in place, so the early work gets none of the benefit, and instructions cannot be added before the project exists."}], "correct": "C", "select": 1}, {"id": "a3-070", "domain": 3, "task_id": "3.3", "objective": "Align model selection with task requirements (cost, speed, quality)", "situation": "After a cost review, a team switches every Claude task to the fastest and cheapest tier.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "Every task will now be completed rather faster than the team is able to review the output properly.", "correct": false, "explanation": "Faster turnaround is not the danger, and review capacity is a separate scheduling matter."}, {"letter": "B", "text": "Responses will be much shorter, so colleagues will have to ask more follow-up questions.", "correct": false, "explanation": "Length is set by what is asked for rather than by the tier, and it is not the substantive exposure."}, {"letter": "C", "text": "The small number of genuinely hard analyses will thin out with no agreed standard to catch it.", "correct": true, "explanation": "A blanket switch removes reasoning depth from the work that needed it, and without a defined quality bar the degradation shows up as slightly weaker answers nobody flags."}, {"letter": "D", "text": "The team will be unable to move any of the tasks back to a deeper tier at a later stage.", "correct": false, "explanation": "Nothing prevents moving back; the tier can be changed whenever the team decides to."}], "correct": "C", "select": 1}, {"id": "a3-071", "domain": 3, "task_id": "3.4", "objective": "Understand and manage context limitations and memory considerations (when to restart, summarize, or persist)", "situation": "A project lead notices that three colleagues each re-explain the same programme background to Claude every week in their own chats.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Ask each colleague to summarise their chats weekly and circulate those summaries by email.", "correct": false, "explanation": "Circulating summaries adds manual work each week without giving Claude the background in any of the chats."}, {"letter": "B", "text": "Ask each colleague to keep one very long chat open so their own context accumulates over time.", "correct": false, "explanation": "Very long personal chats condense their own history and draw more heavily on usage allowance, and the context stays siloed."}, {"letter": "C", "text": "Ask Claude to remember the programme background so nobody has to supply it again in future.", "correct": false, "explanation": "A chat does not carry context to colleagues' conversations; shared background has to be persisted deliberately."}, {"letter": "D", "text": "Gather the background the team keeps repeating into one agreed set of reference documents.", "correct": true, "explanation": "Before choosing where to put shared context, the team needs to agree what that context actually is. A single reference set is what any later Project or hand-off will be built from."}], "correct": "D", "select": 1}, {"id": "a3-072", "domain": 3, "task_id": "3.2", "objective": "Differentiate between Claude model types (Haiku, Sonnet, Opus)", "situation": "A new team member asks how the Claude model tiers actually differ from one another in everyday use. Which two statements are accurate?", "question": "Select two.(Select 2)", "options": [{"letter": "A", "text": "The deepest tier suits genuinely hard problems, and it consumes meaningfully more allowance.", "correct": true, "explanation": "The deepest tier buys reasoning depth for harder problems and costs meaningfully more of the usage allowance."}, {"letter": "B", "text": "The tiers differ only in writing style, so any of them may be used for any task at any time.", "correct": false, "explanation": "The difference is reasoning depth, speed and cost, not writing style, and the tiers are not interchangeable."}, {"letter": "C", "text": "The lightest tier gives answers of the very same depth, only more slowly than the other two.", "correct": false, "explanation": "The lightest tier is the fastest of the three, and it trades some reasoning depth for that speed and low cost."}, {"letter": "D", "text": "The mid tier is the fast, capable default, and it is the right pick for most everyday work.", "correct": true, "explanation": "The middle tier is positioned as the fast, capable default that handles most knowledge work well."}, {"letter": "E", "text": "Moving up one tier is the standard fix whenever a prompt comes back with a vague or shallow answer.", "correct": false, "explanation": "A vague answer usually signals an under-specified prompt, and a deeper tier cannot supply context that was never given."}], "correct": ["A", "D"], "select": 2}, {"id": "a3-073", "domain": 3, "task_id": "3.3", "objective": "Align model selection with task requirements (cost, speed, quality)", "situation": "An operations team must summarise several thousand short product blurbs every night, and the summaries get only a light human check. Which two factors should drive the tier choice?", "question": "Select two.", "options": [{"letter": "A", "text": "How long each finished summary turns out to be, because length is what separates the tiers.", "correct": false, "explanation": "Output length is set by the request, not by the tier, so it tells you nothing about which tier to pick."}, {"letter": "B", "text": "The reasoning depth the hardest problem anywhere in the company would need, as a safety margin.", "correct": false, "explanation": "Sizing for an unrelated hardest-case task overspends depth and quota on work that is simple and lightly reviewed."}, {"letter": "C", "text": "The allowance the run consumes, because this same job repeats every single night of the week.", "correct": true, "explanation": "A nightly repeat multiplies the allowance the job consumes, so recurring cost is a legitimate driver of the decision."}, {"letter": "D", "text": "Whether the deepest tier appears on their plan at all, since availability settles the whole matter.", "correct": false, "explanation": "Availability is not the question; suitability for the task on cost, speed and quality is."}, {"letter": "E", "text": "The throughput needed each night, because the volume makes speed per item a real constraint.", "correct": true, "explanation": "Thousands of items a night makes speed per item a genuine constraint, which is a core input to tier choice."}], "correct": ["C", "E"], "select": 2}, {"id": "a3-074", "domain": 3, "task_id": "3.3", "objective": "Align model selection with task requirements (cost, speed, quality)", "situation": "An analyst must deliver a carefully reasoned briefing by the end of the day and has limited allowance left.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Check it against the brief; cut the scope to what the briefing needs; run that scope on the deepest tier; expand it if time allows.", "correct": false, "explanation": "Checking comes before anything has been produced, which makes the first step impossible."}, {"letter": "B", "text": "Cut the scope to what the briefing actually needs; run that scope on the deepest tier; check it against the brief; expand it if time allows.", "correct": true, "explanation": "Narrowing scope is what makes the deeper tier affordable within the remaining allowance, and the check confirms the reduced briefing still meets the requirement."}, {"letter": "C", "text": "Run the full scope on the deepest tier; cut the scope to what the briefing needs; check it against the brief; expand it if time allows.", "correct": false, "explanation": "Running the full scope on the most expensive tier first spends the constrained allowance before any trimming happens."}, {"letter": "D", "text": "Cut the scope to what the briefing actually needs; expand it if time allows; run that scope on the deepest tier; check it against the brief.", "correct": false, "explanation": "Expanding the scope before it has been produced undoes the trimming that made the deeper tier affordable."}], "correct": "B", "select": 1}, {"id": "a3-075", "domain": 3, "task_id": "3.3", "objective": "Align model selection with task requirements (cost, speed, quality)", "situation": "A colleague proposes running every request on two different tiers and comparing the two answers each time.", "question": "What is the most reasonable assessment?", "options": [{"letter": "A", "text": "It is wasteful, because the two tiers will always return answers that are effectively identical.", "correct": false, "explanation": "They genuinely differ on demanding work, so the objection is cost rather than a lack of difference."}, {"letter": "B", "text": "It is sensible, because using two tiers spreads consumption across two separate allowances.", "correct": false, "explanation": "Usage is measured across all of a person's activity, so nothing is spread by using two tiers."}, {"letter": "C", "text": "It is wasteful, since it doubles the allowance spent on a comparison worth doing only once.", "correct": true, "explanation": "Comparing tiers is worth doing on a representative sample to settle the choice; repeating it on every request pays twice for a decision already made."}, {"letter": "D", "text": "It is sensible, since comparing two answers each time is the surest way to catch any error.", "correct": false, "explanation": "Two answers that agree can both be wrong, so this buys far less assurance than it costs."}], "correct": "C", "select": 1}, {"id": "a3-076", "domain": 3, "task_id": "3.3", "objective": "Align model selection with task requirements (cost, speed, quality)", "situation": "A newsletter team produces about thirty short social posts each week and a person edits every one before it goes out.", "question": "Which tier fits this work best?", "options": [{"letter": "A", "text": "A different tier for each post, chosen according to which social platform it is destined for.", "correct": false, "explanation": "The platform does not change the reasoning the task requires, so varying tier by channel is arbitrary."}, {"letter": "B", "text": "The fastest and cheapest tier, since the posts are short, routine and edited by a person after.", "correct": true, "explanation": "High volume, low complexity and a human editing step is exactly the profile the fastest tier is for; reasoning depth would add little to a two-line post."}, {"letter": "C", "text": "The deepest-reasoning tier, because published content should always use the very strongest option.", "correct": false, "explanation": "The posts are short and human-edited, so the extra depth changes little while consuming much more allowance."}, {"letter": "D", "text": "The default middle tier, because short marketing copy is too nuanced for any lighter option.", "correct": false, "explanation": "It would work, but it spends more allowance than thirty short, edited posts a week justify."}], "correct": "B", "select": 1}, {"id": "a3-077", "domain": 3, "task_id": "3.1", "objective": "Select appropriate Claude product features (Projects, research mode, chat, artifacts)", "situation": "A team moves its live pricing sheet into a project's knowledge so every chat in the project can use it, and then changes its prices two months later.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The pricing sheet will be counted twice, once as knowledge and once in each conversation.", "correct": false, "explanation": "Nothing is double-counted in that way, and this does not describe a real failure mode."}, {"letter": "B", "text": "Every chat in that project will keep quoting the superseded prices with complete confidence.", "correct": true, "explanation": "Making knowledge durable also makes staleness durable. Until the sheet is replaced, the project answers confidently from a document that no longer reflects reality."}, {"letter": "C", "text": "Colleagues outside the project will be unable to see the pricing sheet in their own chats.", "correct": false, "explanation": "That is the expected scope of a project rather than a risk created by the price change."}, {"letter": "D", "text": "The project will refuse any further uploads until the outdated pricing sheet has been removed.", "correct": false, "explanation": "There is no such block, and framing this as an upload problem misses where the damage actually occurs."}], "correct": "B", "select": 1}, {"id": "a3-078", "domain": 3, "task_id": "3.4", "objective": "Understand and manage context limitations and memory considerations (when to restart, summarize, or persist)", "situation": "An operations manager has finished a long planning chat and expects the team to revisit the same topic every month.", "question": "What is the best way to keep that context available?", "options": [{"letter": "A", "text": "Rely on Claude recalling the earlier planning chat automatically whenever she opens a new conversation.", "correct": false, "explanation": "Automatic recall is not a dependable store for decisions the team must rely on months later; durable context should be persisted into project knowledge deliberately."}, {"letter": "B", "text": "Keep reopening the same long chat every month so that all of the history stays inside one conversation.", "correct": false, "explanation": "A conversation that grows for a year will keep condensing its own earlier content and will draw heavily on usage allowance."}, {"letter": "C", "text": "Copy the whole transcript into the first message of every monthly chat so that nothing at all is ever left out.", "correct": false, "explanation": "The transcript is mostly working discussion rather than durable fact, and re-pasting it each month wastes window space and allowance."}, {"letter": "D", "text": "Save the agreed decisions into a Project as knowledge, then start each monthly review as a fresh chat there.", "correct": true, "explanation": "Persisting the durable output into project knowledge keeps the context available to every future chat without carrying months of transcript inside one conversation."}], "correct": "D", "select": 1}, {"id": "a3-079", "domain": 3, "task_id": "3.4", "objective": "Understand and manage context limitations and memory considerations (when to restart, summarize, or persist)", "situation": "A teacher reaches the length limit in a chat while working through a stack of graded feedback.", "question": "What is the appropriate response?", "options": [{"letter": "A", "text": "Start a new chat and carry forward only the material and decisions the remaining work actually needs.", "correct": true, "explanation": "The length limit applies to one conversation, so a fresh chat clears it. Bringing only what is still needed keeps the new conversation lean."}, {"letter": "B", "text": "Buy additional usage credits, which will raise the context window available inside that conversation.", "correct": false, "explanation": "Credits extend how much she can interact with Claude. The context window is fixed by the model and is not enlarged by buying allowance."}, {"letter": "C", "text": "Wait several hours for the length limit to reset, then carry on in the same conversation thread.", "correct": false, "explanation": "Length limits do not reset on a timer; that is how usage limits behave, and the two are easily confused."}, {"letter": "D", "text": "Switch to a shorter reply style, which will free up the window space the earlier messages took.", "correct": false, "explanation": "Shorter future replies slow the fill rate slightly but cannot recover space already occupied by earlier messages."}], "correct": "A", "select": 1}, {"id": "a3-080", "domain": 3, "task_id": "3.3", "objective": "Align model selection with task requirements (cost, speed, quality)", "situation": "An analyst's first draft from Claude is vague, and she wonders whether she should move up to a deeper-reasoning tier to fix it. Which two responses are sound?", "question": "Select two.", "options": [{"letter": "A", "text": "Escalate to the deepest tier right away, since depth is the usual cause of a weak first draft.", "correct": false, "explanation": "Reaching for the deepest tier as a reflex spends quota without fixing the under-specified request underneath."}, {"letter": "B", "text": "Stay on the lightest tier no matter what, because the tier has little effect on answer quality.", "correct": false, "explanation": "The tiers do differ in reasoning depth, so refusing to escalate is as unexamined as escalating by default."}, {"letter": "C", "text": "Check first whether the request was under-specified, since a deeper tier supplies no context.", "correct": true, "explanation": "Vagueness most often comes from a thin prompt, and no tier can invent the audience, format or source material you left out."}, {"letter": "D", "text": "Escalate the tier only when the work truly needs more reasoning depth than it is now getting.", "correct": true, "explanation": "Escalating is the right move when reasoning depth is genuinely the binding constraint, and only then."}, {"letter": "E", "text": "Split the same request across three chats on three tiers and merge whichever parts look best.", "correct": false, "explanation": "Running the same weak prompt three times multiplies the cost and still leaves the original specification problem in place."}], "correct": ["C", "D"], "select": 2}, {"id": "a3-081", "domain": 3, "task_id": "3.2", "objective": "Differentiate between Claude model types (Haiku, Sonnet, Opus)", "situation": "A marketing manager's answers keep coming back shallow, and a colleague suggests moving straight to the deepest-reasoning model.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Check whether the prompt states the depth needed; improve the prompt and retry; move to the deepest-reasoning model; keep the lighter model for routine parts.", "correct": true, "explanation": "A shallow answer is often an under-specified request, so the prompt is checked first, and reasoning depth is escalated only once that has been ruled out."}, {"letter": "B", "text": "Move to the deepest-reasoning model; check whether the prompt states the depth needed; improve the prompt; keep the lighter model for routine parts.", "correct": false, "explanation": "Changing tier before looking at the prompt spends more allowance on what may be an instruction problem rather than a capability one."}, {"letter": "C", "text": "Improve the prompt and retry; move to the deepest-reasoning model; check whether the prompt states the depth needed; keep the lighter model for routine parts.", "correct": false, "explanation": "The prompt is revised before anyone has looked at what it actually says, and the diagnosis arrives after the tier has already been changed."}, {"letter": "D", "text": "Keep the lighter model for routine parts; move to the deepest-reasoning model; improve the prompt and retry; check whether the prompt states the depth needed.", "correct": false, "explanation": "Every diagnostic step happens after both model changes, so the tier decisions are taken with no evidence behind them."}], "correct": "A", "select": 1}, {"id": "a3-082", "domain": 3, "task_id": "3.4", "objective": "Understand and manage context limitations and memory considerations (when to restart, summarize, or persist)", "situation": "An operations manager finds she hits her usage limit before lunch nearly every working day and is frustrated by the interruption.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Upgrade immediately to the highest plan tier available so the daily interruptions simply stop happening.", "correct": false, "explanation": "An upgrade may be justified later, but buying more allowance before understanding what consumes it is guesswork."}, {"letter": "B", "text": "Review the usage view in her settings to see what recent sessions actually consumed, before changing anything.", "correct": true, "explanation": "The fix depends on the cause, and the usage view shows how much of her session and weekly allowance has gone. Diagnosing first avoids paying for a change that would not help."}, {"letter": "C", "text": "Switch every task to the deepest-reasoning model so that fewer follow-up messages are needed to finish.", "correct": false, "explanation": "The deepest-reasoning tier consumes meaningfully more allowance per message, so this is likely to make matters worse."}, {"letter": "D", "text": "Move all of her work into brand-new chats each morning so that every conversation starts with an empty window.", "correct": false, "explanation": "A fresh window addresses a length limit, not a usage limit, which follows her across all conversations."}], "correct": "B", "select": 1}, {"id": "a3-083", "domain": 3, "task_id": "3.3", "objective": "Align model selection with task requirements (cost, speed, quality)", "situation": "A finance team runs an overnight batch of several thousand transaction narratives that must be finished before the six o'clock reporting cut-off. They move the batch to the deepest-reasoning tier because they want the best possible quality.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The team grows used to a higher standard of narrative and finds the earlier output unacceptable if they ever have to revert.", "correct": false, "explanation": "Raised expectations are a soft consequence and only matter if the team later reverts. They have no bearing on whether tonight's batch clears the six o'clock deadline."}, {"letter": "B", "text": "The narratives come back written in a more elaborate style than the reporting template was originally designed to accommodate.", "correct": false, "explanation": "An elaborate style is a formatting mismatch that can be corrected by asking for a tighter format. It does not threaten the report itself in the way a missed cut-off does."}, {"letter": "C", "text": "The batch does not finish inside the overnight window, so the morning report goes out incomplete or does not go out at all.", "correct": true, "explanation": "At this volume the binding constraint is throughput against a fixed deadline. Depth that cannot complete in the window fails the task outright, which is worse than slightly plainer narratives delivered on time."}, {"letter": "D", "text": "The batch produces more detail per transaction than the finance team has time to read through before the reporting cut-off.", "correct": false, "explanation": "Excess detail costs reading time, but the narratives are generated per transaction and are not all read individually. The scheduling failure is the far more serious outcome here."}], "correct": "C", "select": 1}, {"id": "a3-084", "domain": 3, "task_id": "3.3", "objective": "Align model selection with task requirements (cost, speed, quality)", "situation": "An education lead is choosing a model tier to help draft feedback that will contribute to students' marks.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Choose the deepest tier, since anything involving students should use the strongest option.", "correct": false, "explanation": "It may well be right, but choosing it before the quality bar is stated is a guess rather than a judgment."}, {"letter": "B", "text": "Establish how the feedback will be used and what an error in it would cost a student.", "correct": true, "explanation": "The quality bar comes from the consequence of the output. Once she knows an error affects a mark, the tier decision follows from that bar rather than from cost alone."}, {"letter": "C", "text": "Compare the tiers on a sample and pick whichever produces the longest and most detailed drafts.", "correct": false, "explanation": "Length is not the standard, and choosing on it rewards verbosity rather than accuracy."}, {"letter": "D", "text": "Choose the fastest tier, since drafting feedback for a class is a high-volume repetitive task.", "correct": false, "explanation": "Volume is only half the picture, and choosing on throughput ignores what an error here would cost."}], "correct": "B", "select": 1}, {"id": "a3-085", "domain": 3, "task_id": "3.2", "objective": "Differentiate between Claude model types (Haiku, Sonnet, Opus)", "situation": "A startup has a new application idea and wants to minimize development cost and iteration time before deciding whether a more capable model is truly needed.", "question": "Following Anthropic's recommended approach for this situation, which model should they implement first?", "options": [{"letter": "A", "text": "Claude Sonnet, since it should always be the first model tried regardless of budget concerns", "correct": false, "explanation": "Anthropic frames two valid starting approaches based on the use case; Sonnet being the universal first choice regardless of budget priorities is not part of that guidance."}, {"letter": "B", "text": "Whichever model the startup's investors specify in the term sheet", "correct": false, "explanation": "Model selection should follow the technical criteria of capability, speed, and cost fit for the application, not external non-technical instructions unrelated to those criteria."}, {"letter": "C", "text": "Claude Opus, then downgrade later once prompts have been optimized for cost", "correct": false, "explanation": "Starting with Opus and optimizing down over time is the recommended path for complex reasoning tasks where accuracy outweighs cost, not for a startup explicitly prioritizing low initial cost and fast iteration."}, {"letter": "D", "text": "Claude Haiku, then upgrade only if testing reveals a specific capability gap", "correct": true, "explanation": "Anthropic recommends starting with a fast, cost-effective model like Haiku, testing thoroughly, and upgrading only if performance reveals a specific capability gap, which fits a startup minimizing cost and iteration time."}], "correct": "D", "select": 1}, {"id": "a3-086", "domain": 3, "task_id": "3.2", "objective": "Differentiate between Claude model types (Haiku, Sonnet, Opus)", "situation": "A new team member asks how the Claude model tiers actually differ from one another in everyday use. Which two statements are accurate?", "question": "Select two.", "options": [{"letter": "A", "text": "The tiers differ only in writing style, so any of them may be used for any task at any time.", "correct": false, "explanation": "The difference is reasoning depth, speed and cost, not writing style, and the tiers are not interchangeable."}, {"letter": "B", "text": "The deepest tier suits genuinely hard problems, and it consumes meaningfully more allowance.", "correct": true, "explanation": "The deepest tier buys reasoning depth for harder problems and costs meaningfully more of the usage allowance."}, {"letter": "C", "text": "Moving up one tier is the standard fix whenever a prompt comes back with a vague or shallow answer.", "correct": false, "explanation": "A vague answer usually signals an under-specified prompt, and a deeper tier cannot supply context that was never given."}, {"letter": "D", "text": "The mid tier is the fast, capable default, and it is the right pick for most everyday work.", "correct": true, "explanation": "The middle tier is positioned as the fast, capable default that handles most knowledge work well."}, {"letter": "E", "text": "The lightest tier gives answers of the very same depth, only more slowly than the other two.", "correct": false, "explanation": "The lightest tier is the fastest of the three, and it trades some reasoning depth for that speed and low cost."}], "correct": ["B", "D"], "select": 2}, {"id": "a3-087", "domain": 3, "task_id": "3.4", "objective": "Understand and manage context limitations and memory considerations (when to restart, summarize, or persist)", "situation": "A communications lead is drafting an announcement in a chat that is close to its length limit, and the announcement is due in an hour.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Upgrade to a higher plan tier straight away so that the conversation gains a larger context window.", "correct": false, "explanation": "The context window is fixed by the model in use, so buying more allowance does not enlarge the window inside that chat."}, {"letter": "B", "text": "Delete the earliest messages in the conversation so that the newer content has more room to work with.", "correct": false, "explanation": "Removing earlier messages by hand is slow, destroys the record, and is not how room is reclaimed in a chat."}, {"letter": "C", "text": "Switch the conversation to a faster, cheaper model so the remaining exchanges take up less of the window.", "correct": false, "explanation": "Lighter models generally come with smaller context windows, so this shrinks the room she has left rather than extending it."}, {"letter": "D", "text": "Capture the current draft and the outstanding decisions outside the chat before she does anything else.", "correct": true, "explanation": "With a deadline close, the priority is making the work durable. Once the draft and open points are safely out of the conversation, restarting or continuing becomes a low-risk choice."}], "correct": "D", "select": 1}, {"id": "a3-088", "domain": 3, "task_id": "3.2", "objective": "Differentiate between Claude model types (Haiku, Sonnet, Opus)", "situation": "A new joiner asks how the Claude model tiers differ from the point of view of someone using the Claude app.", "question": "Which description is accurate?", "options": [{"letter": "A", "text": "Sonnet reasons most deeply, Opus is the everyday default option, and Haiku handles only images.", "correct": false, "explanation": "Sonnet is the capable default rather than the deepest tier, and Haiku is a general fast model, not an image-only one."}, {"letter": "B", "text": "Haiku gives the deepest reasoning, Sonnet is the fastest option, and Opus suits routine bulk work.", "correct": false, "explanation": "This inverts the tiers. Haiku is the fastest and cheapest rather than the deepest-reasoning option."}, {"letter": "C", "text": "Opus reasons most deeply and uses more allowance, Sonnet is the capable default, Haiku is fastest.", "correct": true, "explanation": "As an end user experiences them, the tiers trade reasoning depth against speed and how much of the usage allowance each message consumes."}, {"letter": "D", "text": "The tiers produce identical answers and differ only in how quickly those answers come back to you.", "correct": false, "explanation": "Speed is one axis, but reasoning depth genuinely differs, which is why the choice matters on hard work."}], "correct": "C", "select": 1}, {"id": "a3-089", "domain": 3, "task_id": "3.2", "objective": "Differentiate between Claude model types (Haiku, Sonnet, Opus)", "situation": "A research lab needs help with a scientific analysis task involving nuanced multi-step reasoning, where correctness matters far more than response time or token cost.", "question": "Which model tier does Anthropic's guidance point to?", "options": [{"letter": "A", "text": "Select Claude Opus, because scientific and mathematical tasks where accuracy outweighs cost are core use cases.", "correct": true, "explanation": "Anthropic explicitly recommends Claude Opus for complex reasoning, scientific, and mathematical tasks where accuracy outweighs cost and speed, exactly matching this scenario."}, {"letter": "B", "text": "Select Claude Sonnet, because it is the best general-purpose choice regardless of how much accuracy is prioritized.", "correct": false, "explanation": "Sonnet is a strong general-purpose model, but when nuanced, multi-step reasoning and accuracy are paramount, Anthropic points to the most capable tier (Opus) rather than a general-purpose default."}, {"letter": "C", "text": "Select Claude Haiku, because near-frontier reasoning at the lowest cost is adequate for standard research tasks.", "correct": false, "explanation": "Haiku is optimized for cost and latency sensitivity; Anthropic's guidance reserves it for tasks where speed or economy is the priority, not nuanced scientific work demanding maximum correctness."}, {"letter": "D", "text": "Select Claude Opus with effort set to its lowest level, because minimizing effort improves reasoning accuracy.", "correct": false, "explanation": "Lowering the effort parameter limits reasoning depth to reduce latency and cost; it does not improve accuracy and would undermine performance on a task requiring nuanced reasoning."}], "correct": "A", "select": 1}, {"id": "a3-090", "domain": 3, "task_id": "3.2", "objective": "Differentiate between Claude model types (Haiku, Sonnet, Opus)", "situation": "A software team wants a coding assistant for daily feature development: reviewing pull requests, generating unit tests, and answering questions about the codebase. They want strong code quality and agentic tool use without paying premium enterprise pricing.", "question": "Which model tier fits best?", "options": [{"letter": "A", "text": "Claude Opus delivers the highest accuracy for pull request reviews and test generation, essential for daily feature development.", "correct": false, "explanation": "While Claude Opus may offer the highest accuracy, it is designed for the most complex, large-scale agentic coding challenges and comes with premium pricing. For routine daily development work, Sonnet provides sufficient accuracy at a more appropriate cost, making Opus unnecessary for this scenario."}, {"letter": "B", "text": "Claude Haiku is the only model with extended thinking, enabling step-by-step code review and test generation at the lowest cost.", "correct": false, "explanation": "While Claude Haiku offers extended thinking that can facilitate step-by-step reasoning, the team requires strong general code quality and agentic tool use for daily development. Haiku's primary strength is speed and cost efficiency, not the frontier-level coding capability needed for thorough pull request reviews and test generation."}, {"letter": "C", "text": "Claude Sonnet delivers frontier-level coding and agentic tool use at a mid-range price, ideal for daily dev tasks.", "correct": true, "explanation": "Claude Sonnet provides frontier-level coding performance and agentic tool use at a price point that balances cost and capability, making it ideal for daily development tasks like code review and test generation. This matches the team's need for strong code quality without enterprise premium pricing."}, {"letter": "D", "text": "Claude Opus in fast mode matches Sonnet's pricing while delivering the highest accuracy for pull request reviews and test generation.", "correct": false, "explanation": "Fast mode on Claude Opus improves output speed but does not reduce pricing to Sonnet's level. Opus fast mode still operates at a premium price point, so it does not meet the team's requirement to avoid premium enterprise pricing."}], "correct": "C", "select": 1}, {"id": "a3-091", "domain": 3, "task_id": "3.1", "objective": "Select appropriate Claude product features (Projects, research mode, chat, artifacts)", "situation": "A game designer asks Claude to build a playable HTML5 prototype of a simple puzzle game with score tracking, and wants to keep tweaking the mechanics over the next hour based on playtesting.", "question": "Which feature keeps the running prototype visible and editable throughout that session?", "options": [{"letter": "A", "text": "A shared team workspace is necessary because individual accounts cannot render and tweak HTML5 prototypes.", "correct": false, "explanation": "Individual accounts can render artifacts including HTML5 prototypes without a shared team workspace. The artifact panel works within a single user’s session to display and iteratively edit the prototype."}, {"letter": "B", "text": "Research mode is required because its engine renders and updates interactive content during iterative tweaks.", "correct": false, "explanation": "Research mode performs web searches and has no engine for rendering or updating interactive HTML5 content. That rendering and iterative edit capability comes from artifacts, not research mode."}, {"letter": "C", "text": "A Project is needed since project knowledge alone renders the HTML5 prototype and supports iterative tweaks.", "correct": false, "explanation": "Project knowledge bases store reference documents and do not render live interactive content. Artifacts—not a Project—are required to render the HTML5 prototype and support iterative tweaks within the conversation."}, {"letter": "D", "text": "The artifact panel keeps the prototype rendered separately while accepting live iterative edits.", "correct": true, "explanation": "The artifact panel is designed to render self-contained content like an HTML5 game prototype separately from the chat, allowing the designer to see a live view while requesting iterative edits to the mechanics. It explicitly supports this interactive, visible feedback loop."}], "correct": "D", "select": 1}, {"id": "a4-001", "domain": 4, "task_id": "4.3", "objective": "Use Claude to support solution design, development, and iteration", "situation": "While designing a Claude-assisted tender-summary arrangement, an analyst finds that every draft needs the same three corrections.", "question": "Which response is most appropriate?", "options": [{"letter": "A", "text": "Keep making the three corrections by hand, since they take only a minute", "correct": false, "explanation": "Hand-correcting works while one person runs it, but the fix vanishes the moment anyone else uses the arrangement."}, {"letter": "B", "text": "Move to a more capable model, which should stop the corrections recurring", "correct": false, "explanation": "A stronger model cannot infer three unstated requirements, so the same three corrections will keep appearing."}, {"letter": "C", "text": "Fold those three corrections into the written instructions and try again", "correct": true, "explanation": "A repeated correction is a specification the instruction is missing. Writing it in makes the fix part of the design instead of something the analyst re-applies every time."}, {"letter": "D", "text": "Ask for three separate drafts each time and pick whichever needs least work", "correct": false, "explanation": "Generating extra drafts increases the work of choosing without addressing why every draft has the same gap."}], "correct": "C", "select": 1}, {"id": "a4-002", "domain": 4, "task_id": "4.5", "objective": "Communicate Claude's value and limitations to stakeholders", "situation": "", "question": "A product manager tells the team, \"Once we roll out Claude for contract drafting, we won't need any legal review at all.\" How should a solutions architect best respond to this claim?", "options": [{"letter": "A", "text": "Confirm that Claude has passed legal certification for autonomous contracts, so review is unnecessary after the pilot.", "correct": false, "explanation": "Claude is not legally certified to autonomously produce binding contracts, and no vendor certification eliminates the need for professional legal review."}, {"letter": "B", "text": "Suggest that legal review can be replaced by running the draft through a second Claude session with a different prompt.", "correct": false, "explanation": "Running a second Claude session does not substitute for qualified human legal judgment and does not eliminate the underlying risk of errors."}, {"letter": "C", "text": "Explain that Claude can speed up drafting, but a qualified legal reviewer should still check every clause before it is used.", "correct": true, "explanation": "Claude can accelerate drafting, but it can still produce plausible-sounding but inaccurate or non-compliant clauses, so human legal review remains essential before contracts are finalized."}, {"letter": "D", "text": "Recommend dropping contract drafting entirely, since Claude cannot produce usable legal language in any context.", "correct": false, "explanation": "Claude is genuinely useful for accelerating contract drafting; the correct guidance is to pair it with review, not abandon the use case."}], "correct": "C", "select": 1}, {"id": "a4-003", "domain": 4, "task_id": "4.4", "objective": "Integrate Claude into existing workflows to augment or redesign them", "situation": "A back-office team plans to remove three handoffs between reviewers by having Claude carry each case straight through. Those handoffs are also where colleagues currently catch one another's mistakes.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "Reviewers may resist a change that removes part of what they currently do each day.", "correct": false, "explanation": "Resistance to change is common and is addressed by involving the reviewers in the redesign. It affects how smoothly the change lands rather than whether the resulting process is safe."}, {"letter": "B", "text": "The informal checks that lived in those handoffs disappear with nothing put in their place.", "correct": true, "explanation": "The handoffs were doing two jobs, and only the visible one is being replaced. Removing them takes out the peer review that was catching errors, so the redesigned process is not just faster but materially less accurate unless a deliberate check is designed back in."}, {"letter": "C", "text": "The redesigned process may be considerably harder to explain to auditors than the version it replaces.", "correct": false, "explanation": "Explaining a new process to auditors takes documentation effort and is a normal part of any change. It becomes a serious problem only because of the missing checks, which is the underlying issue."}, {"letter": "D", "text": "The time saved may be less than expected once the new process is running at full volume.", "correct": false, "explanation": "A smaller saving than hoped would make the change less worthwhile, which is disappointing but recoverable. The team would still be able to see the shortfall and adjust."}], "correct": "B", "select": 1}, {"id": "a4-004", "domain": 4, "task_id": "4.5", "objective": "Communicate Claude's value and limitations to stakeholders", "situation": "An analyst is presenting a Claude-assisted workflow to sceptical senior stakeholders next week. Which two things should the presentation include?", "question": "Select two.", "options": [{"letter": "A", "text": "A comparison of technical benchmark scores across each of the underlying model tiers.", "correct": false, "explanation": "Benchmark scores are not what a business audience needs, and they say nothing about this workflow's value."}, {"letter": "B", "text": "A claim that Claude will not make any factual errors once all its instructions are written well.", "correct": false, "explanation": "Good instructions reduce error but do not eliminate it, so this claim is false and will damage trust later."}, {"letter": "C", "text": "A specific, measured example of what the workflow improved during the pilot it ran on.", "correct": true, "explanation": "A measured pilot result is concrete evidence of value rather than an assertion stakeholders are asked to take on trust."}, {"letter": "D", "text": "Where a human still reviews the output, and why that review step has been kept in place.", "correct": true, "explanation": "Naming where human review remains, and why, is honest limitation-setting and is what makes the value claim credible."}, {"letter": "E", "text": "A promise that the workflow removes any need to check the output from this point onward.", "correct": false, "explanation": "Promising that checking is no longer needed oversells the tool and sets the expectation that will break first."}], "correct": ["C", "D"], "select": 2}, {"id": "a4-005", "domain": 4, "task_id": "4.1", "objective": "Apply Claude to analyze requirements and use cases", "situation": "An HR analyst plans to use Claude to analyse requirements for a new grievance-handling process. The best source material is a folder of named employee case files.", "question": "What should the analyst do first?", "options": [{"letter": "A", "text": "Check the organisation's policy for this category of data and mask identifiers before anything is uploaded.", "correct": true, "explanation": "Requirements analysis is still legitimate here, but grievance case files are sensitive personal data. Confirming what policy allows and removing or masking identifiers first lets the analysis proceed on the patterns that matter while keeping the material within the rules."}, {"letter": "B", "text": "Ask Claude to summarise the case files, then delete the conversation once the analysis has been finished.", "correct": false, "explanation": "Deleting afterwards does not undo the upload, and it treats a data-handling decision as something to clean up later. The judgment about whether identifiable case files may be shared has to be made before, not after."}, {"letter": "C", "text": "Ask Claude to confirm whether the case files contain any personal data before any of them are uploaded to it.", "correct": false, "explanation": "Claude cannot assess files it has not been given, so this either does nothing or requires the very upload that is in question. The analyst already knows the files name employees, so the answer is not the missing piece."}, {"letter": "D", "text": "Upload the case files into a private project workspace so only the analyst can open the resulting chats.", "correct": false, "explanation": "Restricting who can see the resulting chats does not address whether the material should have been uploaded in the first place. Access control inside the tool is not a substitute for the organisation's rule about this class of data."}], "correct": "A", "select": 1}, {"id": "a4-006", "domain": 4, "task_id": "4.2", "objective": "Leverage Claude for research, planning, and process optimization", "situation": "A project manager asks Claude for a delivery plan and receives a confident schedule that quietly assumes full-time availability the team does not have.", "question": "What is the most effective response?", "options": [{"letter": "A", "text": "Ask Claude to compress the proposed schedule so that it still lands on the original date promised to the sponsor.", "correct": false, "explanation": "Compressing a schedule that was already built on an impossible assumption makes it less achievable, not more. The date is preserved on paper while the plan moves further from what the team can deliver."}, {"letter": "B", "text": "Ask Claude to add a contingency buffer to each phase so the plan can absorb the availability problem.", "correct": false, "explanation": "Buffers absorb ordinary variation, not a structural error in the availability the plan assumes. Padding phases spreads the same wrong assumption across the whole schedule and makes it harder to see."}, {"letter": "C", "text": "Give Claude the real constraints and ask for a plan that states its assumptions where facts are missing.", "correct": true, "explanation": "The schedule is wrong because the planning inputs were missing, so the fix is to supply them. Asking Claude to state its assumptions explicitly also makes the next gap visible immediately instead of leaving it buried inside confident dates."}, {"letter": "D", "text": "Ask Claude to explain how it built the schedule so the manager can judge whether to trust the dates.", "correct": false, "explanation": "Understanding the reasoning is a reasonable check and may reveal the flawed assumption, but it stops short of fixing anything. The manager already knows the availability is wrong; what is missing is the correct input."}], "correct": "C", "select": 1}, {"id": "a4-007", "domain": 4, "task_id": "4.3", "objective": "Use Claude to support solution design, development, and iteration", "situation": "A team is setting up a Claude-assisted way of producing meeting notes. Nothing has been agreed or written down, people on the team hold different views about what the notes should contain, and the notes will be circulated to everyone who attended, so the team needs something it can depend on before this goes into regular use.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Run it over one week of real work; collect specific examples of where it fell short; draft the new way of working with the people who do the job; revise the written instructions and run it again.", "correct": false, "explanation": "Starting with a week of live use puts notes in front of every attendee while the team still disagrees about what those notes should contain, so the shortcomings it turns up describe several people's private habits rather than one shared way of working, and there is nothing agreed for the evidence to be measured against."}, {"letter": "B", "text": "Collect specific examples of where it fell short; draft the new way of working with the people who do the job; revise the written instructions and run it again; run it over one week of real work.", "correct": false, "explanation": "There is no agreed practice yet whose shortcomings could be collected, so the sequence opens with evidence it cannot have, and the week of real use lands at the very end, after the written instructions have already been revised, where the evidence it produces feeds nothing."}, {"letter": "C", "text": "Revise the written instructions and run it again; run it over one week of real work; draft the new way of working with the people who do the job; collect specific examples of where it fell short.", "correct": false, "explanation": "Nothing has been written down, so there are no instructions to revise first, and the specific failure examples are gathered last, at the point where there is no remaining step to act on them."}, {"letter": "D", "text": "Draft the new way of working with the people who do the job; run it over one week of real work; collect specific examples of where it fell short; revise the written instructions and run it again.", "correct": true, "explanation": "Drafting with the people who do the job settles the goal and how the work is split between them and Claude before anything is delegated, a week of real use then produces honest evidence against that agreed intent, specific failure examples say what to change, and revising the written instructions and running it again closes the evaluate-and-refine loop for another pass."}], "correct": "D", "select": 1}, {"id": "a4-008", "domain": 4, "task_id": "4.3", "objective": "Use Claude to support solution design, development, and iteration", "situation": "Over three weeks a team refined the instructions for a Claude-drafted internal newsletter until the editor was consistently happy with every draft. Nobody asked the readers.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The three weeks of refinement consumed effort that could have gone into writing newsletters.", "correct": false, "explanation": "The refinement effort is a sunk cost that will be repaid if the arrangement works. Time spent iterating is expected; the problem is what the iteration was aimed at."}, {"letter": "B", "text": "The instructions have grown long enough that new team members will find them hard to follow at first.", "correct": false, "explanation": "Long instructions are an onboarding inconvenience that can be tidied up. They make the arrangement harder to hand over, but they do not make the newsletter less effective for the people reading it."}, {"letter": "C", "text": "The drafts are now tuned to one editor's taste, and no one has checked that they serve the readers.", "correct": true, "explanation": "Iteration improves whatever it is measured against, and here the only measure was one editor's approval. Three weeks of refinement have optimised for a proxy, so the drafts could satisfy the editor completely while serving readers worse than before."}, {"letter": "D", "text": "The editor may become the only person able to judge whether a draft is ready to be published.", "correct": false, "explanation": "Concentrating judgment in one person is a genuine weakness in the design, and it follows from the same root cause. It is a consequence of tuning to one taste rather than the primary danger, which is that readers were never consulted."}], "correct": "C", "select": 1}, {"id": "a4-009", "domain": 4, "task_id": "4.1", "objective": "Apply Claude to analyze requirements and use cases", "situation": "With no internal documents to hand, an analyst asks Claude to list the standard requirements for a customer complaints process and pastes the reply straight into the project brief.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The analyst will have to repeat the request if the conversation is closed before the reply is saved.", "correct": false, "explanation": "Losing the reply costs a few minutes to regenerate. It is an inconvenience rather than a risk to the correctness of the requirements the project will be run against."}, {"letter": "B", "text": "The brief will be rather longer than the sponsor expects and may need trimming before the next review meeting.", "correct": false, "explanation": "Length is easy to fix and easy to notice. A brief that is too long gets edited, whereas plausible requirements nobody sourced tend to survive review precisely because they read like standard practice."}, {"letter": "C", "text": "The brief will carry generic requirements that sound authoritative but reflect no source and no context.", "correct": true, "explanation": "Asked for standard requirements with no material to work from, Claude produces a confident generic list. Pasted straight into the brief it becomes the team's stated scope, even though it reflects no regulator, no system and no customer this organisation actually has."}, {"letter": "D", "text": "The reply may be phrased for a general audience rather than in the complaints team's own vocabulary.", "correct": false, "explanation": "Vocabulary mismatch is a readability problem that a rewrite fixes. It may slow the team down slightly, but the brief would still be built on requirements that were never traced to anything real."}], "correct": "C", "select": 1}, {"id": "a4-010", "domain": 4, "task_id": "4.4", "objective": "Integrate Claude into existing workflows to augment or redesign them", "situation": "An operations director wants Claude added to a customer onboarding process that the team agrees is slow, full of rework, and poorly understood.", "question": "What should the team do first?", "options": [{"letter": "A", "text": "Ask Claude to redesign the whole onboarding process from scratch and compare it with what happens now.", "correct": false, "explanation": "A redesign proposed before the current problems are diagnosed rests on assumptions about why the process fails. It also asks the team to compare today's process against a proposal, when what they lack is an understanding of today's process."}, {"letter": "B", "text": "Give every member of the onboarding team access and let each of them find their own uses for it.", "correct": false, "explanation": "Open access encourages useful experimentation but leaves the process itself untouched. Individuals will speed up their own steps in different ways, which can make an already inconsistent process harder to follow."}, {"letter": "C", "text": "Establish why the current process produces rework, before deciding where Claude would fit into it.", "correct": true, "explanation": "Integrating Claude into a process nobody understands simply performs the existing mistakes faster. Finding out where the rework originates tells the team whether the answer is to augment a step, redesign a sequence, or fix something that has nothing to do with Claude at all."}, {"letter": "D", "text": "Add Claude at the step that generates the most documents, since that is where the volume sits.", "correct": false, "explanation": "Document volume marks where effort is spent, not where the process breaks. If the rework is caused by incomplete information arriving at step one, accelerating the busiest step produces more flawed documents rather than fewer."}], "correct": "C", "select": 1}, {"id": "a4-011", "domain": 4, "task_id": "4.1", "objective": "Apply Claude to analyze requirements and use cases", "situation": "A trade association's events team submits a use-case request stating that Claude should \"take over the speaker selection\" for next year's conference. The business analyst who reviews intake requests must assess it.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Check whether the events team's plan fits inside the association's current allowance", "correct": false, "explanation": "Cost is a real consideration, but it is a downstream question. A use case that should not be delegated at all does not become sound because it is affordable."}, {"letter": "B", "text": "Establish what the selection actually decides and who is accountable for the outcome", "correct": true, "explanation": "Requirements analysis starts with understanding the decision itself: what it determines, for whom, and who must answer for it. Only then can you say which parts are delegable and which must stay with the panel."}, {"letter": "C", "text": "Ask Claude to draft selection criteria from the previous three years of speaker lists", "correct": false, "explanation": "Drafting criteria assumes the use case is sound and that past selections are a fair model; that assumption is exactly what has not been tested yet."}, {"letter": "D", "text": "Run a trial in which Claude ranks this year's applicants and compare with the panel", "correct": false, "explanation": "A comparison trial is a reasonable later step, but running it before anyone has defined what the selection is for gives no basis for judging the result."}], "correct": "B", "select": 1}, {"id": "a4-012", "domain": 4, "task_id": "4.2", "objective": "Leverage Claude for research, planning, and process optimization", "situation": "An analyst asks Claude to redesign a claims workflow using a process map stored in the team's project, unaware that the map has not been updated for two years.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The redesign may optimise a process the team no longer follows, and nothing in the output flags it.", "correct": true, "explanation": "Claude works from the project knowledge it is given, so a map that is two years stale can yield a polished redesign of a process the team has already changed. The help centre warns that Claude can sound convincing while not being grounded in fact, so nothing in the redesign marks its starting point as obsolete, and no amount of reprompting repairs a stale source - only refreshing the map does."}, {"letter": "B", "text": "The project may hold several versions of the map, making it unclear which one Claude actually used.", "correct": false, "explanation": "Ambiguity about which version was used is a housekeeping weakness worth fixing. In this scenario the problem is not which map was chosen but that the only map available may be out of date."}, {"letter": "C", "text": "The analyst may need to rerun the request once an updated version of the process map is available.", "correct": false, "explanation": "Rerunning the request once the map is refreshed is the remedy, not the risk. The cost of a repeated request is trivial next to acting on a redesign nobody realised was obsolete."}, {"letter": "D", "text": "The redesign may use terms for the workflow stages that the current claims team no longer recognises.", "correct": false, "explanation": "Outdated terminology is a visible symptom that would prompt someone to question the output. It is a useful warning sign rather than the underlying danger."}], "correct": "A", "select": 1}, {"id": "a4-013", "domain": 4, "task_id": "4.3", "objective": "Use Claude to support solution design, development, and iteration", "situation": "A team designed its Claude-assisted case-note arrangement around one unusually clean example case, and it worked perfectly. They now plan to put all case types through it.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "Claude will produce case notes that are noticeably longer than the previous ones", "correct": false, "explanation": "Note length is a formatting matter that a reviewer would spot immediately and can be specified away."}, {"letter": "B", "text": "The arrangement was tuned to a case that does not represent the messy real ones", "correct": true, "explanation": "A design validated on one tidy case has been fitted to the easiest conditions it will ever face. The awkward, incomplete and contested cases are where it will fail, and none of them was tested."}, {"letter": "C", "text": "Colleagues will find the written instructions too long to read before using it", "correct": false, "explanation": "Instruction length is a usability nuisance that is easy to notice and easy to shorten."}, {"letter": "D", "text": "The team will exceed its weekly usage allowance within the first fortnight of use", "correct": false, "explanation": "Allowance pressure is a manageable operational cost, and it would show up plainly rather than producing quietly wrong case notes."}], "correct": "B", "select": 1}, {"id": "a4-014", "domain": 4, "task_id": "4.3", "objective": "Use Claude to support solution design, development, and iteration", "situation": "In a pilot of a new Claude-assisted briefing arrangement, three of eight testers report that the output is \"not quite right\" but say nothing more specific.", "question": "What should the pilot's owner do first?", "options": [{"letter": "A", "text": "Rewrite the instructions to cover the areas the owner believes are most likely to be causing the trouble.", "correct": false, "explanation": "Rewriting on a hunch changes the arrangement without knowing what is wrong. If the guess misses, the pilot now has two versions and still no diagnosis, and the testers' original complaint remains unexplained."}, {"letter": "B", "text": "Ask the three testers for specific examples, with what the output said and what it should have said.", "correct": true, "explanation": "\"Not quite right\" is not actionable, and the cheapest way to make it actionable is to ask for cases. A concrete pair of what was produced and what was wanted turns vague dissatisfaction into a specific gap the instructions can address."}, {"letter": "C", "text": "Ask the five satisfied testers what they did differently and circulate that as guidance to everyone.", "correct": false, "explanation": "Learning from the satisfied testers may eventually help, but it assumes the difference lies in how the tool was used rather than in the briefings themselves. It also leaves the three unresolved complaints uninvestigated."}, {"letter": "D", "text": "Pause the pilot until the owner can sit with each tester and watch a full briefing being produced.", "correct": false, "explanation": "Direct observation is thorough but slow, and pausing the pilot stops the flow of evidence just when it is needed. Asking for examples costs an email and usually locates the problem immediately."}], "correct": "B", "select": 1}, {"id": "a4-015", "domain": 4, "task_id": "4.1", "objective": "Apply Claude to analyze requirements and use cases", "situation": "An operations team wants to evaluate whether invoice-query triage, currently done by two people, is a sound use case to hand to Claude.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Compare and fix the guidance; run the small sample; agree what a correct decision looks like; extend to the full queue.", "correct": false, "explanation": "Fixing the guidance before anything has been run means correcting problems nobody has observed, and the standard for a correct decision arrives after the sample has already been judged against it."}, {"letter": "B", "text": "Agree what a correct decision looks like; run the small sample; compare and fix the guidance; extend to the full queue.", "correct": true, "explanation": "The standard comes first so results can be graded, the pilot stays small so mistakes are cheap, the comparison against the current handlers shows where the guidance falls short, and only a corrected approach is extended to the full queue with continued spot checks."}, {"letter": "C", "text": "Extend to the full queue; agree what a correct decision looks like; run the small sample; compare and fix the guidance.", "correct": false, "explanation": "Going to the full queue first is the opposite of piloting. Every later step then happens under live load, so mistakes reach real invoice queries before anyone has agreed what a good triage decision even is."}, {"letter": "D", "text": "Run the small sample; agree what a correct decision looks like; extend to the full queue; compare and fix the guidance.", "correct": false, "explanation": "Running the sample before agreeing what a correct decision looks like leaves the team with output they cannot grade, and extending to the full queue before the comparison means any weakness is discovered at full volume."}], "correct": "B", "select": 1}, {"id": "a4-016", "domain": 4, "task_id": "4.4", "objective": "Integrate Claude into existing workflows to augment or redesign them", "situation": "A claims team is deciding whether to augment its current process with Claude or to redesign the process around what Claude makes possible.", "question": "Which condition most justifies a redesign rather than augmentation?", "options": [{"letter": "A", "text": "Several steps exist only to work around a limitation that no longer applies to this work.", "correct": true, "explanation": "Redesign is warranted when the structure of the process encodes a constraint that has gone away, because no amount of speeding up those steps removes work that no longer needs doing. Augmenting them would preserve the workaround permanently."}, {"letter": "B", "text": "Leadership has asked for a visible improvement that staff and customers will both notice soon.", "correct": false, "explanation": "A demand for visible improvement sets an expectation about impact, not about which approach is appropriate. Augmenting a well-chosen step can be highly visible, and redesigning for visibility invites disruption without a structural reason."}, {"letter": "C", "text": "The team finds the current process tedious and would welcome a change to how the work is done.", "correct": false, "explanation": "Tedium is a good reason to look for delegation candidates within the process, which is augmentation. It does not indicate that the shape of the process itself is wrong."}, {"letter": "D", "text": "The current process has more steps than the equivalent process at a comparable organisation.", "correct": false, "explanation": "A step count comparison ignores why the extra steps exist. Another organisation may face different regulation, volumes or systems, so the difference is a prompt to investigate rather than evidence for a redesign."}], "correct": "A", "select": 1}, {"id": "a4-017", "domain": 4, "task_id": "4.5", "objective": "Communicate Claude's value and limitations to stakeholders", "situation": "A stakeholder is confused that Claude gave two slightly different summaries of the same document on two separate runs, and asks whether the model is broken.", "question": "What is the best explanation to give?", "options": [{"letter": "A", "text": "Explain that document summaries always fail after the first request, so each summary attempt requires a new conversation.", "correct": false, "explanation": "Document summaries do not systematically fail after the first request; this explanation wrongly portrays expected variability as a failure. Each summary attempt can produce slightly different wording without needing a new conversation."}, {"letter": "B", "text": "Explain that Claude's responses are not perfectly deterministic, so minor wording variation between runs is normal and expected.", "correct": true, "explanation": "Claude's responses are not perfectly deterministic because text generation involves random sampling, which can lead to minor wording differences between runs. This variability is normal and expected, not a malfunction."}, {"letter": "C", "text": "Explain that the second summary is correct and the first one should be treated as a bug that support must investigate right away.", "correct": false, "explanation": "Treating the second summary as correct and the first as a bug misrepresents normal output variability. Slight variations between runs are common and do not indicate a defect requiring investigation."}, {"letter": "D", "text": "Explain that variation only occurs when a document exceeds the context window, so shortening the document resolves the issue.", "correct": false, "explanation": "Output variation is not caused by document length or exceeding the context window; it stems from Claude's probabilistic generation process. Shortening the document will not eliminate normal, minor wording differences between runs."}], "correct": "B", "select": 1}, {"id": "a4-018", "domain": 4, "task_id": "4.2", "objective": "Leverage Claude for research, planning, and process optimization", "situation": "A conference organiser wants Claude to help plan the run-up to a 400-delegate event.", "question": "Which input will make the resulting plan most useful?", "options": [{"letter": "A", "text": "The actual dates, the team's availability and the tasks that went wrong last year", "correct": true, "explanation": "Real dates, real capacity and the known failure points from last year are the material that makes a plan specific to this event rather than a template anyone could have written."}, {"letter": "B", "text": "The delegate list and the venue's promotional brochure for this year's conference", "correct": false, "explanation": "A delegate list and a brochure describe who attends and how the venue markets itself, neither of which shapes the run-up schedule."}, {"letter": "C", "text": "A request to produce the most thorough conference plan that Claude is able to write", "correct": false, "explanation": "Asking for thoroughness without supplying facts produces length, not fit; the plan still has to guess the constraints."}, {"letter": "D", "text": "A note that the event is important and that the plan must not miss anything at all", "correct": false, "explanation": "Emphasis is not information. Telling Claude the event matters gives it nothing new to plan around."}], "correct": "A", "select": 1}, {"id": "a4-019", "domain": 4, "task_id": "4.1", "objective": "Apply Claude to analyze requirements and use cases", "situation": "", "question": "What is an effective way to use Claude to analyze requirements and use cases early in the development process?", "options": [{"letter": "A", "text": "Ask Claude to simulate a developer reviewing requirements and ask clarifying questions for ambiguity.", "correct": true, "explanation": "Simulating a developer reviewing requirements and asking clarifying questions allows Claude to proactively identify ambiguities, gaps, and inconsistencies, improving clarity and reducing downstream issues."}, {"letter": "B", "text": "Ask Claude to ignore the current requirements draft and generate a completely new set of use cases from scratch.", "correct": false, "explanation": "Ignoring the current requirements draft to generate new use cases from scratch risks losing valuable context, stakeholder input, and business alignment; instead, Claude should refine and analyze the provided requirements."}, {"letter": "C", "text": "Ask Claude to examine the completed code for missing logic and infer requirements that were not documented.", "correct": false, "explanation": "Examining completed code for missing logic is a reactive approach that addresses issues only after development, missing the chance to catch ambiguities early and leading to costly rework."}, {"letter": "D", "text": "Ask Claude to independently review requirements, resolve ambiguities, and finalize them with no human check.", "correct": false, "explanation": "Having Claude independently review and finalize requirements without human check eliminates necessary oversight; human validation is critical for context, ethics, and domain nuances, so Claude should augment, not replace, human reviewers."}], "correct": "A", "select": 1}, {"id": "a4-020", "domain": 4, "task_id": "4.2", "objective": "Leverage Claude for research, planning, and process optimization", "situation": "An operations planner asks Claude to compress a six-week onboarding schedule into four weeks. The returned plan is detailed and internally consistent, and the planner circulates it as the new schedule.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The plan will be written for a reader who is far more senior than the planner is", "correct": false, "explanation": "Register is easy to adjust and easy to notice. It does not determine whether the compressed schedule is workable."}, {"letter": "B", "text": "The plan will take longer to read than the six-week schedule it is meant to replace", "correct": false, "explanation": "A long plan is an inconvenience. An unexamined assumption inside it is what breaks the onboarding."}, {"letter": "C", "text": "Claude will decline to shorten a schedule without seeing the original planning file", "correct": false, "explanation": "Claude will attempt the compression with whatever it is given; refusing for want of a file is not the failure mode here."}, {"letter": "D", "text": "Steps were dropped or overlapped on assumptions the planner never stated or checked", "correct": true, "explanation": "To fit six weeks into four, something had to give. Because the plan reads as coherent, the compromises it made are invisible, and nobody has confirmed that they are acceptable."}], "correct": "D", "select": 1}, {"id": "a4-021", "domain": 4, "task_id": "4.3", "objective": "Use Claude to support solution design, development, and iteration", "situation": "A team has spent three weeks refining a Claude-assisted weekly digest. The last two rounds of changes produced no difference anyone could notice.", "question": "What is the most appropriate next step?", "options": [{"letter": "A", "text": "Ask Claude to grade its own digests and continue until the grades stop rising", "correct": false, "explanation": "Self-grading tells the team how the digest scores itself, which is not the same as whether readers can use it."}, {"letter": "B", "text": "Keep refining until the digest needs no editing at all before it is circulated", "correct": false, "explanation": "Requiring zero editing sets a standard the arrangement was never designed to meet and delays any benefit indefinitely."}, {"letter": "C", "text": "Start again with a fresh set of instructions written by a different team member", "correct": false, "explanation": "Discarding three weeks of refinement resets the work without any evidence that a fresh start would do better."}, {"letter": "D", "text": "Put the current version into use and review it after a period of real work", "correct": true, "explanation": "Iteration has stopped yielding gains, so the remaining information will come from real use. Adopting it with a review date set keeps the loop open without spending more effort on a plateau."}], "correct": "D", "select": 1}, {"id": "a4-022", "domain": 4, "task_id": "4.4", "objective": "Integrate Claude into existing workflows to augment or redesign them", "situation": "An operations team is embedding Claude into a high-volume workflow that must remain low cost while still handling reasoning tasks such as classifying incoming requests with strong accuracy.", "question": "Which model tier best matches this integration's requirements?", "options": [{"letter": "A", "text": "The most capable, highest-priced model tier, since only it can perform any classification reasoning", "correct": false, "explanation": "Reasoning capability is not exclusive to the most expensive tier; using it here would add unnecessary cost for a workflow prioritizing low cost at high volume."}, {"letter": "B", "text": "A model tier selected purely for its support of computer use screenshots and clicks", "correct": false, "explanation": "Computer use support is unrelated to classifying incoming text requests and is not the deciding factor for this workflow."}, {"letter": "C", "text": "A model tier chosen only because it supports the longest possible context window available", "correct": false, "explanation": "Context window length is not the stated constraint in this scenario, so choosing a tier for that reason alone ignores the cost and volume requirements described."}, {"letter": "D", "text": "The fastest, most economical model tier, since it offers strong reasoning at high volume for low cost", "correct": true, "explanation": "The fastest, most economical model tier is positioned for high-volume intelligent processing and cost-sensitive deployments that still need strong reasoning, matching this workflow's requirements."}], "correct": "D", "select": 1}, {"id": "a4-023", "domain": 4, "task_id": "4.5", "objective": "Communicate Claude's value and limitations to stakeholders", "situation": "A stakeholder asks what a model's \"reliable knowledge cutoff\" means when the solutions architect recommends verifying any recent-event claims Claude makes.", "question": "What is the most accurate explanation?", "options": [{"letter": "A", "text": "It is the date up to which the model was trained, so its awareness may be limited after this date.", "correct": true, "explanation": "The reliable knowledge cutoff is the date through which the model's training data extends, making its knowledge most dependable up to that point. Events occurring after this date may be unknown to the model without external tools, so verification is recommended for recent claims."}, {"letter": "B", "text": "It is the exact date the model will stop being available for use through the API going forward.", "correct": false, "explanation": "This describes a model deprecation or end-of-life date, which is unrelated to the knowledge cutoff. The knowledge cutoff refers to the timeframe of the training data, not the availability of the model via the API."}, {"letter": "C", "text": "It is a setting that administrators can change at any time to extend how current the model's knowledge is.", "correct": false, "explanation": "The knowledge cutoff is fixed by the data used during training and is not a configurable setting. Administrators cannot modify it to extend the model's knowledge after deployment."}, {"letter": "D", "text": "It is the date after which every response becomes randomly generated instead of based on training data.", "correct": false, "explanation": "Responses after the knowledge cutoff are not randomly generated; the model continues to reason based on its training. Its awareness of recent events may be limited, but it does not suddenly produce random output."}], "correct": "A", "select": 1}, {"id": "a4-024", "domain": 4, "task_id": "4.2", "objective": "Leverage Claude for research, planning, and process optimization", "situation": "A school's exams officer has been asked to plan a new revision-support programme for the summer term and intends to use Claude to help build the plan.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Have three colleagues each request their own plan and compare the four versions later", "correct": false, "explanation": "Comparing four unconstrained plans multiplies the work without adding information, because none of them was told what has to be true."}, {"letter": "B", "text": "Request the plan and next year's examination timetable in one long single message", "correct": false, "explanation": "Combining two unrelated planning problems in one request makes both weaker and makes it harder to see which answer belongs to which."}, {"letter": "C", "text": "Ask Claude for a general revision-programme plan and adapt whatever it produces", "correct": false, "explanation": "A generic plan built on unstated constraints has to be unpicked afterwards, and the assumptions it made stay invisible while she adapts it."}, {"letter": "D", "text": "Set out the real constraints — dates, staffing, rooms, budget — then ask for a plan", "correct": true, "explanation": "A plan is only useful if it respects the constraints it has to live inside. Stating dates, staffing, rooms and budget up front is what turns a generic outline into something the school can actually run."}], "correct": "D", "select": 1}, {"id": "a4-025", "domain": 4, "task_id": "4.1", "objective": "Apply Claude to analyze requirements and use cases", "situation": "A shared services team receives a request to use Claude for summarising supplier contracts.", "question": "Which sequence best analyses the use case before any commitment is made?", "options": [{"letter": "A", "text": "Confirm the decision the summaries support; check the upload policy; try a handful of contracts; compare with trusted summaries.", "correct": true, "explanation": "Knowing which decision the summaries feed defines what they must contain, the policy check establishes whether the contracts may be uploaded at all, a small trial gives real evidence cheaply, and comparing against summaries the team already trusts shows whether the quality is good enough to commit to."}, {"letter": "B", "text": "Try a handful of contracts; confirm the decision the summaries support; check the upload policy; compare with trusted summaries.", "correct": false, "explanation": "Trying contracts first means uploading material before anyone has checked whether that is permitted, and the trial is run without knowing what the summaries are for, so there is no basis for judging the results."}, {"letter": "C", "text": "Check the upload policy; compare with trusted summaries; confirm the decision the summaries support; try a handful of contracts.", "correct": false, "explanation": "Comparing with trusted summaries before any have been produced is impossible, and confirming the purpose after that comparison leaves the trial run at the very end with nothing left to inform."}, {"letter": "D", "text": "Compare with trusted summaries; try a handful of contracts; confirm the decision the summaries support; check the upload policy.", "correct": false, "explanation": "Starting with a comparison there is nothing to compare against, and leaving the policy check until last means the contracts have already been uploaded by the time anyone asks whether they should have been."}], "correct": "A", "select": 1}, {"id": "a4-026", "domain": 4, "task_id": "4.5", "objective": "Communicate Claude's value and limitations to stakeholders", "situation": "An operations director presents the results of a two-week Claude pilot to the executive team, showing only the three best outputs the team produced.", "question": "What is the greatest risk in communicating the results this way?", "options": [{"letter": "A", "text": "The pilot's outputs cannot be reused afterwards, because examples shown to executives must be regenerated.", "correct": false, "explanation": "There is no such constraint on reusing output that has been presented."}, {"letter": "B", "text": "Executives will conclude the pilot used the wrong tool and will ask for a different vendor to be evaluated instead.", "correct": false, "explanation": "Strong examples do not prompt a vendor challenge; the problem is that they are not representative."}, {"letter": "C", "text": "The team will be required to repeat the whole pilot from the start before any rollout decision is taken.", "correct": false, "explanation": "A repeat pilot is not the inherent risk; misplaced expectations are."}, {"letter": "D", "text": "Executives set expectations from unrepresentative examples and are surprised when routine output needs editing.", "correct": true, "explanation": "Cherry-picked results overstate the typical case, and the gap surfaces later as disappointment or lost trust."}], "correct": "D", "select": 1}, {"id": "a4-027", "domain": 4, "task_id": "4.1", "objective": "Apply Claude to analyze requirements and use cases", "situation": "A project manager has two hundred pages of stakeholder interview notes and needs to turn them into a requirements draft. Which two uses of Claude fit this task?", "question": "Select two.", "options": [{"letter": "A", "text": "Have Claude list where the notes contradict one another so a person resolves each conflict.", "correct": true, "explanation": "Surfacing contradictions is analysis that helps humans decide; the resolution itself stays with the people accountable for scope."}, {"letter": "B", "text": "Have Claude supply the missing answers for the stakeholders who were never interviewed.", "correct": false, "explanation": "Inventing answers nobody gave is fabrication presented as research, and it corrupts the requirements it feeds."}, {"letter": "C", "text": "Have Claude group the interview notes into themes that the team then reviews and challenges.", "correct": true, "explanation": "Grouping a large volume of text into reviewable themes is well-specified, text-heavy work that a person still checks afterwards."}, {"letter": "D", "text": "Have Claude settle which of the requirements make the final scope, so that the debate is skipped.", "correct": false, "explanation": "Scope is a decision the team owns, and delegating it removes the accountability that has to sit with a person."}, {"letter": "E", "text": "Have Claude sign the requirements off on behalf of the stakeholders who were interviewed.", "correct": false, "explanation": "Sign-off on behalf of stakeholders is not a task to delegate; it misrepresents people who never agreed to it."}], "correct": ["A", "C"], "select": 2}, {"id": "a4-028", "domain": 4, "task_id": "4.2", "objective": "Leverage Claude for research, planning, and process optimization", "situation": "Claude has produced two credible candidate plans for a departmental restructure, and the sponsor asks the planning lead which one is better.", "question": "What is the most appropriate next step?", "options": [{"letter": "A", "text": "Ask Claude to rewrite both plans in the same structure so the sponsor can read them side by side more easily.", "correct": false, "explanation": "A common structure makes the plans easier to read and is worth doing, but presentation is not evaluation. The sponsor would still face two neatly formatted options with no agreed basis for choosing between them."}, {"letter": "B", "text": "Ask Claude which plan it prefers and take that as the department's recommendation to the sponsor.", "correct": false, "explanation": "A stated preference gives an answer without an argument the sponsor can examine. The choice between restructure plans depends on what the department values, which is a judgment the department owns rather than one the tool can settle."}, {"letter": "C", "text": "Combine the strongest parts of both plans into a third plan and present that to the sponsor instead.", "correct": false, "explanation": "Merging attractive pieces of two plans often breaks the internal logic that made each coherent, and it produces a third option nobody has evaluated. It also avoids the comparison the sponsor actually asked for."}, {"letter": "D", "text": "Agree the criteria that matter here, then compare both plans against them and record the reasoning.", "correct": true, "explanation": "Naming the criteria first turns \"which is better\" into a question that can be answered and defended, and recording the reasoning gives the sponsor something to challenge. Claude can then do the comparison work against a standard the humans set."}], "correct": "D", "select": 1}, {"id": "a4-029", "domain": 4, "task_id": "4.3", "objective": "Use Claude to support solution design, development, and iteration", "situation": "Two rounds into designing a Claude-assisted induction pack, an education team finds the drafts are accurate but pitched at the wrong level for new teaching assistants.", "question": "What should they do first?", "options": [{"letter": "A", "text": "Ask the teaching assistants to edit each draft themselves as part of their induction", "correct": false, "explanation": "Pushing the rewrite onto the readers hides the design fault rather than fixing it, and the pack stays mispitched for everyone."}, {"letter": "B", "text": "State who the reader is and what they already know, then try the same case again", "correct": true, "explanation": "The fault is a missing specification, not a reasoning limit. Naming the audience and their starting knowledge addresses the actual cause, and re-running the same case shows whether it worked."}, {"letter": "C", "text": "Move the work to the most capable model available and re-run the same instruction", "correct": false, "explanation": "A stronger model still has no idea who the reader is; escalating tier does not supply information the instruction never contained."}, {"letter": "D", "text": "Rewrite every part of the arrangement at once so the next round is a clean start", "correct": false, "explanation": "Changing everything at once means that if the next draft is better, nobody will know which change was responsible."}], "correct": "B", "select": 1}, {"id": "a4-030", "domain": 4, "task_id": "4.5", "objective": "Communicate Claude's value and limitations to stakeholders", "situation": "A customer support director wants to justify a Claude-based ticket triage pilot to executives and asks how to frame its value.", "question": "What is the best way to present this business case?", "options": [{"letter": "A", "text": "Present only the model's technical architecture details, since executives mainly evaluate proposals on implementation specifics.", "correct": false, "explanation": "Executives generally care about business outcomes and ROI more than low-level technical architecture, so this framing misses the audience's priorities."}, {"letter": "B", "text": "Present concrete metrics such as reduced average handling time and ticket deflection rate compared against the current process.", "correct": true, "explanation": "Framing value in terms of measurable business outcomes, such as handling time or deflection rate, gives executives a concrete basis for evaluating the pilot against the current process."}, {"letter": "C", "text": "Present a claim that the pilot requires no evaluation period, since Claude's output quality never needs to be measured.", "correct": false, "explanation": "Skipping evaluation removes the evidence needed to demonstrate value and identify quality issues before wider rollout."}, {"letter": "D", "text": "Present the pilot as guaranteed to eliminate the entire support team's workload within the first month of deployment.", "correct": false, "explanation": "Promising complete elimination of the team's workload overstates the technology's impact and sets an unrealistic expectation."}], "correct": "B", "select": 1}, {"id": "a4-031", "domain": 4, "task_id": "4.2", "objective": "Leverage Claude for research, planning, and process optimization", "situation": "A shared services director says the invoice-processing team should be \"more efficient\" and asks an analyst to use Claude to work out how.", "question": "What should the analyst do first?", "options": [{"letter": "A", "text": "Ask Claude to list the common inefficiencies in invoice processing and check which of them apply here.", "correct": false, "explanation": "A generic list of inefficiencies gives the analyst somewhere to look but no way to prioritise. Whether a given inefficiency is worth removing depends on which measure the director cares about, which has not been settled."}, {"letter": "B", "text": "Ask the invoice team which parts of their work they would most like to stop doing altogether now.", "correct": false, "explanation": "The team's least favourite tasks are worth hearing and may be a good source of hypotheses, but they indicate irritation rather than cost. The analyst would still not know which measure of efficiency the director wants moved."}, {"letter": "C", "text": "Agree with the director which measure of efficiency matters: cycle time, cost per invoice, or errors.", "correct": true, "explanation": "Efficiency pulls in different directions: faster processing can raise error rates, and cutting cost per invoice can lengthen cycle time. Fixing which measure matters turns a vague instruction into a target the analysis can aim at and later be judged against."}, {"letter": "D", "text": "Ask Claude to redesign the invoice process end to end so the director can see what is possible.", "correct": false, "explanation": "A full redesign produced before the goal is agreed optimises for whatever the tool assumes efficiency means. It also invites the director to react to a finished proposal rather than to state the outcome being sought."}], "correct": "C", "select": 1}, {"id": "a4-032", "domain": 4, "task_id": "4.3", "objective": "Use Claude to support solution design, development, and iteration", "situation": "A workflow owner wants feedback from a four-week pilot that will genuinely help her improve the arrangement.", "question": "Which feedback format is most useful for that purpose?", "options": [{"letter": "A", "text": "Logged cases pairing the input given, the output produced, and what the user needed instead.", "correct": true, "explanation": "Improvement needs cases, not impressions. A logged triple of input, output and the required result shows exactly where the arrangement diverged from what was needed, which is the only feedback that can be turned directly into a change."}, {"letter": "B", "text": "A weekly satisfaction score from each participant, tracked to show the trend across the four weeks.", "correct": false, "explanation": "A satisfaction trend tells the owner whether things are getting better or worse but never why. A falling score gives her no defect to fix, and a rising one may simply reflect participants lowering their expectations."}, {"letter": "C", "text": "A count of how often each participant used the arrangement during each week of the pilot period.", "correct": false, "explanation": "Usage counts show adoption rather than quality. Heavy use can mean the arrangement is valuable or that people are repeatedly retrying it because the first result was unusable."}, {"letter": "D", "text": "A free-text comment box at the end of the pilot for any observations participants wish to record.", "correct": false, "explanation": "An end-of-pilot comment box collects thoughts long after the events that prompted them, so the detail has faded. It also mixes design feedback with unrelated grumbles and offers no link to specific outputs."}], "correct": "A", "select": 1}, {"id": "a4-033", "domain": 4, "task_id": "4.4", "objective": "Integrate Claude into existing workflows to augment or redesign them", "situation": "An insurer's manual claims process involves five separate teams handing a claim off through email at each stage. Leadership decides to rebuild the process so a single Claude-driven agent triages, requests documents, and routes each claim end to end.", "question": "Which term best describes this change?", "options": [{"letter": "A", "text": "Caching the workflow, since previous claim decisions are being reused for new claims", "correct": false, "explanation": "The scenario does not describe reusing prior claim decisions as cached context; it describes restructuring who performs each step."}, {"letter": "B", "text": "Batching the workflow, since claims will now be processed together instead of individually", "correct": false, "explanation": "Nothing in the scenario describes grouping claims for batch processing; the change is about who owns each step of the process."}, {"letter": "C", "text": "Redesigning the workflow, since the multi-team handoff process is being replaced by an agent-driven process", "correct": true, "explanation": "Replacing the multi-team, email-driven handoff structure with a single agent that owns triage, document requests, and routing end to end is a redesign of the workflow rather than a small addition to it."}, {"letter": "D", "text": "Augmenting the workflow, since the existing five-team handoff structure is being kept in place unchanged", "correct": false, "explanation": "The five-team handoff structure is being removed, not preserved, so this is not augmentation of the existing process."}], "correct": "C", "select": 1}, {"id": "a4-034", "domain": 4, "task_id": "4.4", "objective": "Integrate Claude into existing workflows to augment or redesign them", "situation": "A research team spends much of each week manually copying findings between four separate documents. Which two responses fit this situation best?", "question": "Select two.", "options": [{"letter": "A", "text": "Treat the copying itself as the thing to redesign, rather than speeding it up as it stands.", "correct": true, "explanation": "Duplicated copying is a symptom of process design, so this is the case where redesign beats augmentation."}, {"letter": "B", "text": "Have Claude consolidate the four documents into one source the team then maintains together.", "correct": true, "explanation": "Consolidating to a single maintained source removes the duplication instead of automating it."}, {"letter": "C", "text": "Add a fifth document in which Claude records what it copied into each of the other four weekly.", "correct": false, "explanation": "A fifth document adds another artefact to maintain and makes the duplication problem worse than before."}, {"letter": "D", "text": "Have Claude do the copying between all four documents on the team's behalf every week.", "correct": false, "explanation": "Automating the copying preserves four diverging documents and simply performs the waste faster."}, {"letter": "E", "text": "Keep all four documents and ask Claude to double the speed at which each one is updated.", "correct": false, "explanation": "Speeding up a redundant step leaves the structural problem, and four sources still drift apart from each other."}], "correct": ["A", "B"], "select": 2}, {"id": "a4-035", "domain": 4, "task_id": "4.5", "objective": "Communicate Claude's value and limitations to stakeholders", "situation": "A stakeholder wants to send an entire year of raw customer support transcripts, well beyond the model's context window, in a single request and asks why this fails.", "question": "How should this limitation be explained?", "options": [{"letter": "A", "text": "Explain that Claude rejects large inputs purely as a cost-control step, and the limit lifts once the account tier is upgraded.", "correct": false, "explanation": "The context window is a technical limit on how much input the model can process at once, not simply a billing lever tied to account tier."}, {"letter": "B", "text": "Explain that the failure means the account is rate-limited and will refuse further requests for the next 24 hours.", "correct": false, "explanation": "Exceeding the context window is a distinct failure from rate limiting and does not imply a 24-hour account block."}, {"letter": "C", "text": "Explain that transcripts must always be converted into images before Claude will process any amount of text content.", "correct": false, "explanation": "Text does not need to be converted to images to be processed; this misrepresents how context limits work."}, {"letter": "D", "text": "Explain that each model has a maximum context window, so transcripts must be chunked, summarized, or retrieved selectively.", "correct": true, "explanation": "Every model has a fixed context window, so inputs that exceed it must be broken into chunks, summarized, or narrowed via retrieval rather than sent in one request."}], "correct": "D", "select": 1}, {"id": "a4-036", "domain": 4, "task_id": "4.3", "objective": "Use Claude to support solution design, development, and iteration", "situation": "A pilot of a Claude-assisted proposal workflow has produced mixed results across its first month. Which two iteration steps are appropriate now?", "question": "Select two.", "options": [{"letter": "A", "text": "Compare the weak outputs against the strong ones to find what the weak requests were missing.", "correct": true, "explanation": "Comparing weak and strong cases diagnoses the cause instead of guessing, which is what iteration on a pilot is for."}, {"letter": "B", "text": "Roll the workflow out to every team at once so that far more data arrives much faster.", "correct": false, "explanation": "Scaling an unresolved pilot spreads the failure mode rather than diagnosing it, and it makes rollback harder."}, {"letter": "C", "text": "Abandon the pilot right away, because a mixed opening month shows the workflow cannot ever work.", "correct": false, "explanation": "Mixed results in month one are the normal input to iteration, not proof that the workflow is unworkable."}, {"letter": "D", "text": "Fold what worked into the shared instructions so the whole team gets the same improvement.", "correct": true, "explanation": "Making the fix persistent in shared instructions stops the team re-solving the same problem one person at a time."}, {"letter": "E", "text": "Ask each person to keep a private list of tricks rather than changing the shared set-up.", "correct": false, "explanation": "Private tricks leave the shared configuration untouched, so the improvement never reaches anyone else."}], "correct": ["A", "D"], "select": 2}, {"id": "a4-037", "domain": 4, "task_id": "4.3", "objective": "Use Claude to support solution design, development, and iteration", "situation": "A team is designing a Claude-assisted arrangement for handling supplier complaints.", "question": "Which step should most clearly stay with a person?", "options": [{"letter": "A", "text": "Pulling the supplier's complaint history together into a single summary", "correct": false, "explanation": "Assembling history into a summary is repetitive, text-heavy work whose output is reviewed before anyone relies on it."}, {"letter": "B", "text": "Grouping the month's complaints by the type of issue that they describe", "correct": false, "explanation": "Grouping complaints by issue type is classification work that is easy to check and carries no external commitment."}, {"letter": "C", "text": "Deciding whether a goodwill credit is offered and how large it should be", "correct": true, "explanation": "Committing the organisation's money to a supplier is a decision with external consequences that someone must be accountable for, so it stays with a person even when everything around it is delegated."}, {"letter": "D", "text": "Drafting the acknowledgement telling the supplier that work has now begun", "correct": false, "explanation": "An acknowledgement is a routine, low-stakes draft that a person can check quickly before it goes out."}], "correct": "C", "select": 1}, {"id": "a4-038", "domain": 4, "task_id": "4.3", "objective": "Use Claude to support solution design, development, and iteration", "situation": "An events manager is about to hand an end-to-end campaign setup task to Claude in Cowork for the first time, working across her own files and connected tools.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Choose the largest campaign of the quarter, so the value of the new arrangement is obvious to all.", "correct": false, "explanation": "Making the first run the biggest campaign maximises the consequences of an untested arrangement. A first outing should be chosen so that mistakes are cheap, not so that success would be impressive."}, {"letter": "B", "text": "Turn off the automatic checks so that the task can complete without interruption on its first run.", "correct": false, "explanation": "Removing the checks on a first run is the opposite of prudent. Uninterrupted completion is worth little if the manager cannot see what was done to her files and messages along the way."}, {"letter": "C", "text": "Decide which actions in the task a person must approve before they happen, and set that up.", "correct": true, "explanation": "A task that spans files and connected tools will take actions with real consequences, so the first decision is which of those a person must approve. Setting the approval points up front is what keeps the manager in control of the steps that cannot be undone."}, {"letter": "D", "text": "Ask Claude to describe how it would approach the task and keep that as the record of the plan.", "correct": false, "explanation": "A stated plan is a useful reference and worth having, but it is a description rather than a control. The task can still take an action she would not have sanctioned, because nothing stops it to ask."}], "correct": "C", "select": 1}, {"id": "a4-039", "domain": 4, "task_id": "4.2", "objective": "Leverage Claude for research, planning, and process optimization", "situation": "", "question": "Which approach best leverages Claude for prioritizing customer feedback?", "options": [{"letter": "A", "text": "Feed all feedback items to Claude with a broad instruction to prioritize them, such as 'Rank these by urgency,' without providing a scoring rubric, impact categories, or customer segments to guide the prioritization.", "correct": false, "explanation": "Providing only a broad instruction without a scoring rubric, impact categories, or customer segments leaves the prioritization arbitrary and inconsistent. Claude requires clear guidelines to produce reliable, repeatable outputs aligned with business goals."}, {"letter": "B", "text": "Categorize and cluster feedback manually into broad themes such as bugs, usability, and feature requests, then use Claude to summarize each category and extract insights that guide prioritization.", "correct": false, "explanation": "While manual categorization into broad themes is a useful first step, delegating only the summarization to Claude misses the opportunity to scale and standardize the prioritization process. Without defined prioritization criteria, the resulting insights may still lack consistent, actionable guidance."}, {"letter": "C", "text": "Directly ask Claude to evaluate feedback and decide next quarter's features, requesting a prioritized roadmap and engineering estimates, fully automating the decision without human review.", "correct": false, "explanation": "Fully automating strategic decision-making by having Claude directly decide next quarter's features and provide engineering estimates without human review introduces significant risk of unrealistic or misaligned outcomes. Such decisions require product context and human validation, as LLMs cannot replace the nuanced judgment needed for roadmap planning."}, {"letter": "D", "text": "Define clear prioritization criteria (e.g., impact, effort, customer segment), then ask Claude to categorize each feedback item and assign a priority score or rank based on that rubric.", "correct": true, "explanation": "Defining clear prioritization criteria (impact, effort, customer segment) and using Claude to categorize each feedback item against that rubric ensures a structured, objective, and repeatable prioritization. This approach produces reliable, scalable outputs that directly inform planning."}], "correct": "D", "select": 1}, {"id": "a4-040", "domain": 4, "task_id": "4.4", "objective": "Integrate Claude into existing workflows to augment or redesign them", "situation": "", "question": "In a redesigned client-reporting workflow where Claude pulls the analysis together and produces the draft, which step should most clearly remain with a person?", "options": [{"letter": "A", "text": "Deciding which chart style is used for the recurring performance section of each report.", "correct": false, "explanation": "Chart style is a house-style decision with no consequence for accuracy. It is exactly the kind of well-specified, repeatable choice that suits delegation."}, {"letter": "B", "text": "Confirming that the figures and commitments in the report going to the client are correct.", "correct": true, "explanation": "The report is an external commitment the firm will be held to, and once it is sent the claim cannot be withdrawn quietly. Verification of the figures and promises is the accountable, low-reversibility step, so it stays with the person who can answer for it."}, {"letter": "C", "text": "Choosing the order in which the report's sections are arranged for the client's convenience.", "correct": false, "explanation": "Section ordering is a presentation preference that can be specified once and applied consistently. A wrong choice is noticed easily and corrected in the next edition."}, {"letter": "D", "text": "Setting the day of the month on which the report is produced and issued to the client.", "correct": false, "explanation": "The issue date is a scheduling rule that someone sets once. It needs to be agreed, but it carries no judgment that a person must exercise each cycle."}], "correct": "B", "select": 1}, {"id": "a4-041", "domain": 4, "task_id": "4.4", "objective": "Integrate Claude into existing workflows to augment or redesign them", "situation": "A communications team's approval chain is slow, but each of its steps is required by regulation. Which two integration choices fit this situation?", "question": "Select two.", "options": [{"letter": "A", "text": "Replace the reviewers with a checklist prompt that Claude runs at the end of every draft.", "correct": false, "explanation": "A prompt cannot stand in for an accountable human approval that regulation requires a person to give."}, {"letter": "B", "text": "Route the approvals through Claude, so reviewers only see the items that Claude was unsure of.", "correct": false, "explanation": "Filtering what reviewers see means required approvals are no longer genuinely being performed on every item."}, {"letter": "C", "text": "Augment the chain as it stands by having Claude prepare each reviewer's material in advance.", "correct": true, "explanation": "When the steps must stay, augmenting them is the available gain: reviewers arrive prepared and each step takes less time."}, {"letter": "D", "text": "Keep every required approval step in place, since this constraint is legal, not procedural.", "correct": true, "explanation": "A legally required step is not a candidate for removal, so the design has to work within it rather than around it."}, {"letter": "E", "text": "Redesign the chain to drop the approval steps that Claude's drafting quality now makes needless.", "correct": false, "explanation": "Better drafts do not remove a regulatory requirement, and dropping required approvals creates a compliance failure."}], "correct": ["C", "D"], "select": 2}, {"id": "a4-042", "domain": 4, "task_id": "4.3", "objective": "Use Claude to support solution design, development, and iteration", "situation": "A support manager is designing a new way of working in which Claude drafts replies to routine tickets. Which two design decisions are sound?", "question": "Select two.", "options": [{"letter": "A", "text": "Define up front what a good draft looks like, so that drafts can be judged against that bar.", "correct": true, "explanation": "Defining what good looks like before delegating is what makes the output reviewable rather than merely plentiful."}, {"letter": "B", "text": "Measure the pilot only by how many tickets were drafted, since the volume is the whole point.", "correct": false, "explanation": "Volume with no quality measure rewards output the process was never meant to optimise for."}, {"letter": "C", "text": "Let drafts go out unreviewed once the first week of the pilot has produced no complaints.", "correct": false, "explanation": "One quiet week is not evidence of reliability, and removing review is the change most likely to surface publicly."}, {"letter": "D", "text": "Design the process so that Claude also closes each of the tickets it has drafted a reply to.", "correct": false, "explanation": "Closing tickets is an accountable outcome, and bundling it into drafting removes the checkpoint the design needs."}, {"letter": "E", "text": "Keep a human approval step before any drafted reply is sent to the customer who asked.", "correct": true, "explanation": "A customer-facing reply is externally published, so keeping a human approval step in the design is the right call."}], "correct": ["A", "E"], "select": 2}, {"id": "a4-043", "domain": 4, "task_id": "4.5", "objective": "Communicate Claude's value and limitations to stakeholders", "situation": "A trading desk stakeholder wants Claude to report the current live stock price directly from its own knowledge, without connecting any external data source.", "question": "How should this limitation be explained?", "options": [{"letter": "A", "text": "Explain that real-time data requests always trigger a permission error and there is no supported way, such as a webhook, to add live data access.", "correct": false, "explanation": "Real-time data requests do not always trigger a permission error; Claude supports tool integrations like web search or webhooks to access live data. There are supported methods to add live data access."}, {"letter": "B", "text": "Explain that stock prices are static values set at the daily close, so any past training data would reflect the official price and remain accurate today.", "correct": false, "explanation": "Stock prices are not static values set at the daily close; they change constantly during trading hours. Past training data reflects historical prices and would be outdated, not accurate today."}, {"letter": "C", "text": "Explain that Claude has no access to live, real-time data on its own, so a tool like web search must be connected to fetch current prices.", "correct": true, "explanation": "Claude has no built-in ability to access live, real-time data on its own; its knowledge is static, drawn from training data. To retrieve current stock prices, an external tool such as web search must be connected and invoked."}, {"letter": "D", "text": "Explain that Claude continuously streams live market data through a direct internal feed from major exchanges, so no external tool integration is ever required.", "correct": false, "explanation": "Claude does not continuously stream live market data internally; it has no direct feed from major exchanges. Real-time information requires integration with an external data source or tool."}], "correct": "C", "select": 1}, {"id": "a4-044", "domain": 4, "task_id": "4.4", "objective": "Integrate Claude into existing workflows to augment or redesign them", "situation": "A finance team frequently asks Claude to build the same type of formatted spreadsheet report from raw data, and wants this repeated task folded into its existing reporting workflow without re-explaining the steps every time.", "question": "Which capability best supports this?", "options": [{"letter": "A", "text": "Computer use lets Claude click through a spreadsheet application's interface for every report.", "correct": false, "explanation": "Computer Use could interact with a spreadsheet application, but it would require manually clicking through the interface each time, making it inefficient for a repeated task; it lacks the reusability of a packaged skill."}, {"letter": "B", "text": "Agent Skills packages reusable spreadsheet-building instructions and scripts for reuse.", "correct": true, "explanation": "Agent Skills allow you to encapsulate reusable instructions and scripts, enabling the finance team to define the report-building process once and invoke it repeatedly without re-explaining."}, {"letter": "C", "text": "The citations feature makes every cell in the spreadsheet link back to a source sentence.", "correct": false, "explanation": "Citations ground responses in source documents for verifiability, but they do not create links in spreadsheet cells; this feature is unrelated to packaging reusable report-building steps."}, {"letter": "D", "text": "Prompt caching caches the raw data itself instead of storing the report-building instructions.", "correct": false, "explanation": "Prompt caching is meant to reuse prompt prefixes to reduce latency and cost, not to cache raw data; it does not provide a mechanism to store and reuse report-building instructions."}], "correct": "B", "select": 1}, {"id": "a4-045", "domain": 4, "task_id": "4.2", "objective": "Leverage Claude for research, planning, and process optimization", "situation": "", "question": "Which prompt would best leverage Claude's research, planning, and process optimization capabilities to create a comprehensive biology lesson plan?", "options": [{"letter": "A", "text": "Write a biology syllabus that includes a week-by-week topic outline, grading policies with percentages, lab safety rules, and a detailed list of required materials for the semester.", "correct": false, "explanation": "This prompt focuses on syllabus logistics such as grading policies, lab safety, and materials, but omits state standards, learning objectives, activities, and differentiation. It does not fully leverage Claude's research, planning, and process optimization abilities for generating a comprehensive lesson plan as described in the documentation."}, {"letter": "B", "text": "Given state standards for high school biology, generate a week-by-week lesson plan for a 15-week semester with learning objectives, activities, assessments, and differentiation suggestions.", "correct": true, "explanation": "This prompt explicitly includes the subject (high school biology), references state standards, and requests a week-by-week 15-week plan with learning objectives, activities, assessments, and differentiation suggestions. Per Anthropic's Claude for Education documentation, Claude can generate standards-aligned lesson plans and support differentiation; standard backward design also prioritizes clear objectives, varied activities, and formative/summative assessments. By naming biology and incorporating all these elements, this prompt best leverages Claude's research, planning, and optimization capabilities for creating a comprehensive biology lesson plan."}, {"letter": "C", "text": "Create a detailed lesson plan for high school biology covering major topics with unit-by-unit learning objectives, lab activities, and formative assessments for a full year.", "correct": false, "explanation": "This prompt includes the subject (biology) and many components, but it omits alignment to state standards and differentiation suggestions. Anthropic's Claude for Education guidance emphasizes starting with state standards and including differentiation to meet diverse learners, so this prompt does not fully leverage Claude's capabilities for comprehensive curriculum planning."}, {"letter": "D", "text": "List all required high school biology topics as per state standards, organized by unit with major concepts, subtopics, hands-on activities, and formative assessments for a full-year curriculum.", "correct": false, "explanation": "This prompt aligns with state standards and covers biology content, but it does not request week-by-week planning, explicit learning objectives, or differentiation suggestions. Comprehensive lesson planning guidance from Anthropic and standard pedagogy includes week-by-week objectives and differentiation, so this prompt is incomplete for fully leveraging Claude's planning capabilities."}], "correct": "B", "select": 1}, {"id": "a4-046", "domain": 4, "task_id": "4.2", "objective": "Leverage Claude for research, planning, and process optimization", "situation": "A housing association wants to use Claude to help shorten the time it takes to allocate a repair appointment.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Describe the current steps; agree what better looks like; ask Claude where time is lost; trial one change on one patch", "correct": true, "explanation": "You cannot optimise a process you have not described, or judge a suggestion without an agreed target. Describe, agree the target, then ask for suggestions, then test one change on a small scope."}, {"letter": "B", "text": "Trial one change on one patch; ask Claude where time is lost; describe the current steps; agree what better looks like", "correct": false, "explanation": "Beginning with a trial commits the association to a change chosen at random, and leaves the target it is aiming at undefined until the end."}, {"letter": "C", "text": "Ask Claude where time is lost; describe the current steps; trial one change on one patch; agree what better looks like", "correct": false, "explanation": "Asking where time is lost before describing the process means Claude answers from generic assumptions rather than from how this association actually works."}, {"letter": "D", "text": "Agree what better looks like; trial one change on one patch; ask Claude where time is lost; describe the current steps", "correct": false, "explanation": "Trialling a change before any analysis means the change is arbitrary, and describing the current steps last makes the description pointless."}], "correct": "A", "select": 1}, {"id": "a4-047", "domain": 4, "task_id": "4.3", "objective": "Use Claude to support solution design, development, and iteration", "situation": "A charity is designing a way of working in which Claude drafts its monthly supporter update.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Draft one past month with Claude; agree what a good update contains; have the editor mark what is wrong; change one thing and try again", "correct": false, "explanation": "Drafting before agreeing the standard means the editor's marks are personal preference, and the design converges on one person's taste."}, {"letter": "B", "text": "Draft one past month with Claude; have the editor mark what is wrong; change one thing and try again; agree what a good update contains", "correct": false, "explanation": "Leaving the definition of a good update until last means the first three steps were judged against nobody's stated standard."}, {"letter": "C", "text": "Agree what a good update contains; draft one past month with Claude; have the editor mark what is wrong; change one thing and try again", "correct": true, "explanation": "Agree the standard, produce a draft on a known month, have it marked against that standard, then change one thing at a time. Each round then tells you something you can act on."}, {"letter": "D", "text": "Agree what a good update contains; have the editor mark what is wrong; draft one past month with Claude; change one thing and try again", "correct": false, "explanation": "Marking what is wrong before anything has been drafted is impossible; there is no draft to mark."}], "correct": "C", "select": 1}, {"id": "a4-048", "domain": 4, "task_id": "4.4", "objective": "Integrate Claude into existing workflows to augment or redesign them", "situation": "A director tells a service manager to \"roll Claude out across the department next quarter.\" The department runs eleven distinct workflows of varying sensitivity.", "question": "What should the service manager do first?", "options": [{"letter": "A", "text": "Ask Claude to review all eleven workflows and recommend which of them to change first.", "correct": false, "explanation": "Claude can help compare the workflows once someone describes them, but it cannot know which of them the department most needs to fix or which carry unacceptable risk. Handing the prioritisation over skips the manager's own judgment."}, {"letter": "B", "text": "Draft a departmental policy covering how Claude may be used in each of the eleven workflows.", "correct": false, "explanation": "A usage policy will be needed eventually, particularly for the sensitive workflows. Writing it for all eleven before anything has been attempted means guessing at the practices it should govern."}, {"letter": "C", "text": "Give every team access at the start of the quarter so that people can begin finding uses for it.", "correct": false, "explanation": "Broad access with no chosen target produces scattered experiments across eleven workflows, including the sensitive ones. At the end of the quarter the manager will have activity to report but no evidence of improvement."}, {"letter": "D", "text": "Choose one workflow and one outcome to improve, and agree how that will be measured.", "correct": true, "explanation": "A vague rollout instruction becomes deliverable by narrowing it to one workflow and one measurable outcome. That gives a bounded first integration, a clear basis for judging whether it worked, and evidence to take back to the director before spreading further."}], "correct": "D", "select": 1}, {"id": "a4-049", "domain": 4, "task_id": "4.4", "objective": "Integrate Claude into existing workflows to augment or redesign them", "situation": "A hospital wants to integrate Claude into its clinical documentation workflow to help physicians summarize patient charts, where accuracy is far more important than response speed or cost.", "question": "Which starting approach best fits this integration?", "options": [{"letter": "A", "text": "Start with a model selected only for its support of the newest client-side tools", "correct": false, "explanation": "Client-side tool support is unrelated to the summarization accuracy this clinical documentation workflow needs."}, {"letter": "B", "text": "Start with the most capable model, since accuracy outweighing cost considerations favors this choice", "correct": true, "explanation": "Applications where accuracy outweighs cost considerations, such as clinical documentation, are a case for starting with the most capable model rather than a faster, cheaper one."}, {"letter": "C", "text": "Start with the fastest, cheapest model, since most patient charts are short and simple to summarize", "correct": false, "explanation": "Prioritizing speed and low cost does not match a scenario where the stated priority is accuracy over speed or cost."}, {"letter": "D", "text": "Start with a model chosen for its low output token pricing, since pricing is the deciding factor here", "correct": false, "explanation": "Optimizing for output token pricing contradicts the scenario's explicit priority of accuracy over cost."}], "correct": "B", "select": 1}, {"id": "a4-050", "domain": 4, "task_id": "4.1", "objective": "Apply Claude to analyze requirements and use cases", "situation": "A transformation office asks Claude to score two hundred candidate use cases on effort and value from their one-paragraph descriptions, and plans to fund the top twenty exactly as ranked.", "question": "What is the greatest concern?", "options": [{"letter": "A", "text": "The scored list will need reformatting before it can be presented to the executive funding committee.", "correct": false, "explanation": "Reformatting for a committee is routine presentation work. It changes nothing about the reliability of the ranking that the committee would be asked to approve."}, {"letter": "B", "text": "Scoring two hundred use cases in one conversation may exceed what a single chat can comfortably hold.", "correct": false, "explanation": "Volume is a practical constraint that can be handled by working through the list in batches. It affects how the scoring is run, not whether a funding decision should be handed to the scores that come out of it."}, {"letter": "C", "text": "A ranking drawn from short summaries can look authoritative while resting on assumptions nobody checked.", "correct": true, "explanation": "A one-paragraph description does not contain the effort and value evidence the scores imply, so Claude must infer it. The numbers then carry an air of objectivity that a spreadsheet reinforces, and funding follows the ranking without anyone testing the assumptions underneath it."}, {"letter": "D", "text": "Some of the two hundred descriptions were written by staff who are quite unfamiliar with what Claude can do.", "correct": false, "explanation": "Uneven familiarity with the tool affects how well individual submissions are phrased, and it is worth correcting, but it is a quality-of-input issue that sits well below the danger of treating an inferred ranking as a funding decision."}], "correct": "C", "select": 1}, {"id": "a4-051", "domain": 4, "task_id": "4.4", "objective": "Integrate Claude into existing workflows to augment or redesign them", "situation": "A bank wants Claude to prepare a recommendation for each loan application, but a compliance officer must still approve or reject every decision before it takes effect.", "question": "Which integration pattern does this describe?", "options": [{"letter": "A", "text": "A human-in-the-loop pattern where Claude assists with a recommendation and a person retains final decision authority", "correct": true, "explanation": "Requiring a compliance officer to approve or reject every Claude-generated recommendation before it takes effect is a human-in-the-loop integration, where Claude augments the decision process but a person retains authority."}, {"letter": "B", "text": "A tool-use pattern where Claude directly updates the core banking ledger as soon as it drafts a recommendation", "correct": false, "explanation": "Nothing in the scenario has Claude writing directly to the ledger; the compliance officer's approval step sits between the recommendation and any action."}, {"letter": "C", "text": "A fully autonomous pattern where Claude both generates and finalizes the loan decision without further review", "correct": false, "explanation": "The scenario explicitly keeps a human approval step, so the decision is not finalized autonomously by Claude."}, {"letter": "D", "text": "A batch-only pattern where loan applications are processed in bulk once per week with no live interaction", "correct": false, "explanation": "The scenario does not describe a weekly bulk process; it describes ongoing review of individual recommendations."}], "correct": "A", "select": 1}, {"id": "a4-052", "domain": 4, "task_id": "4.4", "objective": "Integrate Claude into existing workflows to augment or redesign them", "situation": "A team is deciding which steps of its onboarding process to hand to Claude and which to keep. Which two steps should stay with a person?", "question": "Select two.", "options": [{"letter": "A", "text": "The final decision on whether a new hire has passed the probation review for this quarter.", "correct": true, "explanation": "A probation outcome affects someone's employment, so the decision needs the accountability only a person can carry."}, {"letter": "B", "text": "The conversation with a new joiner about a personal difficulty raised in their first week.", "correct": true, "explanation": "A personal difficulty raised by a colleague calls for human judgment and duty of care, not a delegated reply."}, {"letter": "C", "text": "Drafting the standard welcome message that each of the new joiners receives on their first day.", "correct": false, "explanation": "A standard welcome message is a low-stakes, repetitive draft, which is a good candidate for delegation."}, {"letter": "D", "text": "Turning the existing onboarding handbook into a shorter checklist for the new joiners.", "correct": false, "explanation": "Condensing an existing handbook is well-specified text work that a person can review before it is published."}, {"letter": "E", "text": "Summarising last year's onboarding survey feedback into a set of recurring themes.", "correct": false, "explanation": "Theming survey feedback is high-volume analysis whose output the team can check against the underlying responses."}], "correct": ["A", "B"], "select": 2}, {"id": "a4-053", "domain": 4, "task_id": "4.3", "objective": "Use Claude to support solution design, development, and iteration", "situation": "To win support for a new Claude-assisted proposal-drafting arrangement, a team demonstrates the best output from about twenty attempts and does not mention the other nineteen.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The demonstration may run longer than the slot the sponsors set aside in the meeting agenda.", "correct": false, "explanation": "Overrunning the agenda is an inconvenience for the meeting chair. It is unrelated to the honesty of what the demonstration implies about everyday performance."}, {"letter": "B", "text": "The twenty attempts consumed usage that the team had allocated to its ordinary client work.", "correct": false, "explanation": "Usage spent preparing a demonstration is a minor and one-off cost. It affects the team's own budget rather than the decision the sponsors are about to make."}, {"letter": "C", "text": "The team will have to repeat the demonstration for other departments that were not present.", "correct": false, "explanation": "Repeating a demonstration costs a little time and is entirely normal when several departments are interested. It has no bearing on whether the expectations being set are accurate."}, {"letter": "D", "text": "Sponsors will fund the arrangement expecting a standard the everyday output does not reach.", "correct": true, "explanation": "Showing the best of twenty sets the expected quality at a level the arrangement reaches roughly one time in twenty. Funding, staffing and review effort are then planned against a standard that will not materialise, and the credibility loss lands on the team."}], "correct": "D", "select": 1}, {"id": "a4-054", "domain": 4, "task_id": "4.5", "objective": "Communicate Claude's value and limitations to stakeholders", "situation": "A stakeholder wants Claude to answer questions about their company's internal product catalog, which was updated last week and was never part of Claude's training data.", "question": "How should this limitation and solution be communicated?", "options": [{"letter": "A", "text": "Explain that the catalog only needs to be emailed to Anthropic once so it joins the base model going forward.", "correct": false, "explanation": "Sending data to Anthropic does not add it to the base model’s training; to keep answers current with recent changes, retrieval-based or tool-based approaches must be used."}, {"letter": "B", "text": "Explain that Claude cannot process proprietary data, so supplying the catalog via retrieval or a connected tool will not help.", "correct": false, "explanation": "Claude can effectively process proprietary data when it is supplied in the prompt or via retrieval; the limitation is a lack of prior knowledge, not an inability to use supplied information."}, {"letter": "C", "text": "Explain that Claude automatically indexes any company website within days, so no integration work is ever required.", "correct": false, "explanation": "Claude does not automatically crawl, index, or learn from any company website; integration work is always required to supply it with proprietary or recently updated data."}, {"letter": "D", "text": "Explain that Claude lacks internal data knowledge, so the catalog must be supplied via retrieval or a connected tool.", "correct": true, "explanation": "Claude lacks built-in knowledge of internal company data, so the updated product catalog must be provided through retrieval mechanisms like RAG or a connected tool for accurate answers."}], "correct": "D", "select": 1}, {"id": "a4-055", "domain": 4, "task_id": "4.5", "objective": "Communicate Claude's value and limitations to stakeholders", "situation": "A sponsor wants to deploy a new Claude-powered feature to all users in every region simultaneously on day one.", "question": "What rollout approach should be recommended instead, and why?", "options": [{"letter": "A", "text": "Recommend delaying the launch indefinitely, since no rollout strategy can reduce risk for AI-powered features.", "correct": false, "explanation": "Indefinite delay is unnecessary; a phased approach allows launch to proceed while managing risk."}, {"letter": "B", "text": "Recommend the simultaneous global launch as proposed, since phased rollouts add no measurable risk-reduction benefit.", "correct": false, "explanation": "Phased rollouts are a standard risk-reduction practice; claiming no benefit contradicts common deployment guidance."}, {"letter": "C", "text": "Recommend a phased rollout to a small user segment first, so issues can be caught and addressed before scaling to everyone.", "correct": true, "explanation": "A phased or pilot rollout to a limited audience lets the team observe real-world performance and fix issues before exposing the full user base to them, reducing overall risk."}, {"letter": "D", "text": "Recommend launching only in the region with the least usage data, since regional differences do not affect quality.", "correct": false, "explanation": "Regional usage patterns and data characteristics can affect performance, so ignoring them is not sound rollout planning."}], "correct": "C", "select": 1}, {"id": "a4-056", "domain": 4, "task_id": "4.5", "objective": "Communicate Claude's value and limitations to stakeholders", "situation": "A team lead must brief a sceptical director on a new Claude-assisted reporting arrangement.", "question": "Which framing is most appropriate?", "options": [{"letter": "A", "text": "Say that results will improve automatically as newer models become available", "correct": false, "explanation": "Promising automatic future improvement asks the director to accept an outcome nobody can commit to."}, {"letter": "B", "text": "Say that the pilot's best output shows what every report will look like now", "correct": false, "explanation": "Presenting the best single output as typical sets an expectation the arrangement will not meet in ordinary use."}, {"letter": "C", "text": "Say that the arrangement removes the need for the team's current review step", "correct": false, "explanation": "Claiming the review step is unnecessary oversells the arrangement and removes the safeguard the director is most likely to ask about."}, {"letter": "D", "text": "Say what it does reliably, what needs checking, and who does the checking", "correct": true, "explanation": "Honest communication pairs capability with limitation and names where human review stays in the loop. That is what lets a sceptical stakeholder judge the arrangement rather than the pitch."}], "correct": "D", "select": 1}, {"id": "a4-057", "domain": 4, "task_id": "4.4", "objective": "Integrate Claude into existing workflows to augment or redesign them", "situation": "An events team has redesigned its exhibitor-liaison workflow so that Claude carries a request from first enquiry through to a confirmed booking slot.", "question": "Where does an approval checkpoint most clearly belong?", "options": [{"letter": "A", "text": "Only if an exhibitor complains about the slot they were allocated", "correct": false, "explanation": "Waiting for a complaint makes the exhibitor the quality control, which is precisely what a checkpoint is meant to avoid."}, {"letter": "B", "text": "After the exhibitor has replied to the confirmation that was sent", "correct": false, "explanation": "A check after the exhibitor has replied comes after the commitment, so it can only manage the consequences."}, {"letter": "C", "text": "At the end of each month, when all bookings are reviewed together", "correct": false, "explanation": "A monthly review catches patterns but arrives long after individual bookings have gone out."}, {"letter": "D", "text": "Before anything is sent to the exhibitor on the team's behalf", "correct": true, "explanation": "The checkpoint belongs at the last moment where a person can still prevent an external, hard-to-retract action. Once the exhibitor has been contacted, the commitment has been made."}], "correct": "D", "select": 1}, {"id": "a4-058", "domain": 4, "task_id": "4.5", "objective": "Communicate Claude's value and limitations to stakeholders", "situation": "A stakeholder wants Claude to read scanned engineering drawings and return exact millimeter measurements for manufacturing tolerances.", "question": "How should the solutions architect set expectations about this vision use case?", "options": [{"letter": "A", "text": "Confirm that Claude's vision accuracy is unaffected by image resolution or scan quality, ensuring consistent measurements across all scenarios.", "correct": false, "explanation": "Image resolution and scan quality can affect how accurately Claude interprets visual details, so its vision accuracy is not independent of these factors."}, {"letter": "B", "text": "Confirm that Claude can read pixel-level measurements directly from any scanned drawing, ensuring guaranteed manufacturing-grade precision.", "correct": false, "explanation": "Claude does not guarantee pixel-level measurements with manufacturing-grade precision, which overstates its vision capabilities for this use case."}, {"letter": "C", "text": "Explain that Claude lacks image processing capabilities, so drawings must be manually transcribed into text by staff before analysis.", "correct": false, "explanation": "Claude does support image input and can reason about visual content, so manual transcription into text is not required; this understates its actual capabilities."}, {"letter": "D", "text": "Explain that Claude can describe and reason about visual content, but exact measurements should be verified with dedicated tools.", "correct": true, "explanation": "Claude can describe and reason about visual content, but it is not a calibrated measurement tool; precise tolerance values should be verified with dedicated measurement instruments before use in manufacturing."}], "correct": "D", "select": 1}, {"id": "a4-059", "domain": 4, "task_id": "4.2", "objective": "Leverage Claude for research, planning, and process optimization", "situation": "A consultancy team on an Enterprise plan wants Claude's research on a client sector to draw on the firm's own past engagement reports, not only on public material.", "question": "What is the most appropriate approach?", "options": [{"letter": "A", "text": "Paste the single most relevant engagement report into each new chat whenever it seems likely to be needed.", "correct": false, "explanation": "Relying on someone to notice the right report and paste it in means the internal evidence is used inconsistently. It also limits each piece of research to one report when the value lies in the pattern across many."}, {"letter": "B", "text": "Put the past engagement reports into the project so every research chat in it works from that material.", "correct": true, "explanation": "Project knowledge is durable background for every chat in that project, so the firm's own engagement history informs each piece of research without anyone re-pasting it. That makes the internal evidence a standing input rather than something remembered on a good day."}, {"letter": "C", "text": "Ask Claude to describe what a firm like this one would probably have learned in past engagements.", "correct": false, "explanation": "Asking Claude what such a firm would probably have learned invites it to invent the firm's experience. The reports exist and can be supplied, so guessing at their contents is the one option that produces confident fiction."}, {"letter": "D", "text": "Summarise the past engagement reports into a single page and rely on that page from then onwards.", "correct": false, "explanation": "A single summary page discards the detail that makes past engagements useful, and it becomes stale the moment new work is completed. The team would be researching against a snapshot rather than the record."}], "correct": "B", "select": 1}, {"id": "a4-060", "domain": 4, "task_id": "4.4", "objective": "Integrate Claude into existing workflows to augment or redesign them", "situation": "A legal team wants to fold Claude into its existing contract-review workflow so that an entire multi-hundred-page merger agreement can be analyzed in a single request instead of being split into many smaller calls.", "question": "Which capability makes this integration possible?", "options": [{"letter": "A", "text": "Batch processing, which lets many small requests run overnight instead of one large request", "correct": false, "explanation": "Batch processing helps with running many separate requests asynchronously; it does not let a single request hold more content than the context window allows."}, {"letter": "B", "text": "The effort parameter, which reduces the number of tokens needed to represent the agreement", "correct": false, "explanation": "The effort parameter trades off response thoroughness for token efficiency in Claude's output; it does not change how much input content a single request can hold."}, {"letter": "C", "text": "Adaptive thinking, which shortens a long agreement before any of it is read by Claude", "correct": false, "explanation": "Adaptive thinking controls how Claude reasons about a request; it does not shrink or summarize the agreement before Claude reads it."}, {"letter": "D", "text": "A large context window, which lets a single request hold the entire lengthy agreement at once", "correct": true, "explanation": "A large context window allows a single request to hold large documents, extensive code bases, or long conversations, which is what lets the entire lengthy merger agreement be analyzed in one request."}], "correct": "D", "select": 1}, {"id": "a4-061", "domain": 4, "task_id": "4.1", "objective": "Apply Claude to analyze requirements and use cases", "situation": "Four analysts on an Enterprise plan are running separate chats to analyse requirements for the same programme, and each keeps re-pasting the same brief, glossary and stakeholder map.", "question": "What is the most appropriate improvement?", "options": [{"letter": "A", "text": "Ask each analyst to paste a shortened summary of the documents rather than the full versions each time.", "correct": false, "explanation": "Shortened summaries mean each analyst works from a different reduction of the same material, which is how definitions quietly diverge. Requirements analysis depends on everyone reading the same glossary, not four paraphrases of it."}, {"letter": "B", "text": "Agree that a single analyst holds the source documents and answers the other three analysts on request each time.", "correct": false, "explanation": "Funnelling the documents through one person creates a bottleneck and leaves the other three analysts analysing from second-hand descriptions. It solves the duplication by removing access rather than by sharing context."}, {"letter": "C", "text": "Merge the four conversations into one long chat so the shared material has to be pasted only once.", "correct": false, "explanation": "One very long shared conversation mixes four unrelated lines of analysis, becomes hard to follow, and pushes against the limits of a single chat. It also gives no way to start a fresh piece of analysis from the same background."}, {"letter": "D", "text": "Put the brief, glossary and stakeholder map into a shared project so every analysis chat starts from them.", "correct": true, "explanation": "Recurring work over a stable set of source material is exactly what a project is for: the brief, glossary and stakeholder map become durable background for every chat in that project, so all four analysts work from the same current version instead of from whatever each last pasted."}], "correct": "D", "select": 1}, {"id": "a4-062", "domain": 4, "task_id": "4.3", "objective": "Use Claude to support solution design, development, and iteration", "situation": "A grants team is designing a new way of working in which Claude prepares the first draft of each funding decision summary. Nothing has been built yet.", "question": "What should the team do first?", "options": [{"letter": "A", "text": "Agree what an acceptable summary contains, so drafts can be judged against it", "correct": true, "explanation": "Defining what \"good\" looks like before delegating is what makes every later round of iteration meaningful; without it, nobody can say whether a draft is acceptable or whether a change helped."}, {"letter": "B", "text": "Announce the new arrangement so that people begin sending their cases through", "correct": false, "explanation": "Announcing an arrangement that has not been designed pushes unfinished work onto colleagues and makes problems harder to withdraw from."}, {"letter": "C", "text": "Write the longest and most detailed instruction the team can manage in one go", "correct": false, "explanation": "Instruction length is not the design. A long instruction written before the team knows what it wants encodes guesses at scale."}, {"letter": "D", "text": "Ask every member of the team to try their own approach and compare next month", "correct": false, "explanation": "Four private approaches produce four private standards, and a month later there is still no agreed standard to compare them against."}], "correct": "A", "select": 1}, {"id": "a4-063", "domain": 4, "task_id": "4.5", "objective": "Communicate Claude's value and limitations to stakeholders", "situation": "A manager asks whether the new Claude-assisted process means the team can cut its review budget. Which two responses are honest and useful?", "question": "Select two.", "options": [{"letter": "A", "text": "Share what the pilot actually changed in the cycle time, and what it did not change at all.", "correct": true, "explanation": "Reporting both what improved and what did not is what sets an expectation the rollout can actually meet."}, {"letter": "B", "text": "Confirm that the review budget can be cut right now, since the pilot produced no visible errors.", "correct": false, "explanation": "Absence of visible errors in a short pilot is weak evidence, and cutting review on it removes the safeguard."}, {"letter": "C", "text": "Say that the process is fully reliable, so the manager's confidence in it is not undermined.", "correct": false, "explanation": "Claiming full reliability is an overstatement, and confidence built on it collapses at the first real error."}, {"letter": "D", "text": "Explain that review is still needed, because outputs continue to need checking before release.", "correct": true, "explanation": "Review remains part of the design, and saying so plainly is the honest answer to a budget question."}, {"letter": "E", "text": "Avoid the question, since discussing limitations now would weaken support for the rollout.", "correct": false, "explanation": "Withholding limitations to protect support is exactly the communication failure this competency warns against."}], "correct": ["A", "D"], "select": 2}, {"id": "a4-064", "domain": 4, "task_id": "4.2", "objective": "Leverage Claude for research, planning, and process optimization", "situation": "An operations team wants Claude to help find where a twelve-step expense reimbursement process loses time.", "question": "Which input will produce the most useful analysis?", "options": [{"letter": "A", "text": "The actual step list with who does each step, the handoffs between them, and typical waiting times.", "correct": true, "explanation": "Delay in a multi-step process almost always sits in the handoffs and the waiting between steps, so the analysis is only as good as the map it is given. Supplying the real steps, owners and typical waits lets Claude point at specific queues instead of generalising."}, {"letter": "B", "text": "A summary of the process written for new starters, which explains its purpose and the policy behind it.", "correct": false, "explanation": "A new-starter summary explains why the process exists and what policy governs it. That is the wrong level of detail for finding delay, because it deliberately omits the handoffs and waiting that create it."}, {"letter": "C", "text": "A description of how the team feels the process performs, written by the team's most experienced approver.", "correct": false, "explanation": "An experienced approver's impression is a useful hypothesis but not evidence. Perceived slowness usually clusters around visible effort rather than around the queues where time is genuinely lost."}, {"letter": "D", "text": "A list of the complaints employees have raised about reimbursement over the past two reporting quarters.", "correct": false, "explanation": "Complaints record where employees felt the pain, which is often the last step before an outcome rather than the step that caused the wait. They are a helpful cross-check but a poor primary input for locating a bottleneck."}], "correct": "A", "select": 1}, {"id": "a4-065", "domain": 4, "task_id": "4.3", "objective": "Use Claude to support solution design, development, and iteration", "situation": "A finance team is designing a way of working in which Claude prepares the monthly variance commentary.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Write the instruction for the rest; try it on last month's figures; decide which parts a person must own; adjust one thing and try again", "correct": false, "explanation": "Writing instructions before deciding what stays human means the instruction may quietly cover work that should never have been handed over."}, {"letter": "B", "text": "Try it on last month's figures; decide which parts a person must own; write the instruction for the rest; adjust one thing and try again", "correct": false, "explanation": "Testing before the boundary or the instruction exists gives a result nobody can interpret, and the design decisions come too late to shape it."}, {"letter": "C", "text": "Decide which parts a person must own; write the instruction for the rest; try it on last month's figures; adjust one thing and try again", "correct": true, "explanation": "Delegation boundaries come first: decide what a person must own, write instructions for the remainder, test on real past figures, then adjust one thing at a time."}, {"letter": "D", "text": "Write the instruction for the rest; decide which parts a person must own; adjust one thing and try again; try it on last month's figures", "correct": false, "explanation": "Adjusting before the arrangement has been tried at all means the first adjustment is made blind, and the test arrives after the changes."}], "correct": "C", "select": 1}, {"id": "a4-066", "domain": 4, "task_id": "4.4", "objective": "Integrate Claude into existing workflows to augment or redesign them", "situation": "A shared services group wants to take a Claude-assisted purchase-request triage from an idea to everyday use across eight teams.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Extend to the other teams with the checks that proved necessary; pick one team and a bounded slice of requests; run it alongside the existing process for a month; compare the two sets of results and fix what failed.", "correct": false, "explanation": "Extending to every team first is a full rollout before any evidence exists, which makes the later pilot and comparison pointless because the risk has already been taken."}, {"letter": "B", "text": "Run it alongside the existing process for a month; compare the two sets of results and fix what failed; pick one team and a bounded slice of requests; extend to the other teams with the checks that proved necessary.", "correct": false, "explanation": "Running alongside the existing process before choosing a team and a slice means the trial has no defined scope, and the scoping decision arrives after the comparison it was supposed to make possible."}, {"letter": "C", "text": "Pick one team and a bounded slice of requests; run it alongside the existing process for a month; compare the two sets of results and fix what failed; extend to the other teams with the checks that proved necessary.", "correct": true, "explanation": "A bounded slice on one team keeps early mistakes cheap, running alongside the existing process gives a fair comparison rather than an impression, fixing what failed happens while the scope is still small, and the wider rollout then carries the human checks the comparison showed were needed."}, {"letter": "D", "text": "Compare the two sets of results and fix what failed; pick one team and a bounded slice of requests; extend to the other teams with the checks that proved necessary; run it alongside the existing process for a month.", "correct": false, "explanation": "There are no results to compare at the start, and the parallel run that would have produced them is left until after the arrangement has already been extended to all eight teams."}], "correct": "C", "select": 1}, {"id": "a4-067", "domain": 4, "task_id": "4.1", "objective": "Apply Claude to analyze requirements and use cases", "situation": "A charity is scoping a use case in which Claude would decide which grant applications are rejected at first screening, removing the review panel's initial pass entirely.", "question": "What is the greatest concern with this use case?", "options": [{"letter": "A", "text": "The screening criteria may be worded differently across the charity's several grant programmes.", "correct": false, "explanation": "Inconsistent criteria across programmes is a real quality problem, and it is worth fixing, but it can be corrected by harmonising the wording. It does not touch the question of who is answerable for a rejection."}, {"letter": "B", "text": "The number of applications in each round may be larger than one conversation can work through.", "correct": false, "explanation": "Volume is a workload constraint that can be handled by processing applications in batches. It affects how the screening would be run rather than whether the decision should be delegated at all."}, {"letter": "C", "text": "Panel members may need extra training before they can interpret the reasoning given for each decision.", "correct": false, "explanation": "Training panel members to read the reasoning would help, but it assumes the panel still reviews the decisions. In the use case as described the panel's initial pass has been removed, so there is nothing left for them to interpret."}, {"letter": "D", "text": "Rejection at first screening is not seen or appealed by anyone, so no person owns the decision made.", "correct": true, "explanation": "A first-pass rejection is effectively final for the applicant and no human sees it, so an error is neither visible nor recoverable and nobody is accountable for it. Work carrying that kind of consequence should keep a person in the decision, with Claude preparing the assessment instead."}], "correct": "D", "select": 1}, {"id": "a4-068", "domain": 4, "task_id": "4.1", "objective": "Apply Claude to analyze requirements and use cases", "situation": "A charity's operations analyst asks Claude to produce a requirements list for a new volunteer scheduling process, giving it only the notes from the two coordinators who happened to be available that week. The list reads as thorough and is circulated as the agreed requirements.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "Claude will refuse to produce a requirements list from notes covering part of the service", "correct": false, "explanation": "Claude will work with whatever it is given; it does not withhold an answer because the input covers only part of the service."}, {"letter": "B", "text": "The two coordinators will be named in the list, exposing internal comments to everyone", "correct": false, "explanation": "Attribution is a housekeeping matter and easily corrected; it does not affect whether the requirements themselves are real."}, {"letter": "C", "text": "The list will present gaps in the source notes as settled requirements nobody stated", "correct": true, "explanation": "Claude will fill an incomplete picture with plausible, well-written material. Circulated as \"the agreed requirements\", inferred content becomes indistinguishable from what stakeholders actually asked for."}, {"letter": "D", "text": "The list will be too long for the sponsor to read before the decision has to be made", "correct": false, "explanation": "Length is a presentation problem. A concise list built on the same partial input would carry exactly the same defect."}], "correct": "C", "select": 1}, {"id": "a4-069", "domain": 4, "task_id": "4.1", "objective": "Apply Claude to analyze requirements and use cases", "situation": "", "question": "What is the most effective way to use Claude to analyze and categorize a list of requirements?", "options": [{"letter": "A", "text": "Ask Claude to design a new spreadsheet layout for organizing the requirements, including columns for type and priority.", "correct": false, "explanation": "Designing a spreadsheet layout focuses on formatting rather than analyzing or categorizing requirements. It does not leverage Claude's ability to classify requirements but instead adds a workflow artifact that doesn't directly address the need for analysis."}, {"letter": "B", "text": "Ask Claude to review each requirement for completeness, clarity, and testability without performing classification.", "correct": false, "explanation": "This approach only evaluates quality attributes but does not categorize requirements, which is the primary goal of analysis. It fails to leverage Claude's classification capabilities, missing the chance to organize and understand requirements effectively."}, {"letter": "C", "text": "Ask Claude to remove fifty percent of the requirements at random to reduce workload, simplifying the review process.", "correct": false, "explanation": "Randomly removing requirements is not a valid analysis or categorization technique and risks losing essential information. It compromises the completeness of the requirements set and fails to achieve proper classification."}, {"letter": "D", "text": "Ask Claude to classify each requirement as functional, non-functional, or constraint with explanations.", "correct": true, "explanation": "This directly uses Claude's strength in analysis and classification by categorizing requirements into functional, non-functional, or constraint, along with explanations. It enhances clarity, traceability, and follows best practices for requirements analysis."}], "correct": "D", "select": 1}, {"id": "a4-070", "domain": 4, "task_id": "4.1", "objective": "Apply Claude to analyze requirements and use cases", "situation": "A product manager is using Claude to help define the minimum viable product (MVP) for a new feature.", "question": "Which approach should they use?", "options": [{"letter": "A", "text": "Ask Claude to prioritize features considering business value, feasibility, and strategic alignment.", "correct": true, "explanation": "Prioritizing features based on business value, feasibility, and strategic alignment ensures the MVP delivers the most impact with the least risk. This structured approach helps define a lean, focused product that addresses key user needs while remaining realistic to build."}, {"letter": "B", "text": "Ask Claude to randomly pick features to include in the MVP without any criteria, by random selection from the list.", "correct": false, "explanation": "Randomly picking features without criteria does not create a principled MVP and can lead to a misaligned product that fails to address core user needs or business goals. This approach lacks structure and strategic thinking."}, {"letter": "C", "text": "Ask Claude to generate even more feature ideas to expand the list for the MVP, then pick the most requested ones.", "correct": false, "explanation": "Expanding the feature list for the MVP goes against the principle of minimalism and focus. An MVP should be lean and targeted, not bloated with additional ideas. The key task is prioritization and selection, not ideation."}, {"letter": "D", "text": "Ask Claude to keep only features endorsed by the most senior stakeholder, using their approval as the deciding factor.", "correct": false, "explanation": "Relying solely on the most senior stakeholder's endorsement can overlook critical user needs, technical constraints, and broader business objectives. It introduces bias and reduces objectivity. Good prioritization should be evidence-based and cross-functional."}], "correct": "A", "select": 1}, {"id": "a4-071", "domain": 4, "task_id": "4.4", "objective": "Integrate Claude into existing workflows to augment or redesign them", "situation": "A team adds Claude to the middle of a workflow that already produces frequent rework, without changing anything else about it.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "Claude will refuse to take part in a workflow that has not first been documented", "correct": false, "explanation": "Claude does not require a workflow to be documented before it will contribute to a step."}, {"letter": "B", "text": "The team will need a different plan tier before Claude can join an existing flow", "correct": false, "explanation": "Plan tier is not what determines whether an existing process can be augmented, and it would not cause rework."}, {"letter": "C", "text": "Colleagues downstream will receive their work later than the old workflow gave it", "correct": false, "explanation": "The likely effect is faster delivery, not slower; the problem is what is being delivered, not when."}, {"letter": "D", "text": "The same faults now appear faster and in greater volume than they did before", "correct": true, "explanation": "Speeding up a broken process scales its defects. The rework was caused by the workflow's design, and inserting Claude without addressing that simply produces more of it, sooner."}], "correct": "D", "select": 1}, {"id": "a4-072", "domain": 4, "task_id": "4.2", "objective": "Leverage Claude for research, planning, and process optimization", "situation": "A procurement lead asks Claude to recommend one of three suppliers, giving it only the brochures and web copy that each supplier provided.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The comparison can only reflect what each supplier says about itself, with no independent evidence.", "correct": true, "explanation": "Every input is written by the party being assessed, so the comparison measures marketing quality rather than supplier quality. The output will read as a balanced analysis while containing no reference checks, pricing evidence or performance history at all."}, {"letter": "B", "text": "Comparing three suppliers at once may produce an answer that is too long for the committee pack.", "correct": false, "explanation": "Length can be controlled by asking for a shorter comparison or a summary table. A recommendation that is too long is inconvenient; one built on marketing claims is misleading."}, {"letter": "C", "text": "The recommendation may be written in a tone that is too informal for a procurement committee.", "correct": false, "explanation": "Tone is adjusted by asking for a different register. It affects how the recommendation reads to the committee, not whether the underlying comparison is sound."}, {"letter": "D", "text": "The brochures may each use different section headings, which makes the comparison harder to lay out neatly.", "correct": false, "explanation": "Inconsistent headings make the material messier to align and may need a little manual mapping. This is a presentation problem rather than a defect in the evidence base."}], "correct": "A", "select": 1}, {"id": "a4-073", "domain": 4, "task_id": "4.1", "objective": "Apply Claude to analyze requirements and use cases", "situation": "A stakeholder has submitted the requirement \"make monthly reporting easier.\" An analyst wants Claude to help prepare for the next stakeholder meeting.", "question": "Which use of Claude is most appropriate here?", "options": [{"letter": "A", "text": "Ask Claude to rewrite the requirement in formal language so it can be entered into the requirements log.", "correct": false, "explanation": "Formalising the language makes a vague requirement look rigorous without adding any information. The log then contains a polished statement that still cannot be built or tested, and the ambiguity survives into design."}, {"letter": "B", "text": "Ask Claude to describe how comparable organisations define easier reporting and adopt that wording.", "correct": false, "explanation": "Adopting an external definition replaces this stakeholder's need with a generic one. It also produces confident-sounding wording that nobody in the organisation has agreed to, which makes the ambiguity harder to spot later, not easier."}, {"letter": "C", "text": "Ask Claude to decide which reading the stakeholder most likely meant, and then build the entire delivery plan on it.", "correct": false, "explanation": "Letting Claude pick the most likely reading converts an open question into a hidden assumption. The stakeholder is available to answer, so guessing on their behalf trades a cheap clarification for an expensive misunderstanding."}, {"letter": "D", "text": "Ask Claude to produce several plausible readings of the requirement plus questions to put to the stakeholder.", "correct": true, "explanation": "Ambiguity is resolved by the stakeholder, not by the tool. Claude is best used here to widen the analyst's view of what the phrase could mean and to generate the specific questions that will close it, so the meeting produces a decision rather than another vague line."}], "correct": "D", "select": 1}, {"id": "a4-074", "domain": 4, "task_id": "4.2", "objective": "Leverage Claude for research, planning, and process optimization", "situation": "A finance team wants Claude to help streamline month-end close, but the process has never been written down and each of the four preparers does parts of it differently.", "question": "What should the team do first?", "options": [{"letter": "A", "text": "Write down how the close actually runs today, including exceptions and the preparers' own variations.", "correct": true, "explanation": "There is nothing to optimise until the current process is visible, and the exceptions and per-preparer variations are exactly where the inefficiency lives. Capturing them gives Claude a real process to analyse and gives the team a baseline to improve against."}, {"letter": "B", "text": "Ask Claude to describe a typical month-end close process and use that description as the starting point.", "correct": false, "explanation": "A typical close describes the average of other organisations, not this one. Optimising against it produces recommendations aimed at steps this team may not perform and misses the variations that are the actual source of friction."}, {"letter": "C", "text": "Ask Claude to list the month-end steps that are most often automated in finance teams of a similar size.", "correct": false, "explanation": "A list of commonly automated steps is a solution catalogue offered before the problem is known. It also encourages the team to chase whichever steps are easy to automate rather than the ones costing them time."}, {"letter": "D", "text": "Ask each preparer to try Claude on their own part of the close and then share whatever improvements they find.", "correct": false, "explanation": "Four uncoordinated experiments deepen the existing problem, which is that four people already do the close differently. Local improvements to inconsistent practice make the overall process harder to reason about."}], "correct": "A", "select": 1}, {"id": "a4-075", "domain": 4, "task_id": "4.1", "objective": "Apply Claude to analyze requirements and use cases", "situation": "", "question": "When a stakeholder provides an ambiguous requirement like 'secure login', what is the best approach when using Claude for requirements analysis?", "options": [{"letter": "A", "text": "Ask Claude to build a login feature from the requirement, assuming it is sufficiently clear.", "correct": false, "explanation": "Asking Claude to build a login feature from an ambiguous requirement like 'secure login' risks producing a misaligned or incomplete solution. It is better to first clarify the requirement by gathering details about authentication methods, security standards, and user constraints."}, {"letter": "B", "text": "Ask Claude to interpret 'secure login' as two-factor authentication and document that assumption.", "correct": false, "explanation": "Interpreting 'secure login' as two-factor authentication without validation risks making an incorrect assumption. While documenting assumptions can be useful, arbitrarily locking in a design choice like two-factor authentication may introduce a requirement that does not match stakeholder intent."}, {"letter": "C", "text": "Ask Claude to generate specific clarifying questions about what 'secure login' entails to stakeholders.", "correct": true, "explanation": "Asking Claude to generate specific clarifying questions helps stakeholders define the scope and specifics of 'secure login', reducing assumptions. This aligns with requirements analysis best practices by ensuring the solution addresses actual needs before implementation."}, {"letter": "D", "text": "Ask Claude to confirm that 'secure login' is clear and does not require further elaboration.", "correct": false, "explanation": "Confirming that 'secure login' is clear without further elaboration overlooks its inherent ambiguity. The phrase is not specific enough to be complete; skipping a discovery step can lead to misunderstandings and a solution that does not meet real needs."}], "correct": "C", "select": 1}, {"id": "a4-076", "domain": 4, "task_id": "4.2", "objective": "Leverage Claude for research, planning, and process optimization", "situation": "A strategy lead is planning a market-entry study and wants Claude to help with the research phase of it. Which two approaches are sound?", "question": "Select two.", "options": [{"letter": "A", "text": "Ask for the finished conclusion in a single prompt and publish it with no intermediate step.", "correct": false, "explanation": "Collapsing a study into one prompt and publishing the result removes every point where a human could catch an error."}, {"letter": "B", "text": "Ask Claude to draft the research plan and its open questions, then have the team refine it.", "correct": true, "explanation": "Drafting the plan and the open questions is planning work Claude does well, and the team refining it keeps ownership where it belongs."}, {"letter": "C", "text": "Use Research mode and check the citations it returns back against the original sources.", "correct": true, "explanation": "Research mode returns citations, and checking them against the originals is how a claim is actually verified."}, {"letter": "D", "text": "Leave out the decision the study must support, so Claude is not steered toward an answer.", "correct": false, "explanation": "Stating the decision the study supports is what makes the research useful; withholding it produces unfocused output."}, {"letter": "E", "text": "Accept the market figures Claude gives as final, since the answer reads confident and full.", "correct": false, "explanation": "Confident phrasing is not evidence, and market figures used for a real decision need checking against sources."}], "correct": ["B", "C"], "select": 2}, {"id": "a4-077", "domain": 4, "task_id": "4.1", "objective": "Apply Claude to analyze requirements and use cases", "situation": "Two departments have submitted requirements for the same intake queue that directly contradict each other about who approves exceptions. The analyst is preparing the requirements pack for a decision meeting.", "question": "What should the analyst do first?", "options": [{"letter": "A", "text": "Ask Claude to set out every point at which the two submissions conflict, so they can be tabled openly.", "correct": true, "explanation": "The analyst's job is to make the conflict visible and decidable, not to remove it. A precise list of where the two submissions diverge turns a vague disagreement into a short agenda that the decision meeting can actually settle."}, {"letter": "B", "text": "Ask Claude to merge the two submissions into one wording that both departments could plausibly accept.", "correct": false, "explanation": "Merging contradictory statements into agreeable wording hides the disagreement rather than resolving it. The conflict then resurfaces during delivery, when the cost of settling who approves exceptions is far higher."}, {"letter": "C", "text": "Ask Claude to recommend the approval route that is more common in organisations of a comparable size.", "correct": false, "explanation": "What comparable organisations do is not evidence about this organisation's accountability for exceptions. It also asks Claude to settle a governance question that belongs to the two departments and their sponsor."}, {"letter": "D", "text": "Ask Claude to drop the approval requirements from the pack and raise them later during the design phase.", "correct": false, "explanation": "Removing the contested requirement leaves the pack looking complete while the hardest question is unanswered. Deferring it to design guarantees the argument happens later, with a partly built process depending on the outcome."}], "correct": "A", "select": 1}, {"id": "a4-078", "domain": 4, "task_id": "4.2", "objective": "Leverage Claude for research, planning, and process optimization", "situation": "", "question": "Which of the following is the most effective method for using Claude to produce an executive summary of a lengthy market report?", "options": [{"letter": "A", "text": "Break the report into sections, prompt Claude to summarize each section separately, and then manually compile those summaries.", "correct": false, "explanation": "While breaking the report into sections can help process long documents, manually compiling separate summaries adds extra work and can introduce inconsistency. This approach is less efficient than prompting Claude to generate a structured executive summary directly from the full report."}, {"letter": "B", "text": "Provide the report and ask Claude to produce a single-paragraph executive summary without providing any explicit structure or focus areas.", "correct": false, "explanation": "Providing the report without any explicit structure or focus areas may result in a generic or unfocused single-paragraph summary. Claude benefits from clear guidance to produce a summary that is structured and relevant for an executive audience."}, {"letter": "C", "text": "Supply the entire report, specify a desired structure (e.g., key trends, risks, opportunities), and request a concise structured summary.", "correct": true, "explanation": "Supplying the entire report and specifying a desired structure (e.g., key trends, risks, opportunities) ensures a concise, well-organized, and relevant executive summary. This leverages Claude's ability to follow detailed instructions and improves efficiency."}, {"letter": "D", "text": "Upload the report and ask Claude to list all action items directly mentioned in the text, ignoring the general market context.", "correct": false, "explanation": "Asking Claude to list only action items directly mentioned in the text is too narrow for an executive summary and ignores broader context such as trends, risks, and opportunities. This makes the output less useful for strategic decision-making."}], "correct": "C", "select": 1}, {"id": "a4-079", "domain": 4, "task_id": "4.2", "objective": "Leverage Claude for research, planning, and process optimization", "situation": "A team asks Claude to draw up options for reorganising its weekly reporting cycle, and Claude returns three credible options.", "question": "What is the most appropriate next step?", "options": [{"letter": "A", "text": "Adopt the option Claude lists first, since it ordered them by likely effectiveness", "correct": false, "explanation": "Order of presentation carries no ranking, and treating it as one substitutes a formatting artefact for analysis."}, {"letter": "B", "text": "Ask for a fourth option, on the basis that more choices produce a better decision", "correct": false, "explanation": "A fourth option adds volume without resolving anything, because nothing has yet been checked against what the team needs."}, {"letter": "C", "text": "Ask Claude to pick one of the three and put that choice to the sponsor as agreed", "correct": false, "explanation": "Asking Claude to choose hands over the accountable part of the decision while leaving the constraints untested."}, {"letter": "D", "text": "Test each option against the constraints the team already knows it must satisfy", "correct": true, "explanation": "Generating options is delegable; choosing between them against the team's real constraints is the judgement the team has to own, and it is what turns three suggestions into a decision."}], "correct": "D", "select": 1}, {"id": "a4-080", "domain": 4, "task_id": "4.2", "objective": "Leverage Claude for research, planning, and process optimization", "situation": "An analyst uses Claude's research capability to produce a market briefing with citations, then pastes the briefing into a board paper without opening any of the cited sources.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The research may have taken several minutes to run and consumed a noticeable share of the monthly usage.", "correct": false, "explanation": "Time and usage consumed are costs the analyst has already paid. They say nothing about the reliability of what the board is about to rely on."}, {"letter": "B", "text": "A cited claim may not be supported by the source, and the citations make the paper look already checked.", "correct": true, "explanation": "Citations are what makes verification possible, not a substitute for it. A reference can point to a source that says something narrower, older or different from the claim it supports, and the presence of the reference discourages anyone at the board table from checking."}, {"letter": "C", "text": "Board members may prefer a short slide deck rather than a written paper when the material is this dense.", "correct": false, "explanation": "Format preference is a matter of house style and can be resolved by converting the material. The board would still be making a decision on claims that nobody has traced back to their sources."}, {"letter": "D", "text": "The briefing may be longer than the board's standard paper format allows and require heavy trimming.", "correct": false, "explanation": "Length is visible and easy to correct before the paper is issued. It affects how comfortably the board reads the briefing, not whether the statements in it are true."}], "correct": "B", "select": 1}, {"id": "a4-081", "domain": 4, "task_id": "4.2", "objective": "Leverage Claude for research, planning, and process optimization", "situation": "Given a written description of an approvals workflow, Claude suggests removing a second review step that appears redundant. The team plans to remove it next week.", "question": "What is the greatest concern?", "options": [{"letter": "A", "text": "Removing a step may require retraining the reviewers who currently perform the second review.", "correct": false, "explanation": "Retraining is a normal cost of any process change and is planned for openly. It is a consequence of the decision rather than a reason the decision might be wrong."}, {"letter": "B", "text": "The description given to Claude may not use the same step names that the workflow tracking tool records.", "correct": false, "explanation": "Mismatched naming is a mapping nuisance that surfaces quickly and is easy to reconcile. It might make the analysis harder to read, but it does not make removing the step dangerous."}, {"letter": "C", "text": "The time saved may be smaller than the team expects once the change has been rolled out fully.", "correct": false, "explanation": "A disappointing saving means the change was not worth much, which is a mild outcome. The team would be no worse off than before, whereas removing a control leaves them exposed."}, {"letter": "D", "text": "The step may exist for a control reason that was never written into the description Claude was given.", "correct": true, "explanation": "Claude can only reason about what the description contains, and the reasons for control steps are usually the part nobody writes down. A step that looks redundant on paper may be the segregation of duties or fraud check an auditor relies on, and removing it is not easily undone."}], "correct": "D", "select": 1}, {"id": "a4-082", "domain": 4, "task_id": "4.3", "objective": "Use Claude to support solution design, development, and iteration", "situation": "Six weeks in, a team must decide whether to keep a Claude-assisted intake summarising arrangement or return to the previous approach.", "question": "Which comparison best supports that decision?", "options": [{"letter": "A", "text": "How much faster the summaries are produced now than the team managed before the trial.", "correct": false, "explanation": "Speed alone answers half the question. Faster summaries that are less accurate would push cost downstream into misdirected cases, which is why speed has to be weighed against the agreed quality criteria."}, {"letter": "B", "text": "How the summaries compare with the team's own on the criteria agreed before the trial began.", "correct": true, "explanation": "The decision is a comparison against the realistic alternative, measured on criteria fixed in advance so the standard cannot drift to fit the result. That is what turns six weeks of experience into an evidenced choice rather than an impression."}, {"letter": "C", "text": "Whether the summaries are entirely free of error, since a wrong summary can misdirect a case.", "correct": false, "explanation": "A demand for zero errors sets a standard the previous human process never met either. Judged that way every arrangement fails, so the test cannot distinguish a good option from a bad one."}, {"letter": "D", "text": "Whether participants say they enjoy the new arrangement more than the approach it replaced.", "correct": false, "explanation": "Enjoyment affects adoption and is worth knowing, but people can prefer an arrangement that produces weaker summaries. Preference is a secondary signal, not the basis for keeping it."}], "correct": "B", "select": 1}, {"id": "a4-083", "domain": 4, "task_id": "4.2", "objective": "Leverage Claude for research, planning, and process optimization", "situation": "A strategy lead is planning a market-entry study and wants Claude to help with the research phase of it. Which two approaches are sound?", "question": "Select two.(Select 2)", "options": [{"letter": "A", "text": "Accept the market figures Claude gives as final, since the answer reads confident and full.", "correct": false, "explanation": "Confident phrasing is not evidence, and market figures used for a real decision need checking against sources."}, {"letter": "B", "text": "Use Research mode and check the citations it returns back against the original sources.", "correct": true, "explanation": "Research mode returns citations, and checking them against the originals is how a claim is actually verified."}, {"letter": "C", "text": "Ask for the finished conclusion in a single prompt and publish it with no intermediate step.", "correct": false, "explanation": "Collapsing a study into one prompt and publishing the result removes every point where a human could catch an error."}, {"letter": "D", "text": "Leave out the decision the study must support, so Claude is not steered toward an answer.", "correct": false, "explanation": "Stating the decision the study supports is what makes the research useful; withholding it produces unfocused output."}, {"letter": "E", "text": "Ask Claude to draft the research plan and its open questions, then have the team refine it.", "correct": true, "explanation": "Drafting the plan and the open questions is planning work Claude does well, and the team refining it keeps ownership where it belongs."}], "correct": ["B", "E"], "select": 2}, {"id": "a4-084", "domain": 4, "task_id": "4.2", "objective": "Leverage Claude for research, planning, and process optimization", "situation": "A service manager wants to use Claude to improve a customer refund process that everyone agrees is too slow.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Map how the process actually runs today; have Claude propose where time is being lost; test those proposals against real cycle-time records; agree the changes with the people who run the steps.", "correct": true, "explanation": "The map grounds the analysis in the real process, Claude then does the pattern work of pointing at likely losses, the cycle-time records separate genuine bottlenecks from plausible-sounding ones, and the people who run the steps agree changes that have already been evidenced."}, {"letter": "B", "text": "Map how the process actually runs today; agree the changes with the people who run the steps; have Claude propose where time is being lost; test those proposals against real cycle-time records.", "correct": false, "explanation": "Agreeing changes immediately after mapping commits the team before Claude has identified anything, and the evidence check lands at the very end when the changes are already settled."}, {"letter": "C", "text": "Test those proposals against real cycle-time records; have Claude propose where time is being lost; map how the process actually runs today; agree the changes with the people who run the steps.", "correct": false, "explanation": "There are no proposals to test at the start, so the first step has nothing to act on, and mapping the process arrives after the analysis that depended on it."}, {"letter": "D", "text": "Have Claude propose where time is being lost; map how the process actually runs today; agree the changes with the people who run the steps; test those proposals against real cycle-time records.", "correct": false, "explanation": "Asking for proposals before the process is mapped means Claude reasons from an assumed refund process rather than this one, and the changes are agreed with the team before anyone has tested whether they hold."}], "correct": "A", "select": 1}, {"id": "a4-085", "domain": 4, "task_id": "4.5", "objective": "Communicate Claude's value and limitations to stakeholders", "situation": "An HR stakeholder proposes letting Claude autonomously make final hiring decisions from resumes with no human involvement, to speed up recruiting.", "question": "What limitation should be raised, and what should be recommended instead?", "options": [{"letter": "A", "text": "Raise that Claude cannot read resume text at all, so the proposal is technically impossible to build.", "correct": false, "explanation": "Claude can process and reason about resume text, so this claim is factually wrong and not the relevant concern here."}, {"letter": "B", "text": "Raise that hiring speed is the only relevant factor, so fairness and bias considerations can be dropped entirely.", "correct": false, "explanation": "Fairness and legal compliance are material concerns in hiring and should not be deprioritized in favor of speed alone."}, {"letter": "C", "text": "Raise that model outputs can reflect biases in input data, and recommend keeping a human reviewer in the final decision.", "correct": true, "explanation": "Like any model trained on real-world data, Claude's outputs can reflect biases present in that data, so high-stakes decisions like hiring should keep a human reviewer involved rather than being fully automated."}, {"letter": "D", "text": "Raise no concerns, since removing human involvement from hiring eliminates all risk of biased outcomes.", "correct": false, "explanation": "Removing human oversight does not eliminate bias risk; unsupervised automation can actually amplify it if left unchecked."}], "correct": "C", "select": 1}, {"id": "a4-086", "domain": 4, "task_id": "4.2", "objective": "Leverage Claude for research, planning, and process optimization", "situation": "A policy officer must produce a short evidence review for a decision meeting later in the week.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Point Claude at the sources that count as evidence here; check the load-bearing claims before it is circulated; state the question and what a usable answer must cover; have Claude draft the review with references to passages.", "correct": false, "explanation": "Checking claims before anything has been drafted leaves nothing to check, and stating the question only after that verification means the sources were chosen without knowing what was being asked."}, {"letter": "B", "text": "Check the load-bearing claims before it is circulated; have Claude draft the review with references to passages; state the question and what a usable answer must cover; point Claude at the sources that count as evidence here.", "correct": false, "explanation": "This inverts the whole order: verification comes first with nothing to verify, and the question that should have framed the work is settled after the review has already been written."}, {"letter": "C", "text": "State the question and what a usable answer must cover; point Claude at the sources that count as evidence here; have Claude draft the review with references to passages; check the load-bearing claims before it is circulated.", "correct": true, "explanation": "The question defines what the review must cover, the agreed sources bound what counts as evidence, the referenced draft makes each claim traceable, and verifying the claims the decision turns on happens while there is still time to correct them."}, {"letter": "D", "text": "Have Claude draft the review with references to passages; state the question and what a usable answer must cover; point Claude at the sources that count as evidence here; check the load-bearing claims before it is circulated.", "correct": false, "explanation": "Drafting before the question is stated produces a review aimed at no particular decision, and the sources arrive after the draft that was supposed to rest on them."}], "correct": "C", "select": 1}, {"id": "a4-087", "domain": 4, "task_id": "4.3", "objective": "Use Claude to support solution design, development, and iteration", "situation": "After several rounds of instruction changes, a Claude-assisted complaints-response arrangement still gets the tone wrong on the cases that involve bereavement.", "question": "What is the most appropriate change to the design?", "options": [{"letter": "A", "text": "Route bereavement cases to a person to write, and keep the arrangement for the routine ones.", "correct": true, "explanation": "Iterating a design includes changing the division of labour, not only the instructions. Bereavement responses need judgment and care that the team must own, so returning that slice to a person while keeping the routine volume delegated is the sound redesign."}, {"letter": "B", "text": "Ask Claude to flag bereavement cases so that reviewers know to read those responses carefully.", "correct": false, "explanation": "Flagging improves the odds that a reviewer pays attention, but Claude still writes the response and a person still has to rescue it. The step with the real consequence remains delegated."}, {"letter": "C", "text": "Add more instruction detail about tone until the sensitive cases are handled acceptably as well.", "correct": false, "explanation": "Several rounds of instruction changes have already failed, which is evidence that the problem is the nature of the task rather than the wording. Continuing to add detail spends effort on the least promising fix."}, {"letter": "D", "text": "Accept the current quality and ask reviewers to correct the tone on those cases before sending.", "correct": false, "explanation": "Relying on reviewers to repair tone in the most sensitive cases puts the heaviest correction burden exactly where mistakes are least forgivable. One missed review reaches a bereaved complainant."}], "correct": "A", "select": 1}, {"id": "a4-088", "domain": 4, "task_id": "4.5", "objective": "Communicate Claude's value and limitations to stakeholders", "situation": "An analyst is presenting a Claude-assisted workflow to sceptical senior stakeholders next week. Which two things should the presentation include?", "question": "Select two.(Select 2)", "options": [{"letter": "A", "text": "A comparison of technical benchmark scores across each of the underlying model tiers.", "correct": false, "explanation": "Benchmark scores are not what a business audience needs, and they say nothing about this workflow's value."}, {"letter": "B", "text": "Where a human still reviews the output, and why that review step has been kept in place.", "correct": true, "explanation": "Naming where human review remains, and why, is honest limitation-setting and is what makes the value claim credible."}, {"letter": "C", "text": "A claim that Claude will not make any factual errors once all its instructions are written well.", "correct": false, "explanation": "Good instructions reduce error but do not eliminate it, so this claim is false and will damage trust later."}, {"letter": "D", "text": "A specific, measured example of what the workflow improved during the pilot it ran on.", "correct": true, "explanation": "A measured pilot result is concrete evidence of value rather than an assertion stakeholders are asked to take on trust."}, {"letter": "E", "text": "A promise that the workflow removes any need to check the output from this point onward.", "correct": false, "explanation": "Promising that checking is no longer needed oversells the tool and sets the expectation that will break first."}], "correct": ["B", "D"], "select": 2}, {"id": "a4-089", "domain": 4, "task_id": "4.1", "objective": "Apply Claude to analyze requirements and use cases", "situation": "A programme sponsor asks who is accountable for the requirements catalogue now that most of it is drafted with Claude.", "question": "Which answer best reflects sound practice?", "options": [{"letter": "A", "text": "The named business owner still signs it off; Claude speeds the drafting but cannot carry accountability.", "correct": true, "explanation": "Delegating the drafting does not move the accountability. The business owner remains the person who can commit to the requirements being right, and keeping sign-off there is what makes the faster drafting safe to rely on."}, {"letter": "B", "text": "The analyst who wrote the prompts signs it off, since they know best how the catalogue was put together.", "correct": false, "explanation": "Knowing how the catalogue was produced is not the same as being able to commit the business to it. Prompt authorship is a process detail; sign-off is a statement that these are the needs the organisation will fund."}, {"letter": "C", "text": "Claude signs the catalogue off, and the business owner reviews only the items Claude marks as uncertain.", "correct": false, "explanation": "A tool cannot be accountable for a commitment the organisation makes, and reviewing only self-flagged items means the errors Claude is confident about, which are the dangerous ones, go unexamined."}, {"letter": "D", "text": "Sign-off can wait while the catalogue is a draft, because every item will be reviewed again in delivery.", "correct": false, "explanation": "Deferring sign-off pushes accountability into a phase where changing a requirement is far more expensive. Delivery review checks whether the build matches the catalogue, not whether the catalogue was ever right."}], "correct": "A", "select": 1}, {"id": "a4-090", "domain": 4, "task_id": "4.5", "objective": "Communicate Claude's value and limitations to stakeholders", "situation": "A stakeholder asks why a complex multi-step reasoning task takes longer and costs more than a simple lookup task run on the same model.", "question": "What tradeoff should be explained?", "options": [{"letter": "A", "text": "Explain that all tasks cost the same regardless of complexity, so the difference must come from a network outage.", "correct": false, "explanation": "Cost and latency do scale with task complexity and token usage; attributing the difference to an outage misrepresents normal behavior."}, {"letter": "B", "text": "Explain that deeper reasoning consumes more tokens and time, so settings that improve quality on hard tasks add cost.", "correct": true, "explanation": "Increasing reasoning depth (for example through higher effort or thinking settings) generally uses more tokens and processing time, which is the direct tradeoff between response quality and cost/latency."}, {"letter": "C", "text": "Explain that complex reasoning tasks are always cheaper, since they need fewer output tokens than simple lookups.", "correct": false, "explanation": "Complex reasoning tasks typically use more output tokens, not fewer, making them more expensive rather than cheaper."}, {"letter": "D", "text": "Explain that latency differences are random and unrelated to how much reasoning a given request requires.", "correct": false, "explanation": "Latency is meaningfully related to reasoning depth and token volume, not simply random variation."}], "correct": "B", "select": 1}, {"id": "a4-091", "domain": 4, "task_id": "4.2", "objective": "Leverage Claude for research, planning, and process optimization", "situation": "A strategy analyst is told, \"Research our competitors and come back with something useful.\" She has two days and intends to use Claude for most of the work.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Ask Claude to produce a broad overview of every competitor in the sector and narrow the list afterwards.", "correct": false, "explanation": "A broad sweep first spends most of the two days on competitors that will turn out to be irrelevant. Narrowing afterwards requires the very criteria the analyst has not yet agreed, so the effort has to be repeated."}, {"letter": "B", "text": "Collect the competitors' public material into one place so Claude has something concrete to work from.", "correct": false, "explanation": "Gathering source material is a sensible early move, but doing it before the question is fixed means gathering the wrong material. Which brochures, filings and reviews matter depends entirely on what the requester needs to decide."}, {"letter": "C", "text": "Agree with the requester which decision the research supports and which competitors and factors matter.", "correct": true, "explanation": "\"Something useful\" is not a research question, and no amount of good prompting rescues an undefined one. Fixing the decision the work feeds and the comparison factors that bear on it determines what to gather, what to ignore, and when the analyst can stop."}, {"letter": "D", "text": "Ask Claude which competitor comparison factors are most commonly used in this kind of market study.", "correct": false, "explanation": "Common comparison factors describe what other studies did, not what this requester needs. Adopting them substitutes a generic frame for the specific business question and leaves the analyst no closer to knowing what useful means here."}], "correct": "C", "select": 1}, {"id": "a4-092", "domain": 4, "task_id": "4.3", "objective": "Use Claude to support solution design, development, and iteration", "situation": "A department wants Claude to take over a large slice of its quarterly reporting work. Which two steps belong before any full rollout?", "question": "Select two.", "options": [{"letter": "A", "text": "Start with the reports carrying the highest legal exposure, since they save the most time.", "correct": false, "explanation": "Starting where legal exposure is greatest maximises the blast radius of a mistake in an untested process."}, {"letter": "B", "text": "Pilot the change on one report, so the result can be compared with the current output.", "correct": true, "explanation": "Piloting on a small scope gives a comparison against today's output before anything is committed department-wide."}, {"letter": "C", "text": "Agree what counts as an acceptable result before any of the work is handed over at all.", "correct": true, "explanation": "Defining what good looks like before delegating is what makes the pilot's result interpretable at all."}, {"letter": "D", "text": "Announce the change as settled, so that the department commits to the new way of working.", "correct": false, "explanation": "Announcing the change as finished commits the department before there is any evidence the new way works."}, {"letter": "E", "text": "Remove the existing review step first, so the pilot measures the fully automated version.", "correct": false, "explanation": "Stripping review out of a pilot removes the safeguard exactly when the risk of an unnoticed error is highest."}], "correct": ["B", "C"], "select": 2}, {"id": "a4-093", "domain": 4, "task_id": "4.2", "objective": "Leverage Claude for research, planning, and process optimization", "situation": "A team is about to change a shift handover process on Claude's suggestion, and wants to be able to tell in a month whether the change actually helped.", "question": "What most makes that judgement possible?", "options": [{"letter": "A", "text": "A baseline of the old process's timings and error rates, captured before the change is made.", "correct": true, "explanation": "Improvement is a comparison, and the comparison is impossible once the old process has gone. Capturing timings and error rates beforehand is the only step that must happen now, and it turns a month of impressions into evidence."}, {"letter": "B", "text": "A note of how confident Claude was that its suggested change would improve the handover process.", "correct": false, "explanation": "Claude's stated confidence is a prediction, not a measurement, and it was formed before the change met real shifts. Treating it as evidence of improvement substitutes expectation for observation."}, {"letter": "C", "text": "A list of the other changes that Claude suggested and the team decided not to make at this time.", "correct": false, "explanation": "A record of rejected suggestions is useful for the next round of improvement work. It tells the team what they chose not to try, not whether what they did try worked."}, {"letter": "D", "text": "A record of the prompts used when Claude was asked to analyse the handover process originally.", "correct": false, "explanation": "Keeping the prompts helps the team repeat or explain the analysis, which is good practice. It says nothing about whether the process is faster or more accurate than it used to be."}], "correct": "A", "select": 1}, {"id": "a4-094", "domain": 4, "task_id": "4.3", "objective": "Use Claude to support solution design, development, and iteration", "situation": "A team lead is about to hand a multi-step campaign preparation task to Claude in Cowork, where it will work across her files and connected tools.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Describe the outcome wanted and the constraints; decide which actions need approval before they happen; let Claude work through the task; review what was actually done before anything is final.", "correct": true, "explanation": "Describing the outcome and constraints gives the task a target, deciding in advance which actions need approval keeps the consequential steps under human control, and reviewing what was done before anything is finalised keeps the team lead accountable for the result."}, {"letter": "B", "text": "Decide which actions need approval before they happen; review what was actually done before anything is final; describe the outcome wanted and the constraints; let Claude work through the task.", "correct": false, "explanation": "Reviewing before the task has run leaves nothing to review, and the outcome is described only after the approval points have been fixed, so those points were chosen without knowing what the task involves."}, {"letter": "C", "text": "Let Claude work through the task; describe the outcome wanted and the constraints; decide which actions need approval before they happen; review what was actually done before anything is final.", "correct": false, "explanation": "Starting the work before the outcome and the approval points are set means Claude acts on files and tools under no agreed constraint, which is exactly the situation the approval decision exists to prevent."}, {"letter": "D", "text": "Review what was actually done before anything is final; let Claude work through the task; describe the outcome wanted and the constraints; decide which actions need approval before they happen.", "correct": false, "explanation": "This puts the review first and the description of the goal third, so the work runs with no stated outcome and the approval decision is made after everything has already happened."}], "correct": "A", "select": 1}, {"id": "a4-095", "domain": 4, "task_id": "4.2", "objective": "Leverage Claude for research, planning, and process optimization", "situation": "A finance team wants Claude's help making its slow month-end reporting process faster. Which two opening moves make sense?", "question": "Select two.", "options": [{"letter": "A", "text": "Describe the current process step by step so Claude can point at the slowest handoffs in it.", "correct": true, "explanation": "Optimisation starts from the actual process, and pasting the real steps is what lets Claude find where time is lost."}, {"letter": "B", "text": "Ask Claude to approve the final numbers so that the human reviewer step can be removed.", "correct": false, "explanation": "Sign-off on financial figures is accountable human work, and removing the reviewer is exactly the wrong optimisation."}, {"letter": "C", "text": "Ask Claude which of the team members is the real cause of the delay in the closing cycle.", "correct": false, "explanation": "Claude cannot judge individual performance from a process description, and framing delay as blame is the wrong question."}, {"letter": "D", "text": "Ask Claude to rewrite the entire process, then adopt the result at the very next month-end close.", "correct": false, "explanation": "Adopting a wholesale rewrite untested at the next close bets the month-end on a change nobody has validated."}, {"letter": "E", "text": "Ask Claude for options with their trade-offs, which the team then tests on a single cycle.", "correct": true, "explanation": "Options with trade-offs, piloted on one cycle, is how a process change is tested before it is trusted at scale."}], "correct": ["A", "E"], "select": 2}, {"id": "a4-096", "domain": 4, "task_id": "4.3", "objective": "Use Claude to support solution design, development, and iteration", "situation": "A project lead is demonstrating a new Claude-assisted reporting workflow to executives who could fund its wider use.", "question": "What should the demonstration make clear alongside the benefits?", "options": [{"letter": "A", "text": "The wording of the instructions used, so that executives can see how the workflow was actually built.", "correct": false, "explanation": "Showing the instructions is transparent about the mechanics but not about the risk. Executives are not well placed to infer failure modes from wording, and detail here can create false reassurance."}, {"letter": "B", "text": "Where the workflow still needs a person to check the output, and the kinds of error it can produce.", "correct": true, "explanation": "Honest communication of capability and limitation is what makes the funding decision sound: executives need to know which checks remain human and what kinds of error to expect, or they will plan for a level of automation the workflow does not deliver."}, {"letter": "C", "text": "The number of hours the project lead spent designing and refining the workflow before the demonstration.", "correct": false, "explanation": "Design effort is a cost already incurred and tells executives nothing about whether the workflow is dependable. It answers a question about the team rather than about the decision being funded."}, {"letter": "D", "text": "The other tools the team considered and the reasons each of them was set aside before Claude was chosen.", "correct": false, "explanation": "A tool comparison explains how the team arrived here, which may be of passing interest. It does not tell executives what they will be relying on or where it can go wrong."}], "correct": "B", "select": 1}, {"id": "a4-097", "domain": 4, "task_id": "4.5", "objective": "Communicate Claude's value and limitations to stakeholders", "situation": "A sponsor wants to skip evaluation and move straight to full production rollout of a new Claude-powered feature to save time.", "question": "How should the solutions architect explain the risk of this plan?", "options": [{"letter": "A", "text": "Explain that skipping evaluation is safe as long as the feature is reviewed by the engineering team after launch instead of before.", "correct": false, "explanation": "Reviewing after full launch means issues are discovered only once they have already reached all users, defeating the purpose of pre-launch evaluation."}, {"letter": "B", "text": "Explain that benchmark testing against real prompts and data is needed first to catch accuracy issues before they reach every user.", "correct": true, "explanation": "Building benchmark tests and evaluating against representative prompts and data before rollout is the standard way to catch quality or accuracy issues before they affect the full user base."}, {"letter": "C", "text": "Explain that evaluation can be skipped since production issues can always be reversed instantly with no impact to users.", "correct": false, "explanation": "Production issues involving inaccurate or inappropriate outputs are not instantly reversible in terms of user trust or downstream impact."}, {"letter": "D", "text": "Explain that evaluation is optional because Claude's output quality is guaranteed to be identical across every possible use case.", "correct": false, "explanation": "Output quality varies by task and data, so there is no guarantee of uniform performance across different use cases without evaluation."}], "correct": "B", "select": 1}, {"id": "a4-098", "domain": 4, "task_id": "4.3", "objective": "Use Claude to support solution design, development, and iteration", "situation": "A newly designed way of working has Claude approve expense claims under a set threshold outright, with no person seeing the ones it approves.", "question": "What is the greatest concern with this design?", "options": [{"letter": "A", "text": "Approved payments leave the organisation with no one accountable for the decision to pay.", "correct": true, "explanation": "Once money has left the organisation the action is not reversible, and because nobody reviews the approved claims there is no person who can answer for the decision. Payment approval is the kind of accountable, irreversible step that should keep a human in it."}, {"letter": "B", "text": "The threshold may need adjusting as expense patterns change over the course of the year.", "correct": false, "explanation": "Threshold maintenance is routine tuning of a control that already exists. It is a reason to review the design periodically rather than a reason the design is unsound."}, {"letter": "C", "text": "Employees may submit more claims once they learn that approvals are handled automatically.", "correct": false, "explanation": "A rise in claim volume is a behavioural effect worth monitoring, and it can be managed by sampling or by adjusting the threshold. It does not describe what happens when an approval is wrong."}, {"letter": "D", "text": "Claims written in unusual formats may take the arrangement longer to work through fully.", "correct": false, "explanation": "Unusual formats slow processing and may produce the occasional query. That is a throughput problem, not a governance one, and it is visible enough to be corrected."}], "correct": "A", "select": 1}, {"id": "a4-099", "domain": 4, "task_id": "4.3", "objective": "Use Claude to support solution design, development, and iteration", "situation": "A marketing team has always written its weekly client status note by hand. They now plan to switch to a new arrangement in which Claude drafts the note each week and an account manager edits it before sending.", "question": "Before the first Claude-drafted note under the new arrangement reaches a client, what should the team do first?", "options": [{"letter": "A", "text": "Agree in advance exactly how many minutes of editing per note the account manager should expect to spend on this every week.", "correct": false, "explanation": "An editing-time budget measures effort, which only means something once the drafts are known to be usable. Agreeing on one before that check risks pressuring the account manager to spend less time editing than the drafts actually need."}, {"letter": "B", "text": "Send the first Claude-drafted note straight to the client, along with a short note explaining that Claude helped produce it.", "correct": false, "explanation": "Disclosure is good practice, but it does not test anything the team needs to know yet. The client would still receive the first output of an arrangement nobody has tried, and explaining that Claude helped does not make an inaccurate status note acceptable."}, {"letter": "C", "text": "Run Claude on a past month's inputs and compare its drafts with the hand-written notes actually sent to clients that month.", "correct": true, "explanation": "Feeding Claude the same weekly inputs the account manager already used, for a month that has already happened, produces drafts the team can check against a known right answer: the hand-written note the client actually received that week, before Claude was involved. That lets the team see where Claude's drafts fall short before any client sees output from the new arrangement."}, {"letter": "D", "text": "Ask Claude to check its own draft against the account manager's usual style before it goes out to the client.", "correct": false, "explanation": "Self-review can catch obvious slips, but Claude judging its own draft uses the same understanding that produced the draft in the first place. That is no substitute for comparing drafts against notes the team already knows were accurate."}], "correct": "C", "select": 1}, {"id": "a4-100", "domain": 4, "task_id": "4.3", "objective": "Use Claude to support solution design, development, and iteration", "situation": "A department has used Claude to redraft a board paper over six rounds. Each round was reviewed by whoever was free that week, and each reviewer applied their own idea of what a good board paper looks like.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "No shared standard of quality, so the team cannot tell whether the paper improved", "correct": true, "explanation": "Six reviewers applying six private ideas of quality means each round pulls the draft in a different direction, so the department has no agreed measure against which to judge round six versus round one. Anthropic's AI Fluency material defines Product Discernment as evaluating the quality of what the model produces for accuracy, appropriateness, coherence and relevance, and describes Discernment as working hand-in-hand with Description in a continuous feedback loop; that loop only converges when the criteria being applied stay the same from round to round."}, {"letter": "B", "text": "The reviewers may use up their monthly usage allowance before the sixth round ends", "correct": false, "explanation": "Usage allowance caps how much work can be sent to Claude in a period, and reviewers would notice they were running low well before it halted the cycle. It says nothing about whether the drafting process is producing a better board paper."}, {"letter": "C", "text": "Claude may carry wording the reviewers rejected in earlier rounds into later drafts", "correct": false, "explanation": "Wording from an earlier round can resurface when every round shares the same conversation, but a reviewer reading the draft will see it and ask for it to go. It does not explain why the department cannot tell whether the paper is getting better."}, {"letter": "D", "text": "The paper may end up shorter than the format the board has asked for in the past", "correct": false, "explanation": "A draft drifting away from the length the board is used to is visible on the page and corrected with one more instruction. It is a symptom of an unmanaged review cycle rather than the thing that makes six rounds of work impossible to evaluate."}], "correct": "A", "select": 1}, {"id": "a4-101", "domain": 4, "task_id": "4.5", "objective": "Communicate Claude's value and limitations to stakeholders", "situation": "A finance team wants to use Claude to calculate and report final loan interest figures directly to customers with no further checks.", "question": "What risk should be communicated about this workflow?", "options": [{"letter": "A", "text": "Communicate that Claude's arithmetic has been certified as error-free for all financial and regulatory calculations, making separate validation unnecessary.", "correct": false, "explanation": "Claude's arithmetic is not certified as error-free for financial and regulatory calculations. Claiming such certification overstates its reliability and could lead to unvalidated outputs being used in high-stakes contexts."}, {"letter": "B", "text": "Communicate that manual validation is unnecessary because customers are legally obligated to review all figures and will identify any calculation errors on their own.", "correct": false, "explanation": "Relying on customers to catch errors is not a responsible validation approach; financial outputs should be verified internally before they reach customers. Legal obligations on customers do not absolve the provider from ensuring accuracy."}, {"letter": "C", "text": "Communicate that Claude is not designed for precise financial calculations, so loan interest determinations must be delegated to a dedicated accounting system.", "correct": false, "explanation": "While Claude is not a dedicated financial engine, it can assist with such workflows when outputs are validated. The risk is not total incapability but the need for verification before use."}, {"letter": "D", "text": "Communicate that outputs involving precise financial calculations should be validated against a deterministic calculation system before use.", "correct": true, "explanation": "Precise financial calculations demand high accuracy, and language model outputs are not inherently deterministic or error-free. Therefore, any output involving such calculations should be validated against a deterministic calculation system before being used."}], "correct": "D", "select": 1}, {"id": "a4-102", "domain": 4, "task_id": "4.5", "objective": "Communicate Claude's value and limitations to stakeholders", "situation": "A finance director asks why the team is proposing a smaller, faster Claude model for a high-volume invoice classification workflow instead of the most capable model available.", "question": "What is the best way to justify this choice?", "options": [{"letter": "A", "text": "Explain that the most capable model cannot process invoices at all, so a smaller model is the only technically possible option.", "correct": false, "explanation": "The most capable model can process invoices; it is simply not the most cost- or latency-efficient choice for this task."}, {"letter": "B", "text": "Explain that classification is a straightforward task, so a faster and cheaper model can meet accuracy needs at lower cost and latency.", "correct": true, "explanation": "For high-volume, relatively simple tasks like classification, a faster and cheaper model often delivers sufficient accuracy while reducing cost and latency, which is the standard model-selection tradeoff to communicate."}, {"letter": "C", "text": "Explain that model choice has no effect on cost, so picking the smaller model is purely a matter of team preference.", "correct": false, "explanation": "Model choice directly affects per-token cost and response speed, so it is not merely a matter of preference."}, {"letter": "D", "text": "Explain that smaller models are always more accurate than larger models for every business task the company might have.", "correct": false, "explanation": "Smaller models are not universally more accurate; capability, task complexity, and cost must be weighed case by case."}], "correct": "B", "select": 1}, {"id": "a4-103", "domain": 4, "task_id": "4.2", "objective": "Leverage Claude for research, planning, and process optimization", "situation": "An executive asks a planner to use Claude to help answer whether the company should enter a new market over the next five years.", "question": "Which use of Claude is most appropriate?", "options": [{"letter": "A", "text": "Ask Claude for a single clear recommendation so the executive receives an unambiguous answer to act on.", "correct": false, "explanation": "A single recommendation hides the reasoning the executive most needs to test. Market entry turns on the company's risk appetite and capacity, which the executive holds and Claude does not, so a clean answer here is false confidence."}, {"letter": "B", "text": "Ask Claude to draft the market entry announcement so the timeline can then be worked backwards from it.", "correct": false, "explanation": "Drafting the announcement assumes the decision has been made and then reverse-engineers a plan to fit it. That is a good technique for communicating a chosen course, not for deciding whether to take it."}, {"letter": "C", "text": "Ask Claude to predict the market's size in five years so the decision can rest on one concrete figure.", "correct": false, "explanation": "A five-year market size figure is a projection presented as a fact. Resting the decision on one number invites precision that the underlying uncertainty cannot support and crowds out the trade-offs that actually matter."}, {"letter": "D", "text": "Ask Claude to set out the realistic options with the trade-offs and evidence behind each one.", "correct": true, "explanation": "Claude is strongest here as an analyst that widens and structures the option space, surfacing trade-offs and the evidence behind each path. The executive keeps the strategic choice, which is where the accountability sits."}], "correct": "D", "select": 1}, {"id": "a4-104", "domain": 4, "task_id": "4.5", "objective": "Communicate Claude's value and limitations to stakeholders", "situation": "A legal operations stakeholder plans to send Claude-generated contract redlines directly to clients without any attorney review, to reduce turnaround time.", "question": "What should the solutions architect recommend?", "options": [{"letter": "A", "text": "Recommend keeping attorney review before delivery, since Claude can miss jurisdiction-specific or contextual nuances.", "correct": true, "explanation": "Claude can accelerate redlining, but it may miss jurisdiction-specific nuances or context only a qualified attorney would catch, so attorney review before external delivery reduces legal risk."}, {"letter": "B", "text": "Recommend sending Claude's redlines directly to clients, since its legal reasoning always equals an attorney's judgment.", "correct": false, "explanation": "Claude's outputs are not guaranteed to be equivalent to attorney-level legal judgment, so skipping review is risky."}, {"letter": "C", "text": "Recommend removing Claude from the workflow entirely, since it cannot generate any useful redline suggestions.", "correct": false, "explanation": "Claude can meaningfully assist with drafting and redlining; the concern is about unsupervised external delivery, not usefulness."}, {"letter": "D", "text": "Recommend having clients verify the redlines themselves instead of routing them through internal legal review.", "correct": false, "explanation": "Shifting verification onto clients does not address the firm's own responsibility for quality and risk in what it sends out."}], "correct": "A", "select": 1}, {"id": "a4-105", "domain": 4, "task_id": "4.1", "objective": "Apply Claude to analyze requirements and use cases", "situation": "A marketing operations lead is deciding which steps of the campaign brief intake process to hand to Claude.", "question": "Which characteristic most reliably marks a step as a good candidate for delegation?", "options": [{"letter": "A", "text": "The step is repetitive and text-heavy, and the team can describe clearly what a good result looks like.", "correct": true, "explanation": "Repetitive, text-heavy work with a describable standard of success is the classic strong delegation candidate: there is volume to save, the material suits language work, and the team can tell whether the result is acceptable because they agreed in advance what acceptable means."}, {"letter": "B", "text": "The step depends on one manager's undocumented sense of which stakeholders need careful handling.", "correct": false, "explanation": "Work that rests on one person's undocumented judgment cannot be specified well enough to delegate, and there is no written standard against which the output could be checked. Sole-source judgment is a reason to keep a step human."}, {"letter": "C", "text": "The step commits agency spend externally as soon as it completes, so speed matters more than review.", "correct": false, "explanation": "A step that spends money externally the moment it finishes is irreversible, so an error cannot be caught and corrected. Irreversibility is one of the standard markers of a step that should stay with a person, not move to Claude."}, {"letter": "D", "text": "The step is the one that team members complain about most often during retrospectives and planning days.", "correct": false, "explanation": "How much a step is disliked says nothing about whether it can be delegated well. Unpopular steps are often unpopular precisely because they demand awkward judgment calls, which is the profile of a poor candidate rather than a good one."}], "correct": "A", "select": 1}, {"id": "a4-106", "domain": 4, "task_id": "4.3", "objective": "Use Claude to support solution design, development, and iteration", "situation": "A Claude-assisted tender-response arrangement runs well, but it depends on adjustments that the one person who built it makes by hand each time and has never written down.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The manual adjustments may take that person longer as tender volumes continue to increase.", "correct": false, "explanation": "Growing effort as volumes rise is a scaling concern that would show up gradually and can be planned for. It is a reason to invest further, not the thing that could remove the capability overnight."}, {"letter": "B", "text": "Other teams may build slightly different versions of the same arrangement for their own use.", "correct": false, "explanation": "Divergent local versions cause inconsistency across teams and are worth avoiding. That is an organisational tidiness problem rather than a threat to this team's ability to respond to tenders."}, {"letter": "C", "text": "The person may find it hard to explain the arrangement's benefits when funding is next reviewed.", "correct": false, "explanation": "Difficulty explaining the benefits could cost future funding, which matters. It is a communication weakness rather than the structural fragility of a process nobody else can run."}, {"letter": "D", "text": "The arrangement stops working the moment that person is unavailable, and nobody can rebuild it.", "correct": true, "explanation": "The working design lives in one person's head, so leave, illness or a change of role removes the capability entirely and leaves no way to reconstruct it. An arrangement that cannot be handed over has not really been designed, only performed."}], "correct": "D", "select": 1}, {"id": "a4-107", "domain": 4, "task_id": "4.2", "objective": "Leverage Claude for research, planning, and process optimization", "situation": "A researcher wants to use Claude to synthesize findings from multiple academic papers.", "question": "Which workflow is most effective for leveraging Claude's capabilities?", "options": [{"letter": "A", "text": "Ask Claude to rank all papers by citation count, then perform a thorough analysis and synthesis of only the top three, ignoring the rest.", "correct": false, "explanation": "Ranking papers by citation count may overlook highly relevant but less-cited studies, such as recent or niche works. Limiting analysis to only the top three papers discards potentially important insights, resulting in a skewed or incomplete synthesis."}, {"letter": "B", "text": "Manually summarize each paper first, then ask Claude to paraphrase those summaries into a single narrative without further analysis.", "correct": false, "explanation": "Manual summarization adds unnecessary effort and prevents Claude from working with original content directly. Having Claude only paraphrase those summaries into a single narrative forgoes deeper comparison and analysis, reducing the depth of the synthesis."}, {"letter": "C", "text": "Feed all papers to Claude at once and request a bullet list of interesting quotes without any thematic grouping or structure.", "correct": false, "explanation": "Asking for a bullet list of interesting quotes without thematic grouping or structure limits Claude's ability to provide a coherent synthesis. This approach may produce a superficial and disorganized collection of excerpts rather than meaningful insights."}, {"letter": "D", "text": "Extract key findings from each paper individually, then synthesize them via comparison and contrast, and finally request a gap analysis.", "correct": true, "explanation": "Extracting key findings from each paper individually ensures reliable source-level capture before synthesis. Comparing and contrasting those findings, then requesting a gap analysis, yields a structured, comprehensive synthesis that leverages Claude's analytical strengths."}], "correct": "D", "select": 1}, {"id": "a4-108", "domain": 4, "task_id": "4.3", "objective": "Use Claude to support solution design, development, and iteration", "situation": "A team has revised its Claude-assisted reporting arrangement four times in two months. The latest version is worse for certain reports, and nobody can work out which change caused it.", "question": "What practice would have helped most?", "options": [{"letter": "A", "text": "Testing each revision on a larger sample of reports before it replaced the earlier version.", "correct": false, "explanation": "Wider testing would have caught the regression sooner, which is valuable. It still would not tell the team which change was responsible, so they could not undo the right one."}, {"letter": "B", "text": "Having a single person own all revisions so the changes are made consistently over time.", "correct": false, "explanation": "A single owner makes changes more consistent and is good practice. It does not help anyone reconstruct which of four revisions introduced the regression once the memory of them has faded."}, {"letter": "C", "text": "Limiting revisions to one per month so that each change has time to settle before the next.", "correct": false, "explanation": "Slowing the pace of change reduces how often revisions collide, but four undocumented changes over eight months would leave the team in the same position. The problem is that nothing was recorded, not that it happened quickly."}, {"letter": "D", "text": "Recording what changed at each revision and why, so a step backwards can be traced.", "correct": true, "explanation": "Iteration only compounds if each change is traceable. A short record of what was altered and why lets the team match the regression to the revision that caused it and roll back precisely instead of guessing."}], "correct": "D", "select": 1}, {"id": "a4-109", "domain": 4, "task_id": "4.4", "objective": "Integrate Claude into existing workflows to augment or redesign them", "situation": "A school office wants to bring Claude into its existing parental-enquiry workflow.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Map the workflow as it runs now; run it alongside the old way; choose one step to hand over; compare the two and then decide", "correct": false, "explanation": "Running the new way before deciding which step it replaces leaves the parallel run undirected, and the choice is then made after the evidence."}, {"letter": "B", "text": "Choose one step to hand over; map the workflow as it runs now; compare the two and then decide; run it alongside the old way", "correct": false, "explanation": "Choosing a step before the workflow is mapped means the choice is guesswork, and comparing before running has nothing to compare."}, {"letter": "C", "text": "Run it alongside the old way; compare the two and then decide; map the workflow as it runs now; choose one step to hand over", "correct": false, "explanation": "Running and comparing before the workflow is understood produces a result nobody can interpret, and the mapping arrives too late to matter."}, {"letter": "D", "text": "Map the workflow as it runs now; choose one step to hand over; run it alongside the old way; compare the two and then decide", "correct": true, "explanation": "Map what exists, choose one step, run the new way in parallel so nothing is lost, then compare before committing. Parallel running is what makes the comparison safe."}], "correct": "D", "select": 1}, {"id": "a4-110", "domain": 4, "task_id": "4.1", "objective": "Apply Claude to analyze requirements and use cases", "situation": "", "question": "Which task is most aligned with analyzing requirements and use cases for a time management application?", "options": [{"letter": "A", "text": "Ask Claude to start developing the software code for the time management application.", "correct": false, "explanation": "Starting to develop software code is a later implementation step, not part of analyzing requirements and use cases. Analysis should always precede coding to ensure the solution is grounded in user needs, and skipping this step risks building features without a clear problem understanding."}, {"letter": "B", "text": "Ask Claude to generate user personas and high-level use cases for a time management app.", "correct": true, "explanation": "Generating user personas and high-level use cases directly supports the analysis of requirements and use cases by clarifying who the app serves, what problems it solves, and which capabilities are most valuable. This foundational understanding is essential for designing an effective time management solution."}, {"letter": "C", "text": "Ask Claude to provide a formal definition and explanation of the key time management concepts.", "correct": false, "explanation": "Providing a formal definition and explanation of key concepts offers background knowledge, but it does not directly address the practical analysis of user requirements and use cases. This task is too abstract and academic, lacking the user-centered insights needed to design the application's features and workflows."}, {"letter": "D", "text": "Ask Claude to create a marketing plan with user personas for the time management app.", "correct": false, "explanation": "Creating a marketing plan with user personas is a marketing and go-to-market activity, not a step in analyzing requirements and use cases. While it may later inform promotion, it does not help define the core functionality or technical requirements of the application."}], "correct": "B", "select": 1}, {"id": "a4-111", "domain": 4, "task_id": "4.5", "objective": "Communicate Claude's value and limitations to stakeholders", "situation": "A hospital operations lead proposes letting Claude autonomously approve treatment plans without any clinician sign-off, citing faster patient throughput.", "question": "What is the most appropriate response to this proposal?", "options": [{"letter": "A", "text": "Approve the proposal as written, since faster throughput cuts wait times and Claude's output is reliable enough for sign-off, based on internal testing.", "correct": false, "explanation": "Removing clinician sign-off entirely introduces unacceptable patient-safety risks; no internal testing can fully guarantee Claude's reliability for unsupervised medical decisions."}, {"letter": "B", "text": "Recommend disabling Claude for any healthcare task, since its training data excludes recent treatment guidelines, leading to outdated recommendations.", "correct": false, "explanation": "Claude can still add significant value in clinical support roles, such as summarizing records or drafting plans for review; the concern here is unsupervised autonomy, not its usefulness in healthcare tasks."}, {"letter": "C", "text": "Recommend keeping a licensed clinician in the loop, since Claude can support analysis but should not make unsupervised medical decisions.", "correct": true, "explanation": "High-stakes medical decisions require a human expert in the loop; Claude can assist with analysis and accelerate workflows but should not make autonomous treatment approvals without clinician oversight."}, {"letter": "D", "text": "Recommend that clinicians only review treatment plans once per quarter, since periodic oversight ensures safety while minimizing clinician burden.", "correct": false, "explanation": "Reviewing treatment plans only once per quarter leaves the vast majority of decisions unsupervised for extended periods, failing to address the fundamental safety requirement of consistent clinician oversight."}], "correct": "C", "select": 1}, {"id": "a4-112", "domain": 4, "task_id": "4.1", "objective": "Apply Claude to analyze requirements and use cases", "situation": "A department is picking which parts of its work to hand to Claude first. Which two tasks are the stronger candidates for that first handover?", "question": "Select two.", "options": [{"letter": "A", "text": "Approving contract terms for the company without any lawyer reviewing what was agreed.", "correct": false, "explanation": "Binding the company to contract terms without legal review is both irreversible and outside what should be delegated."}, {"letter": "B", "text": "Sending payments out to suppliers automatically once an invoice arrives in the inbox.", "correct": false, "explanation": "Sending money is irreversible, which is the textbook sign that a step should not be handed over unattended."}, {"letter": "C", "text": "Drafting first-pass summaries of long vendor documents that a specialist then checks over.", "correct": true, "explanation": "Text-heavy first drafts with a specialist review afterwards are the classic good candidate: repetitive work, recoverable errors."}, {"letter": "D", "text": "Making the final call on which of the employees will be promoted in this year's review round.", "correct": false, "explanation": "Promotion decisions affect people's livelihoods and need human accountability, so Claude may inform but not decide."}, {"letter": "E", "text": "Reformatting a high volume of well-specified status updates into one agreed house template.", "correct": true, "explanation": "High-volume, well-specified reformatting is delegable precisely because the standard is clear and the output is checkable."}], "correct": ["C", "E"], "select": 2}, {"id": "a4-113", "domain": 4, "task_id": "4.1", "objective": "Apply Claude to analyze requirements and use cases", "situation": "", "question": "Reviewing a backlog of intake forms, an analyst notices that most entries name a solution rather than a need, for example \"we need a dashboard.\" How can Claude best help with this backlog?", "options": [{"letter": "A", "text": "Ask Claude to design the dashboard described in each entry so submitters can confirm what they wanted.", "correct": false, "explanation": "Designing what was asked for skips the analysis entirely. If the submitter needed a weekly exception alert rather than a dashboard, building the dashboard confirms only that both parties misread the need together."}, {"letter": "B", "text": "Ask Claude to reject entries that name a tool and ask the submitters to resubmit them in the right form.", "correct": false, "explanation": "Pushing the work back onto submitters is slow and often futile, since people describe needs in terms of solutions naturally. It also discards information the analyst already has in the entries."}, {"letter": "C", "text": "Ask Claude to group the entries by the department that submitted them and report the totals to sponsors.", "correct": false, "explanation": "Counting entries by department produces a useful management view but leaves every entry still expressed as a solution. The backlog would be tidier and no better understood."}, {"letter": "D", "text": "Ask Claude to restate each entry as the outcome sought, then flag entries where that outcome is unclear.", "correct": true, "explanation": "The analytical move is to convert stated solutions back into the outcomes behind them, which is language work Claude does well at volume. Flagging the entries where no outcome can be recovered tells the analyst exactly which submitters to go back to."}], "correct": "D", "select": 1}, {"id": "a4-114", "domain": 4, "task_id": "4.5", "objective": "Communicate Claude's value and limitations to stakeholders", "situation": "A stakeholder asks the solutions architect to define, in plain terms for a non-technical audience, what it means when Claude \"hallucinates\" during a research summarization task.", "question": "Which explanation is most accurate?", "options": [{"letter": "A", "text": "It means the input document was corrupted, containing garbled or missing sections, which is the only possible reason that factual errors could appear in a summary.", "correct": false, "explanation": "Hallucinations can happen even when the input document is perfectly clean and complete, so corruption is not the only or primary cause. The model may still generate plausible-sounding but incorrect information regardless of input quality."}, {"letter": "B", "text": "It means Claude can generate text that sounds plausible and confident but is factually incorrect or not supported by the source material.", "correct": true, "explanation": "Hallucination means the model generates text that sounds confident and plausible but is factually incorrect or not supported by the source material. This behavior underscores the importance of verifying Claude's outputs, especially in research summarization where accuracy is critical."}, {"letter": "C", "text": "It means Claude has detected a technical error in its own systems, such as a conflict in its training data, and automatically halted the response before completion.", "correct": false, "explanation": "Claude does not detect hallucinations as technical errors and halt responses; it is unaware of its own inaccuracies. Hallucination refers to generating incorrect content, not to system-level error detection."}, {"letter": "D", "text": "It means Claude is intentionally inserting false information, such as a made-up date or name, into the summary to test whether the user is paying close attention.", "correct": false, "explanation": "Claude does not intentionally insert false information as a test; hallucinations are unintended and arise from the model's generation process. The model has no awareness or intent behind its errors."}], "correct": "B", "select": 1}, {"id": "a4-115", "domain": 4, "task_id": "4.3", "objective": "Use Claude to support solution design, development, and iteration", "situation": "A finance manager is running a Cowork task that will reorganise files and send messages on her behalf.", "question": "Which approach best matches sound practice for work that has real consequences?", "options": [{"letter": "A", "text": "Stay close and approve the actions that send or change things, reviewing what Claude does.", "correct": true, "explanation": "No approval setting replaces the person's judgment, and this task touches things with real consequences. Staying close and approving the sending and changing actions keeps her in control of exactly the steps that cannot be walked back."}, {"letter": "B", "text": "Let the task run without any pauses, then review the whole session once it has finished completely.", "correct": false, "explanation": "Reviewing afterwards is fine for work that can be undone, but messages already sent in her name cannot be recalled. The review would document the damage rather than prevent it."}, {"letter": "C", "text": "Approve every action once at the outset so the task is not interrupted while it is running.", "correct": false, "explanation": "Blanket approval at the start grants permission for actions nobody has seen yet, which defeats the purpose of an approval step. Convenience is bought by removing the control the task most needs."}, {"letter": "D", "text": "Let the task run unattended overnight and check the results at the start of the next day.", "correct": false, "explanation": "Running unattended overnight maximises the time between a mistake and its discovery. Recipients will have read the messages long before anyone looks at what happened."}], "correct": "A", "select": 1}, {"id": "a4-116", "domain": 4, "task_id": "4.2", "objective": "Leverage Claude for research, planning, and process optimization", "situation": "A research team has delegated literature gathering and first-pass summarising to Claude.", "question": "Which part of the work should most clearly stay with the researchers themselves?", "options": [{"letter": "A", "text": "Deciding the order in which the summarised sources appear in the final written research output.", "correct": false, "explanation": "Ordering the sources is a presentation decision that can be delegated and then reviewed. Getting it wrong makes the output slightly harder to read and is corrected in a moment."}, {"letter": "B", "text": "Deciding which findings actually matter for this organisation's decision and what follows from them.", "correct": true, "explanation": "Gathering and summarising are volume tasks that suit delegation, but working out which findings bear on this organisation's decision requires context Claude does not hold and carries consequences the researchers must answer for."}, {"letter": "C", "text": "Deciding how many separate sources to gather before the first-pass summaries are considered complete.", "correct": false, "explanation": "How many sources to gather is a scoping rule that can be agreed once and applied mechanically. It matters, but it is far less consequential than deciding what the evidence means for the organisation."}, {"letter": "D", "text": "Deciding which file format the collected material should be stored in for the rest of the project.", "correct": false, "explanation": "File format is an administrative convention with no bearing on the quality of the conclusions. It is exactly the kind of low-stakes, easily specified choice that does not need senior attention."}], "correct": "B", "select": 1}, {"id": "a4-117", "domain": 4, "task_id": "4.2", "objective": "Leverage Claude for research, planning, and process optimization", "situation": "A market researcher needs Claude to incorporate current, real-world information published this week into a competitive-landscape summary, since this data postdates the model's training.", "question": "Which capability should the researcher enable?", "options": [{"letter": "A", "text": "Structured outputs, which guarantees the response conforms to a defined JSON schema", "correct": false, "explanation": "Structured outputs controls response formatting, not access to current external information."}, {"letter": "B", "text": "The web search tool, which augments Claude's knowledge with current data from across the web", "correct": true, "explanation": "The web search tool augments Claude's comprehensive knowledge with current, real-world data from across the web, which is needed for this week's information."}, {"letter": "C", "text": "Prompt caching, which reduces cost by reusing previously cached background context", "correct": false, "explanation": "Prompt caching lowers cost and latency for repeated context; it does not fetch new external information."}, {"letter": "D", "text": "The memory tool, which stores and retrieves information across separate conversations", "correct": false, "explanation": "The memory tool persists information across sessions but does not retrieve fresh external web content."}], "correct": "B", "select": 1}, {"id": "a4-118", "domain": 4, "task_id": "4.1", "objective": "Apply Claude to analyze requirements and use cases", "situation": "A project lead has a single sentence from a director: \"We need faster case handling.\" She wants to use Claude to turn this into a requirement set that goes to stakeholders for sign-off with no open questions left once it is drafted.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "List the open gaps; draft the requirement set; state the outcome and who is affected; take the gaps to stakeholders.", "correct": false, "explanation": "This option drafts the requirement set in step two, before the outcome is even stated in step three and before the gaps are ever taken to stakeholders in step four -- the draft goes out carrying the same open questions the project lead wanted resolved first, and the closing stakeholder step happens too late to change a draft that already exists."}, {"letter": "B", "text": "State the outcome and who is affected; list the open gaps; take the gaps to stakeholders; draft the requirement set.", "correct": true, "explanation": "Naming the outcome and who is affected turns the director's one sentence into a target the gaps can be measured against; listing what is still unknown surfaces the open questions; taking those to stakeholders closes them before anything is written down; only then does the draft go out with no open questions left, which is exactly what the project lead needs."}, {"letter": "C", "text": "Take the gaps to stakeholders; list the open gaps; draft the requirement set; state the outcome and who is affected.", "correct": false, "explanation": "This option takes gaps to stakeholders in step one, before any gap has been listed in step two, so there is nothing yet to ask stakeholders about; it also drafts the requirement set before the outcome and audience are named, so the draft is written without knowing what it is meant to achieve or for whom."}, {"letter": "D", "text": "Draft the requirement set; state the outcome and who is affected; list the open gaps; take the gaps to stakeholders.", "correct": false, "explanation": "The project lead wants a requirement set with no open questions left once it is drafted, so drafting cannot be the first step: nothing is known yet about the outcome, who is affected, or what is missing when this option's draft is written, and that draft is never revisited after the outcome is stated or the gaps are surfaced."}], "correct": "B", "select": 1}, {"id": "a4-119", "domain": 4, "task_id": "4.4", "objective": "Integrate Claude into existing workflows to augment or redesign them", "situation": "A finance shared-service team already has a working month-end checklist and wants to bring Claude into it without disrupting the close.", "question": "What should they do first?", "options": [{"letter": "A", "text": "Pick one checklist step that is text-heavy and repetitive and try Claude there", "correct": true, "explanation": "Augmenting a process that already works means starting with the smallest, most suitable slice, where a failure is contained and the rest of the close carries on unchanged."}, {"letter": "B", "text": "Ask Claude to rewrite the whole checklist before the next month-end close runs", "correct": false, "explanation": "Rewriting the checklist before anyone has tried a single step changes the whole close on an untested assumption."}, {"letter": "C", "text": "Replace the checklist with an arrangement built around what Claude can do now", "correct": false, "explanation": "Redesigning around the tool is the wrong move when the existing process works; it puts a hard deadline at risk for no demonstrated gain."}, {"letter": "D", "text": "Have every preparer use Claude however they prefer during the next close cycle", "correct": false, "explanation": "Uncoordinated individual use during a deadline-driven close produces inconsistent output and no learning the team can reuse."}], "correct": "A", "select": 1}, {"id": "a4-120", "domain": 4, "task_id": "4.2", "objective": "Leverage Claude for research, planning, and process optimization", "situation": "A manufacturing engineer has a detailed transcript of a production process and wants to convert it into a Standard Operating Procedure (SOP) using Claude.", "question": "Which approach best leverages Claude's capabilities to create an accurate, complete, and company-standard SOP?", "options": [{"letter": "A", "text": "Ask Claude to generate a generic manufacturing SOP, then insert specific details from the transcript into the template, relying on manual review to align steps with the process.", "correct": false, "explanation": "Starting with a generic SOP template and inserting details from the transcript can lead to misalignment between the template's structure and the actual process, requiring extensive manual review to correct. This approach fails to capitalize on Claude's capacity to directly generate a tailored, accurate SOP from the source material."}, {"letter": "B", "text": "Provide the transcript to Claude and request a complete SOP without specifying details, letting the model infer all necessary steps and formatting.", "correct": false, "explanation": "Providing the transcript without specifying details relies entirely on Claude's inference, which may overlook critical safety notes, tools, or quality checks necessary for a complete SOP. This approach also lacks iterative refinement to ensure the output matches company-specific formatting and content requirements, increasing the risk of omissions or inaccuracies."}, {"letter": "C", "text": "Ask Claude to extract steps from the transcript, then iteratively add safety notes, tools, and quality checks, and format as a company SOP template.", "correct": true, "explanation": "This method leverages Claude's capabilities by breaking the task into logical steps: extracting process steps, then iteratively adding safety notes, tools, and quality checks, and finally formatting into a company SOP template. This iterative process ensures the SOP is accurate, complete, and aligned with company standards."}, {"letter": "D", "text": "Manually rewrite the transcript into bullet points, then structure them into an SOP draft, and use Claude only to check grammar and style, not to suggest or organize content.", "correct": false, "explanation": "Limiting Claude to only grammar and style checks underutilizes its ability to analyze the transcript, suggest organizational improvements, and ensure procedural completeness. Manual rewriting and structuring risk introducing errors or missing details that Claude could otherwise catch and enhance."}], "correct": "C", "select": 1}, {"id": "a4-121", "domain": 4, "task_id": "4.2", "objective": "Leverage Claude for research, planning, and process optimization", "situation": "", "question": "What is the best approach for using Claude to create a detailed timeline from a marketing campaign brief?", "options": [{"letter": "A", "text": "Provide the brief, specify the exact launch date, ask for a backward-scheduled timeline with dependencies and responsible teams, and request that Claude flag any unrealistic constraints.", "correct": true, "explanation": "This approach anchors the timeline to the exact launch date, uses backward scheduling, and includes dependencies and responsible teams. Flagging unrealistic constraints allows for proactive adjustments, and grounding the plan in the provided brief ensures it is comprehensive and actionable. It fully leverages Claude for structured planning."}, {"letter": "B", "text": "Tell Claude to create a timeline based on typical marketing best practices, specify a generic campaign structure, ask for a forward-scheduled plan with standard phases, and ignore any specific dates from the brief.", "correct": false, "explanation": "Ignoring specific dates from the brief prevents the timeline from reflecting actual project constraints. A generic forward-scheduled plan based only on best practices lacks the necessary context to be actionable. Claude should incorporate the brief’s details to create a relevant and accurate timeline."}, {"letter": "C", "text": "Use Claude to rewrite the brief as a press release, specify the dateline and boilerplate, ask for a narrative flow, and then extract a timeline from the generated release for the campaign's key phases.", "correct": false, "explanation": "Rewriting the brief as a press release is an unnecessary transformation that doesn’t directly support timeline creation. Extracting a timeline from narrative prose risks losing key details and is less accurate than having Claude produce a structured schedule from the outset."}, {"letter": "D", "text": "Instruct Claude to list all possible marketing activities, categorize them by phase, and output a sequential plan without specific dates, relying on the team to schedule execution using their own project management tools.", "correct": false, "explanation": "Listing activities by phase without specific dates does not produce a workable timeline. Relying on the team to schedule execution later shifts critical planning decisions away from Claude, missing the opportunity to integrate temporal dependencies directly into the output."}], "correct": "A", "select": 1}, {"id": "a4-122", "domain": 4, "task_id": "4.4", "objective": "Integrate Claude into existing workflows to augment or redesign them", "situation": "A department switches all fifty of its staff to a redesigned, Claude-assisted case-handling process on the same Monday morning, with no trial group kept back.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The department's usage may rise sharply in the first week as fifty people start using Claude.", "correct": false, "explanation": "A jump in usage is a predictable consequence of fifty people starting at once and can be planned for. It affects cost and capacity, not the soundness of the redesigned process."}, {"letter": "B", "text": "Some staff will need considerably longer than others to become comfortable with the new way of working.", "correct": false, "explanation": "Uneven adoption is expected in any change and is managed with support and coaching. It slows the benefit rather than exposing the department to a failure it cannot see."}, {"letter": "C", "text": "Any flaw hits every case at once, and there is no unchanged group to compare results against.", "correct": true, "explanation": "A full switch removes both safeguards at once: every case is exposed to a design that has never run live, and with no group still on the old process there is nothing to compare against, so the team cannot even tell whether results have got worse."}, {"letter": "D", "text": "Training all fifty staff in one week will strain the two people who designed the new process.", "correct": false, "explanation": "Straining the two designers is a real resourcing problem during the first week. It is a delivery pressure rather than the reason a simultaneous switch is dangerous."}], "correct": "C", "select": 1}, {"id": "a4-123", "domain": 4, "task_id": "4.1", "objective": "Apply Claude to analyze requirements and use cases", "situation": "A council's improvement team has four candidate use cases and can take only one forward this quarter.", "question": "Which is the soundest first candidate for Claude?", "options": [{"letter": "A", "text": "Setting the annual budget allocation across all six of the council's neighbourhood teams", "correct": false, "explanation": "A budget allocation is a sole-source judgement with accountability attached, and the reasoning behind it must be owned by named people."}, {"letter": "B", "text": "Deciding which of the housing repair requests are downgraded to a lower priority band", "correct": false, "explanation": "Downgrading a repair affects a resident's service directly and needs judgement someone must answer for; Claude may inform that decision but should not make it."}, {"letter": "C", "text": "Summarising the free-text comments on the monthly residents' survey for service leads", "correct": true, "explanation": "It is repetitive, text-heavy and well specified, the output is reviewable before anyone acts on it, and a mistake is recoverable. That is the profile of a good delegation candidate."}, {"letter": "D", "text": "Issuing the final refusal letters for licence applications directly to the applicants", "correct": false, "explanation": "Sending a refusal to an applicant is an outward-facing, effectively irreversible act, which is precisely what should not be handed over unreviewed."}], "correct": "C", "select": 1}, {"id": "a4-124", "domain": 4, "task_id": "4.3", "objective": "Use Claude to support solution design, development, and iteration", "situation": "A Claude-assisted case-summary way of working produces good results for the analyst who designed it, but two colleagues get noticeably worse results from the same written instructions.", "question": "What should the designer do first?", "options": [{"letter": "A", "text": "Add examples of good summaries to the instructions so the expected standard is unmistakable.", "correct": false, "explanation": "Worked examples usually improve an arrangement and may well be part of the eventual fix. Adding them before the divergence has been observed is still a guess at the cause."}, {"letter": "B", "text": "Rewrite the instructions in much more detail so that less is left to the individual using them.", "correct": false, "explanation": "More detail is a plausible fix but an untargeted one. Without knowing which part of the colleagues' use diverges, the designer may add paragraphs about something that was never the problem."}, {"letter": "C", "text": "Ask the two colleagues to use the designer's own saved wording word for word every time.", "correct": false, "explanation": "Copying the designer's wording exactly hides the difference rather than explaining it, and it will not transfer to the next case. It also suggests the arrangement only works when one person operates it."}, {"letter": "D", "text": "Watch each colleague run the work once and see where their inputs differ from the designer's.", "correct": true, "explanation": "The same instructions producing different results means the difference is in what people actually do with them. Watching one run each is the quickest way to find the unstated step the designer performs without noticing, and it is what makes any later rewrite land in the right place."}], "correct": "D", "select": 1}, {"id": "a4-125", "domain": 4, "task_id": "4.3", "objective": "Use Claude to support solution design, development, and iteration", "situation": "A team is choosing which slice of work to use for a two-week trial of a new Claude-assisted contract summarising arrangement.", "question": "Which slice is the best choice for the trial?", "options": [{"letter": "A", "text": "The contracts causing the most delay right now, so the trial also clears the urgent backlog.", "correct": false, "explanation": "Using the trial to clear a backlog mixes two goals and puts urgent work in the hands of an untested arrangement. The most delayed contracts are also often the least typical, so they are a poor sample."}, {"letter": "B", "text": "A random sample drawn from the whole archive, including contracts that are no longer in force.", "correct": false, "explanation": "A random archive sample sounds rigorous but includes expired contracts the team no longer handles. Effort goes into cases that will never recur, and the awkward live cases may not appear at all."}, {"letter": "C", "text": "A representative mix, including the awkward cases the team knows the arrangement must survive.", "correct": true, "explanation": "A pilot is only useful if it is a fair rehearsal of the real work, which means deliberately including the hard cases. Two weeks on a representative mix shows both what the arrangement handles well and where it must keep a person involved."}, {"letter": "D", "text": "The simplest contracts, so the trial is not derailed by problems that affect only unusual cases.", "correct": false, "explanation": "An easy slice produces a flattering result that does not predict live performance. The arrangement would then meet its first difficult contract after the team has already committed to it."}], "correct": "C", "select": 1}, {"id": "a4-126", "domain": 4, "task_id": "4.1", "objective": "Apply Claude to analyze requirements and use cases", "situation": "A communications team has decided that Claude will draft its recurring stakeholder update, but the team cannot yet say how they will judge whether the arrangement is working.", "question": "Which action best prepares them to delegate the task?", "options": [{"letter": "A", "text": "Write down what an acceptable draft must contain and avoid, and agree it before any drafting is handed over.", "correct": true, "explanation": "Defining what good looks like before delegating is what makes the delegation reviewable: the team gains a shared standard to check drafts against, a basis for correcting the instructions, and evidence for whether the arrangement is working rather than an argument about taste."}, {"letter": "B", "text": "Agree a rota naming who will paste the source material into Claude for each reporting cycle in turn.", "correct": false, "explanation": "A rota settles who performs the task, which is useful operationally, but it does not give anyone a way to judge the output. The team would still have no agreed answer to whether a given draft is acceptable."}, {"letter": "C", "text": "Draw up a shortlist of other tools to fall back on if the drafts disappoint during the first month.", "correct": false, "explanation": "Planning a fallback before defining success means the team cannot tell whether the fallback is ever needed. Without a quality standard, disappointment is a matter of opinion and switching tools would not fix it."}, {"letter": "D", "text": "Set a target for how long each draft should take compared with the time the current manual version takes today.", "correct": false, "explanation": "A time target measures effort saved but says nothing about whether the drafts are fit to send. A fast draft that misses required content is worse than the slower manual version it replaced."}], "correct": "A", "select": 1}, {"id": "a4-127", "domain": 4, "task_id": "4.1", "objective": "Apply Claude to analyze requirements and use cases", "situation": "A programme team asks Claude to condense forty stakeholder interview transcripts into a single requirements list, then plans to send that list to a supplier as the final agreed scope without further review.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The supplier may raise follow-up questions that the programme team cannot answer without rereading material.", "correct": false, "explanation": "Follow-up questions from a supplier are normal in any scoping exercise and are answerable by returning to the material. This is an inconvenience rather than the central danger of treating an unverified summary as agreed scope."}, {"letter": "B", "text": "Working through forty long transcripts may take several minutes and consume a large share of the team's usage.", "correct": false, "explanation": "Processing time and usage consumption are real but recoverable costs. They do not affect the correctness of the scope being sent out, and they would not survive a comparison with a defect that only surfaces after the supplier has started building."}, {"letter": "C", "text": "Needs raised in only one interview may be lost, and nobody has checked the list against the transcripts.", "correct": true, "explanation": "Summarising forty transcripts naturally favours what was said repeatedly, so a genuine but rarely voiced requirement can quietly disappear. Because the list becomes contractual scope with no verification pass against the sources, an omission is both likely and expensive to discover later."}, {"letter": "D", "text": "The list may be written as flowing prose when the supplier's template expects a numbered table of items.", "correct": false, "explanation": "Formatting is a presentation issue that is trivially fixed by asking for a different layout. It has no bearing on whether the scope the supplier is being asked to build is actually the scope the stakeholders described."}], "correct": "C", "select": 1}, {"id": "a4-128", "domain": 4, "task_id": "4.4", "objective": "Integrate Claude into existing workflows to augment or redesign them", "situation": "A manager is about to tell her team that Claude will now produce the first draft of every incident report.", "question": "What should the announcement make clear above all?", "options": [{"letter": "A", "text": "How much time the change is expected to save the team over the course of a full year.", "correct": false, "explanation": "A projected annual saving is the business case rather than the operating instruction. It tells the team why the change is happening but leaves them unsure what they are now responsible for."}, {"letter": "B", "text": "What changes, what stays with each person, and where their review is still required.", "correct": true, "explanation": "People need to know exactly where the boundary now sits: which part Claude drafts, which judgments remain theirs, and that their review is still a required step. Without that, reports are either double-handled or sent out with nobody checking them."}, {"letter": "C", "text": "Which other departments have already adopted the same approach to their own reports.", "correct": false, "explanation": "Naming other departments builds confidence in the decision but says nothing about how this team should work. It is reassurance, not direction."}, {"letter": "D", "text": "How the manager tested the new arrangement before deciding to introduce it to the team.", "correct": false, "explanation": "Describing the testing is good practice and supports trust in the change. It still leaves the team without a clear statement of their own remaining responsibilities."}], "correct": "B", "select": 1}, {"id": "a4-129", "domain": 4, "task_id": "4.1", "objective": "Apply Claude to analyze requirements and use cases", "situation": "An operations manager has collected about sixty requirement statements from five departments for a new supplier onboarding process. Many overlap, several contradict each other, and some are one-line wishes. She wants Claude to help her make sense of the list.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Ask Claude to rank all sixty statements by business value so the team can start on the top ten items.", "correct": false, "explanation": "Ranking is premature while duplicates and contradictions are still in the list. A value ranking built over sixty unclean statements will rank the same need twice and score wishes that nobody can define, so the ordering would have to be redone once the list is tidied."}, {"letter": "B", "text": "Paste the full list into Claude and ask it to group related items and flag statements too vague to act on.", "correct": true, "explanation": "Analysis has to start with understanding what is actually in the list. Grouping related statements and surfacing the ones that are too vague to act on gives the manager a clean picture of overlaps, conflicts and gaps, which is exactly the work Claude accelerates well and which every later step depends on."}, {"letter": "C", "text": "Ask Claude to estimate how long each of the sixty requirements would take to deliver before reviewing.", "correct": false, "explanation": "Effort estimates depend on requirements being unambiguous. Estimating vague one-line wishes produces numbers with no basis, and the manager would be attaching delivery figures to items that may turn out to be duplicates of one another."}, {"letter": "D", "text": "Ask Claude to draft the finished onboarding process document so the departments can react to something real.", "correct": false, "explanation": "Drafting the finished process before the requirements are understood invites departments to react to Claude's assumptions rather than to their own stated needs, and it makes the analysis harder because disagreements now attach to a document instead of to the underlying requirement."}], "correct": "B", "select": 1}, {"id": "a4-130", "domain": 4, "task_id": "4.5", "objective": "Communicate Claude's value and limitations to stakeholders", "situation": "A department head asks for a one-page summary of what Claude will and will not do in a redesigned review workflow.", "question": "What is the most important thing for that summary to include?", "options": [{"letter": "A", "text": "A technical comparison of the model options so the department can judge the quality of the choice.", "correct": false, "explanation": "A model comparison is not what a department head needs to understand value and limits."}, {"letter": "B", "text": "A ranked list of every Claude feature the department could adopt over the next financial year.", "correct": false, "explanation": "A feature roadmap does not tell the department what the workflow will and will not do."}, {"letter": "C", "text": "A statement that the outputs are reliable enough for review to be dropped once staff are trained.", "correct": false, "explanation": "Promising that review disappears oversells the tool and removes the accountability step."}, {"letter": "D", "text": "Where Claude drafts and where a named person reviews, with the checks that stay in human hands.", "correct": true, "explanation": "Stakeholders need the boundary between assisted work and human accountability stated plainly."}], "correct": "D", "select": 1}, {"id": "a4-131", "domain": 4, "task_id": "4.1", "objective": "Apply Claude to analyze requirements and use cases", "situation": "A department head tells an analyst, \"I want Claude to handle our client onboarding.\" The analyst has been given no other detail.", "question": "What should the analyst do first?", "options": [{"letter": "A", "text": "Draw up a list of onboarding steps that Claude could not perform, and send that list to the department head.", "correct": false, "explanation": "Cataloguing limitations before the goal is known produces an answer to an unasked question. Some of those steps may be irrelevant to the department head's actual concern, and the list reads as resistance rather than analysis."}, {"letter": "B", "text": "Set up a shared project workspace and upload the current onboarding documents so Claude has background.", "correct": false, "explanation": "Loading the current documents is useful eventually, but doing it first assumes the existing process is the thing worth automating. If the real problem is that onboarding starts too late or asks clients for the wrong information, the documents encode the problem rather than the solution."}, {"letter": "C", "text": "Ask the department head which onboarding outcome is failing today, and for whom, before naming any tool.", "correct": true, "explanation": "The request names a tool, not a problem. Establishing which onboarding outcome is failing and who feels it defines the problem to be solved, and that definition determines which steps are worth delegating at all and how success would later be judged."}, {"letter": "D", "text": "Run a series of trial prompts to see how well Claude drafts onboarding emails, then share the results.", "correct": false, "explanation": "Trial prompts show what Claude can produce, not what the department needs. Demonstrating a capability before the problem is defined tends to anchor everyone on whichever step happens to demo well rather than on the step that is actually hurting."}], "correct": "C", "select": 1}, {"id": "a4-132", "domain": 4, "task_id": "4.4", "objective": "Integrate Claude into existing workflows to augment or redesign them", "situation": "A charity's donation-thank-you process works well overall but is slow only at the drafting step.", "question": "Which integration approach is most appropriate?", "options": [{"letter": "A", "text": "Augment the process by having Claude draft, with the same person signing off", "correct": true, "explanation": "When one step is the constraint and the rest of the process is sound, augment that step and leave the controls in place. The sign-off is not what was slow."}, {"letter": "B", "text": "Redesign the process so donors receive whatever Claude drafts straight away", "correct": false, "explanation": "Sending unreviewed drafts to donors is an outward-facing, effectively irreversible step and should not be delegated unchecked."}, {"letter": "C", "text": "Keep the process as it is and offer Claude as a tool people may use or ignore", "correct": false, "explanation": "Optional, unstructured use leaves the bottleneck exactly where it was and produces inconsistent letters to donors."}, {"letter": "D", "text": "Redesign the process around Claude and drop the sign-off step that slows it", "correct": false, "explanation": "Removing the sign-off removes the check that makes the process safe, in exchange for time that was not being lost there."}], "correct": "A", "select": 1}, {"id": "a4-133", "domain": 4, "task_id": "4.3", "objective": "Use Claude to support solution design, development, and iteration", "situation": "A pilot has worked well, and the team wants the new way of working to survive the departure of the person who created it.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Correct the instructions wherever they got stuck; have a second person run the work from those instructions alone; capture the working instructions in one shared place; write down what a good result looks like.", "correct": false, "explanation": "This begins by correcting instructions that do not yet exist in a shared place and ends by defining the standard, so every earlier step was taken without knowing what success meant."}, {"letter": "B", "text": "Write down what a good result looks like; capture the working instructions in one shared place; have a second person run the work from those instructions alone; correct the instructions wherever they got stuck.", "correct": true, "explanation": "The quality standard has to exist before anyone can tell whether a handover succeeded, the instructions must be captured somewhere shared to be handed over at all, a second person running from them alone is the real test, and their sticking points show precisely what to correct."}, {"letter": "C", "text": "Capture the working instructions in one shared place; correct the instructions wherever they got stuck; write down what a good result looks like; have a second person run the work from those instructions alone.", "correct": false, "explanation": "Correcting the instructions before anyone has tried to follow them means correcting imagined problems, and the standard for a good result is written after the corrections it should have guided."}, {"letter": "D", "text": "Have a second person run the work from those instructions alone; write down what a good result looks like; correct the instructions wherever they got stuck; capture the working instructions in one shared place.", "correct": false, "explanation": "The second person cannot run the work from instructions that have not yet been captured, and writing down what good looks like after the test leaves them no way to judge their own output."}], "correct": "B", "select": 1}, {"id": "a4-134", "domain": 4, "task_id": "4.4", "objective": "Integrate Claude into existing workflows to augment or redesign them", "situation": "A team plans to bring Claude into a workflow that three other departments feed into.", "question": "Which step is most important before the change goes live?", "options": [{"letter": "A", "text": "Wait until the other departments adopt the same arrangement themselves", "correct": false, "explanation": "Making the change conditional on three other departments adopting it first blocks a workflow improvement that is entirely within this team's control."}, {"letter": "B", "text": "Tell the feeding departments what changes for them and what does not", "correct": true, "explanation": "Integration touches everyone who feeds the workflow. Being explicit about what changes for them, and what stays the same, is what stops the redesign breaking at its handovers."}, {"letter": "C", "text": "Give each department its own version of the new written instructions", "correct": false, "explanation": "Separate instruction versions per department guarantee that the workflow behaves differently depending on who fed it."}, {"letter": "D", "text": "Ask the departments to stop sending work until the change is settled", "correct": false, "explanation": "Halting incoming work to make an internal change imposes the cost on other departments unnecessarily."}], "correct": "B", "select": 1}, {"id": "a5-001", "domain": 5, "task_id": "5.3", "objective": "Create effective system-level instructions", "situation": "A training manager is setting up a project so that every chat produces course outlines in the department's standard format.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Write the project instructions; refine the wording; define the audience and format; test with a real request.", "correct": false, "explanation": "Refining before the standard is defined means editing wording with no agreed target to edit towards."}, {"letter": "B", "text": "Define the audience and format; write the project instructions; test with a real request; refine the wording.", "correct": true, "explanation": "Agree the standard, encode it as durable direction, try it on real work, then tighten what missed."}, {"letter": "C", "text": "Refine the wording; test with a real request; write the project instructions; define the audience and format.", "correct": false, "explanation": "Nothing exists to refine at the start, and the format is agreed only after everything is already built."}, {"letter": "D", "text": "Test with a real request; define the audience and format; refine the wording; write the project instructions.", "correct": false, "explanation": "Testing before any instructions exist tells you nothing about whether the instructions work."}], "correct": "B", "select": 1}, {"id": "a5-002", "domain": 5, "task_id": "5.4", "objective": "Inform, maintain, and update Claude configurations, knowledge sources, and instructions", "situation": "A project shared across a department has no named owner. Several people hold edit access and each has added documents over the past year.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "Edit access is revoked automatically after a long period of inactivity, so the project becomes read-only.", "correct": false, "explanation": "Access is not withdrawn on its own because a project has been quiet."}, {"letter": "B", "text": "Shared projects reset their instructions whenever a second person saves a change to the knowledge.", "correct": false, "explanation": "Editing knowledge does not clear the project's instructions."}, {"letter": "C", "text": "Each new document replaces the previous one, so the project only ever holds the most recent upload.", "correct": false, "explanation": "Uploads accumulate; they do not overwrite one another."}, {"letter": "D", "text": "Nobody is accountable for removing outdated material, so contradictory documents accumulate unchecked.", "correct": true, "explanation": "Shared editing without ownership is precisely how a knowledge base fills with conflicting versions."}], "correct": "D", "select": 1}, {"id": "a5-003", "domain": 5, "task_id": "5.3", "objective": "Create effective system-level instructions", "situation": "An analyst has a specific quarterly business review deck outline that they need occasionally, in several different projects and sometimes in plain chats.", "question": "Where does this routine best belong?", "options": [{"letter": "A", "text": "Put it into every project's instructions, so the deck outline is imposed on all of the chats in each one.", "correct": false, "explanation": "This forces an occasional format onto every conversation in those projects and must be maintained in each."}, {"letter": "B", "text": "Keep it in a document inside one project's knowledge base and copy that file across into the other projects as needed.", "correct": false, "explanation": "Copying a file around leaves several versions to maintain and does not make the routine act on demand."}, {"letter": "C", "text": "Store it in the account-wide instructions, so the deck outline shapes every conversation the analyst has.", "correct": false, "explanation": "An occasional deck format would then influence unrelated work in every conversation."}, {"letter": "D", "text": "Package it as a skill, so the routine is available across projects and applies when the task calls for it.", "correct": true, "explanation": "Skills are reusable procedures that load when relevant, rather than always-on direction tied to one workspace."}], "correct": "D", "select": 1}, {"id": "a5-004", "domain": 5, "task_id": "5.3", "objective": "Create effective system-level instructions", "situation": "An operations lead runs a project whose instructions describe the team's standard weekly status report. This week only, a board member has asked for a one-page version.", "question": "What is the best way to handle the request?", "options": [{"letter": "A", "text": "Create a second project duplicating the same knowledge base but with one-page reports as its house rule.", "correct": false, "explanation": "A whole parallel workspace is heavy maintenance for a single request, and the two projects will drift apart."}, {"letter": "B", "text": "Add a permanent line to the project instructions saying every status report must be exactly one page long.", "correct": false, "explanation": "Baking a one-off requirement into durable direction changes every future report for a single exception."}, {"letter": "C", "text": "Ask for the one-page version in that chat, leaving the project instructions describing the normal report.", "correct": true, "explanation": "A one-off need belongs in the individual request; durable instructions should keep describing the standing format."}, {"letter": "D", "text": "Upload the board member's request into project knowledge so Claude applies the shorter format from now on.", "correct": false, "explanation": "Knowledge is background material, not direction, and this still makes a temporary need permanent."}], "correct": "C", "select": 1}, {"id": "a5-005", "domain": 5, "task_id": "5.4", "objective": "Inform, maintain, and update Claude configurations, knowledge sources, and instructions", "situation": "A shared project supports a process that changes every quarter.", "question": "Which practice best keeps the project trustworthy over time?", "options": [{"letter": "A", "text": "Name an owner and a review date so the instructions and the documents are checked each quarter.", "correct": true, "explanation": "Accountability plus a scheduled review is what keeps configuration aligned with a changing process."}, {"letter": "B", "text": "Re-upload every document once a week so that the knowledge base is refreshed whether or not it changed.", "correct": false, "explanation": "Re-uploading unchanged files costs effort without detecting anything that actually changed."}, {"letter": "C", "text": "Ask anyone who spots an error to correct the knowledge base themselves whenever they happen to notice.", "correct": false, "explanation": "Ad hoc fixes by whoever notices leave gaps and produce inconsistent, unrecorded changes."}, {"letter": "D", "text": "Rely on members to say in their prompts which parts of the documents are now out of date.", "correct": false, "explanation": "This assumes every member already knows what changed, which is rarely true."}], "correct": "A", "select": 1}, {"id": "a5-006", "domain": 5, "task_id": "5.4", "objective": "Inform, maintain, and update Claude configurations, knowledge sources, and instructions", "situation": "A logistics coordinator depends on a carrier rate sheet that the finance team edits every week.", "question": "Which maintenance approach is better?", "options": [{"letter": "A", "text": "Upload one copy of the rate sheet, since an uploaded copy follows the original whenever it changes.", "correct": false, "explanation": "An uploaded document is a snapshot; it does not update when the original is edited."}, {"letter": "B", "text": "Keep the rate sheet reachable through the connected drive so Claude works from the current version.", "correct": true, "explanation": "A connected source reflects the file as it stands, which removes a weekly manual refresh."}, {"letter": "C", "text": "Ask Claude to remember the rates in one chat and to reuse those figures in all later chats.", "correct": false, "explanation": "Detail mentioned in one conversation is not carried into other chats as project context."}, {"letter": "D", "text": "Paste the rate figures into the project instructions, where they are least likely to be edited by others.", "correct": false, "explanation": "Figures that change weekly are the worst possible content for durable instructions."}], "correct": "B", "select": 1}, {"id": "a5-007", "domain": 5, "task_id": "5.4", "objective": "Inform, maintain, and update Claude configurations, knowledge sources, and instructions", "situation": "A project owner has been sent the replacement version of a document that is already in the project's knowledge base.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Test a question whose answer changed; remove the superseded file; upload the replacement; identify the outdated document.", "correct": false, "explanation": "Verification happens before anything has been fixed, and identification comes last of all."}, {"letter": "B", "text": "Upload the replacement; test a question whose answer changed; identify the outdated document; remove the superseded file.", "correct": false, "explanation": "Testing while both versions are loaded produces an unreliable result."}, {"letter": "C", "text": "Identify the outdated document; upload the replacement; remove the superseded file; test a question whose answer changed.", "correct": true, "explanation": "Find it, replace it, remove the old copy so both cannot be cited, then verify with a question that changed."}, {"letter": "D", "text": "Remove the superseded file; identify the outdated document; test a question whose answer changed; upload the replacement.", "correct": false, "explanation": "The file is deleted before it has been identified, and the project has no current source during the test."}], "correct": "C", "select": 1}, {"id": "a5-008", "domain": 5, "task_id": "5.1", "objective": "Configure Claude Projects with instructions and knowledge sources", "situation": "A manager wants Claude to help draft a weekly customer newsletter, always drawing on the brand guide, the past issues and a fixed tone.", "question": "What should the manager do first?", "options": [{"letter": "A", "text": "Paste the brand guide into a single long chat and keep returning to that same conversation each week.", "correct": false, "explanation": "One growing conversation is fragile and does not carry the direction into any new chat."}, {"letter": "B", "text": "Ask Claude for a first newsletter and add the brand guide only if the draft misses the intended tone.", "correct": false, "explanation": "This defers the setup and produces a first draft with none of the context that matters."}, {"letter": "C", "text": "Put the brand guide into the account-wide instructions so that it shapes every conversation they ever have.", "correct": false, "explanation": "Newsletter-specific material would then influence unrelated work across the whole account."}, {"letter": "D", "text": "Create a project for the newsletter work so the guide, the past issues and the tone rules sit together.", "correct": true, "explanation": "Recurring work with shared context and standing rules is exactly what a project is for."}], "correct": "D", "select": 1}, {"id": "a5-009", "domain": 5, "task_id": "5.1", "objective": "Configure Claude Projects with instructions and knowledge sources", "situation": "A support team lead creates a Claude Project and wants Claude to always answer in a formal tone and respond as a senior support engineer for every conversation in that workspace, without repeating those directions in each chat.", "question": "What should the lead configure?", "options": [{"letter": "A", "text": "Upload a text file describing the tone to the knowledge base and name the file 'instructions.txt'", "correct": false, "explanation": "A file named like instructions is still treated as knowledge base content, not as project-level behavioral instructions."}, {"letter": "B", "text": "Create a separate project for every tone and role combination the support team might ever need", "correct": false, "explanation": "Creating a project per tone combination is unnecessary overhead when a single project's instructions field already covers this need."}, {"letter": "C", "text": "Paste the tone and role guidance into the opening message of each new chat inside the project", "correct": false, "explanation": "Repeating guidance in every chat defeats the purpose of project instructions, which persist automatically across chats."}, {"letter": "D", "text": "Set project instructions defining the tone and role, since these apply automatically to every chat", "correct": true, "explanation": "Project instructions are built specifically to establish behavioral guidelines, such as tone and role, that Claude applies to all chats within that project."}], "correct": "D", "select": 1}, {"id": "a5-010", "domain": 5, "task_id": "5.2", "objective": "Manage uploaded knowledge and connectors (e.g., Google Drive, Gmail)", "situation": "Claude keeps quoting a superseded pricing sheet that someone uploaded to a shared Project several months ago. Which two actions address the cause?", "question": "Select two.", "options": [{"letter": "A", "text": "Remove the superseded pricing sheet from the Project's knowledge so it stops being used.", "correct": true, "explanation": "The stale document is the cause, so removing it from the knowledge base is what actually stops the wrong figures."}, {"letter": "B", "text": "Switch the Project over to a deeper-reasoning tier so the wrong figures get caught in future.", "correct": false, "explanation": "A deeper tier reasons better over whatever it is given, and it is still being given the superseded sheet."}, {"letter": "C", "text": "Upload the current pricing sheet so that the Project draws on the version now in force.", "correct": true, "explanation": "Replacing it with the current sheet gives the Project the right source, which is the other half of the same fix."}, {"letter": "D", "text": "Add a line to every chat telling Claude to ignore any pricing figure that looks wrong.", "correct": false, "explanation": "Re-prompting cannot fix a wrong source document; Claude has no way to know which figure is out of date."}, {"letter": "E", "text": "Start a brand new Project every month so that old uploads never have time to accumulate.", "correct": false, "explanation": "Discarding the workspace monthly throws away the accumulated context and still does not curate the sources."}], "correct": ["A", "C"], "select": 2}, {"id": "a5-011", "domain": 5, "task_id": "5.1", "objective": "Configure Claude Projects with instructions and knowledge sources", "situation": "", "question": "Which statement accurately reflects how a project's chat history relates to that project's workspace?", "options": [{"letter": "A", "text": "Chat history is automatically discarded at the end of every chat session within a project", "correct": false, "explanation": "Chat history is retained as part of the project workspace, not discarded automatically at session end."}, {"letter": "B", "text": "Each project is a self-contained workspace that maintains its own chat history apart from other projects", "correct": true, "explanation": "Projects function as self-contained workspaces, each maintaining its own separate chat history and knowledge base distinct from other projects."}, {"letter": "C", "text": "Chat history is stored globally across the account and is identical no matter which project it belongs to", "correct": false, "explanation": "Chat history is scoped per project rather than merged into one identical global history across all projects."}, {"letter": "D", "text": "Chat history can only be viewed after re-uploading it to the project's knowledge base first", "correct": false, "explanation": "Chat history is viewable directly within the project without needing to be uploaded to the knowledge base."}], "correct": "B", "select": 1}, {"id": "a5-012", "domain": 5, "task_id": "5.2", "objective": "Manage uploaded knowledge and connectors (e.g., Google Drive, Gmail)", "situation": "An analyst has connected her work Google Drive to Claude and asks for last quarter's regional sales file. Claude replies that it cannot find it, although she knows the file exists because a colleague showed it to her.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Rephrase the request repeatedly until Claude eventually locates the file on the connected drive.", "correct": false, "explanation": "Rewording cannot reveal a file the connected account has no permission to open in the first place."}, {"letter": "B", "text": "Remove the connector and set it up again, as a broken connection is the usual cause of this.", "correct": false, "explanation": "Reconnecting does not widen access, and nothing here suggests the connection itself has failed."}, {"letter": "C", "text": "Check whether that file is shared with her own account, since the connector uses her own access.", "correct": true, "explanation": "A connector reaches only what the connecting account can already open, so a file a colleague has not shared with her is simply outside its scope."}, {"letter": "D", "text": "Ask her colleague to run the same request and paste his answer back into her conversation.", "correct": false, "explanation": "This works around the question without establishing whether she is entitled to the file at all."}], "correct": "C", "select": 1}, {"id": "a5-013", "domain": 5, "task_id": "5.4", "objective": "Inform, maintain, and update Claude configurations, knowledge sources, and instructions", "situation": "A reorganisation has changed how a team's workflow runs, and a shared project's instructions no longer describe it.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Revise the project instructions; gather what changed in the process; tell the members; check the outputs against the revised rules.", "correct": false, "explanation": "The instructions are rewritten before anyone has established what actually changed."}, {"letter": "B", "text": "Check the outputs against the revised rules; tell the members; revise the project instructions; gather what changed in the process.", "correct": false, "explanation": "Outputs are checked against rules that do not exist yet, and the change is investigated last."}, {"letter": "C", "text": "Gather what changed in the process; revise the project instructions; check the outputs against the revised rules; tell the members.", "correct": true, "explanation": "Understand the change, encode it, confirm it produces the right output, then inform the people who rely on it."}, {"letter": "D", "text": "Tell the members; check the outputs against the revised rules; gather what changed in the process; revise the project instructions.", "correct": false, "explanation": "Members are told about a change that has not been made, and the revision happens last."}], "correct": "C", "select": 1}, {"id": "a5-014", "domain": 5, "task_id": "5.2", "objective": "Manage uploaded knowledge and connectors (e.g., Google Drive, Gmail)", "situation": "An employee keeps a connector to a shared team drive in place after moving to a completely different department.", "question": "What is the greatest concern?", "options": [{"letter": "A", "text": "The drive's files are duplicated into the employee's own personal account and count against their storage.", "correct": false, "explanation": "Connectors do not copy files into the individual's own storage."}, {"letter": "B", "text": "Claude keeps reaching material the person no longer needs, so access is broader than their current role.", "correct": true, "explanation": "A connection left in place after a role change quietly extends access beyond what the job now requires."}, {"letter": "C", "text": "The organisation's owner is blocked from enabling any further connectors until this one is removed.", "correct": false, "explanation": "One connection does not prevent others from being enabled."}, {"letter": "D", "text": "The connector stops working entirely, so any project relying on it returns an error to every member.", "correct": false, "explanation": "A department move does not by itself break a connector."}], "correct": "B", "select": 1}, {"id": "a5-015", "domain": 5, "task_id": "5.2", "objective": "Manage uploaded knowledge and connectors (e.g., Google Drive, Gmail)", "situation": "A user asks Claude, connected via the Drive connector, to save a report Claude just generated into a specific Drive folder.", "question": "What can Claude do?", "options": [{"letter": "A", "text": "Claude can save files to Drive only when code execution is separately turned on first.", "correct": false, "explanation": "Code execution is required for XLSX chat uploads, not for the Drive connector's save and folder features."}, {"letter": "B", "text": "Claude can only save generated files to the root of Drive, not a folder.", "correct": false, "explanation": "Folder creation and organized saving are supported, not just root-level file placement."}, {"letter": "C", "text": "Claude can create folders and save Claude-generated files directly into Drive itself.", "correct": true, "explanation": "The Drive connector supports creating folders and saving Claude-generated files, so the report can go into the requested folder."}, {"letter": "D", "text": "Claude cannot save any files to Drive, only read files already there.", "correct": false, "explanation": "Saving generated files to Drive is an explicit connector capability beyond just reading existing files."}], "correct": "C", "select": 1}, {"id": "a5-016", "domain": 5, "task_id": "5.1", "objective": "Configure Claude Projects with instructions and knowledge sources", "situation": "A researcher mentions in a chat that they prefer citations in APA format, but does not add this preference to the project knowledge base or project instructions. In a different chat within the same project, Claude does not apply APA formatting by default.", "question": "Why does this occur?", "options": [{"letter": "A", "text": "Claude supports only one citation style per account for the entire lifetime of that account", "correct": false, "explanation": "Citation style is not fixed for the lifetime of an account; it can be guided per project through instructions."}, {"letter": "B", "text": "The researcher's current plan does not support citation formatting inside project chats at all", "correct": false, "explanation": "This behavior reflects how context persistence works, not a plan-based restriction on formatting styles."}, {"letter": "C", "text": "Preferences mentioned only in a single chat are not shared across chats unless added to instructions", "correct": true, "explanation": "Context shared in one chat is not shared across chats within a project unless it is captured in project instructions or added to the knowledge base."}, {"letter": "D", "text": "APA formatting is a feature that must be enabled separately in account-wide settings outside projects", "correct": false, "explanation": "There is no separate account-wide setting for citation style; persistence within a project is controlled through instructions or knowledge base content."}], "correct": "C", "select": 1}, {"id": "a5-017", "domain": 5, "task_id": "5.1", "objective": "Configure Claude Projects with instructions and knowledge sources", "situation": "A user on a free Claude account has already created five projects and wants to start a sixth one for a new personal task.", "question": "What should they expect?", "options": [{"letter": "A", "text": "They can create a sixth project only after deleting the oldest chat from an existing project", "correct": false, "explanation": "Deleting a chat does not lift the five-project cap; the limit applies to the count of projects, not chats."}, {"letter": "B", "text": "They can create unlimited projects because free accounts carry no cap on project count at all", "correct": false, "explanation": "Free accounts are capped at five projects; the cap is not unlimited."}, {"letter": "C", "text": "They can create a sixth project but that project will have no knowledge base feature available", "correct": false, "explanation": "The limitation described is about the number of projects allowed, not about disabling the knowledge base feature on new ones."}, {"letter": "D", "text": "They cannot create a sixth project because free accounts are limited to a maximum of five projects", "correct": true, "explanation": "Free Claude accounts can create a maximum of five projects, so a sixth would require removing or upgrading rather than simply creating a new one."}], "correct": "D", "select": 1}, {"id": "a5-018", "domain": 5, "task_id": "5.3", "objective": "Create effective system-level instructions", "situation": "An HR partner finds that a project is producing answers in the wrong tone for its audience.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Identify the rule that is vague or missing; read what the instructions currently say; rerun a typical request; revise that rule.", "correct": false, "explanation": "A rule cannot be identified as weak before the instructions have been read, and the fix comes last."}, {"letter": "B", "text": "Revise that rule; rerun a typical request; read what the instructions currently say; identify the rule that is vague or missing.", "correct": false, "explanation": "Revising before reading means editing direction nobody has actually examined."}, {"letter": "C", "text": "Rerun a typical request; revise that rule; read what the instructions currently say; identify the rule that is vague or missing.", "correct": false, "explanation": "The rule is revised before it has been identified, and the instructions are read last of all."}, {"letter": "D", "text": "Read what the instructions currently say; identify the rule that is vague or missing; revise that rule; rerun a typical request.", "correct": true, "explanation": "Diagnose the existing direction, isolate the weak rule, fix it, then confirm with a real request."}], "correct": "D", "select": 1}, {"id": "a5-019", "domain": 5, "task_id": "5.4", "objective": "Inform, maintain, and update Claude configurations, knowledge sources, and instructions", "situation": "An HR project built eighteen months ago is still in daily use. Its benefits summary has since been superseded by a newer edition, and one uploaded policy was retired outright in a restructure. Which two actions should the project owner take?", "question": "Select two.", "options": [{"letter": "A", "text": "Remove the retired policy, because an obsolete document still gets used in answers.", "correct": true, "explanation": "and this is a different maintenance action on a different document: nothing replaces the retired policy, so it is removed rather than refreshed. Claude has no way to know a policy no longer applies unless it stops being available."}, {"letter": "B", "text": "Ask Claude which uploaded documents are out of date and act on whatever it reports.", "correct": false, "explanation": "Claude can compare documents it has been given, but it cannot know which reflects current company reality. Delegating the currency judgment to the model is precisely the human review step that project maintenance requires."}, {"letter": "C", "text": "Upload the current benefits summary and delete the superseded edition it replaces.", "correct": true, "explanation": "Refreshing the document is the fix for a superseded source, and deleting the old edition matters as much as adding the new one — left in place, it competes with the current figures and produces confidently wrong answers."}, {"letter": "D", "text": "Add an instruction telling Claude to disregard any document older than twelve months.", "correct": false, "explanation": "Uploaded files do not carry a reliable age Claude can filter on, and the rule would be wrong in both directions — discarding accurate long-standing policies while keeping a recently uploaded document that is already superseded."}, {"letter": "E", "text": "Keep both benefits editions in place so Claude can compare the old and new wording.", "correct": false, "explanation": "Two live editions of the same summary is the situation that causes staff to be quoted last year's benefits. If a historical version is genuinely needed, it belongs outside the project knowledge that answers everyday questions."}], "correct": ["A", "C"], "select": 2}, {"id": "a5-020", "domain": 5, "task_id": "5.2", "objective": "Manage uploaded knowledge and connectors (e.g., Google Drive, Gmail)", "situation": "A manager asks Claude, connected via the Calendar connector, to find a time when five attendees are all free next week.", "question": "What can Claude do?", "options": [{"letter": "A", "text": "Claude can identify mutual availability across attendees using the connector directly.", "correct": true, "explanation": "The Calendar connector explicitly supports identifying mutual availability across attendees, finding a shared open slot for all five."}, {"letter": "B", "text": "Claude can list each attendee's events but cannot compute their overlapping free time.", "correct": false, "explanation": "Computing mutual availability, not just listing separate events, is specifically supported."}, {"letter": "C", "text": "Claude can only check the manager's own calendar, not the attendees' ones.", "correct": false, "explanation": "The connector is designed to view shared calendars and attendee availability, not just the manager's events."}, {"letter": "D", "text": "Claude cannot access calendar data for anyone besides the account owner.", "correct": false, "explanation": "Access to shared calendars for availability checks is a core connector feature."}], "correct": "A", "select": 1}, {"id": "a5-021", "domain": 5, "task_id": "5.2", "objective": "Manage uploaded knowledge and connectors (e.g., Google Drive, Gmail)", "situation": "A user has connected the Gmail connector and asks Claude to reply to a client email.", "question": "What happens once Claude has drafted the reply?", "options": [{"letter": "A", "text": "Claude holds the message for a workspace administrator to approve before anything leaves the account", "correct": false, "explanation": "The approval prompt goes to the person making the request, not to an administrator. Administrators control whether a connector is available at all, not individual messages."}, {"letter": "B", "text": "Claude sends it straight away, because authorising a connector authorises every action that connector offers", "correct": false, "explanation": "Authorising a connector does not hand over unattended permission to act. Write actions are gated by an approval step by default, which is the point of that design."}, {"letter": "C", "text": "Claude can send it, but by default it asks the user to approve the action before the message goes out", "correct": true, "explanation": "Sending, replying and forwarding are write actions the mail connector supports, and by default Claude asks for approval before each one. The user stays in control of what actually leaves the account without having to copy the draft out by hand."}, {"letter": "D", "text": "Claude can only produce a draft, because a mail connector is limited to searching and reading messages", "correct": false, "explanation": "Mail connectors are not read-only. Alongside searching and reading, they support sending, replying to and forwarding messages."}], "correct": "C", "select": 1}, {"id": "a5-022", "domain": 5, "task_id": "5.1", "objective": "Configure Claude Projects with instructions and knowledge sources", "situation": "A bid team on a Team plan is setting up a shared project for public-sector tender responses. Two writers will draft in it and four managers will review. Which two set-up choices are appropriate?", "question": "Select two.", "options": [{"letter": "A", "text": "Paste the tender house rules at the top of each new chat instead of storing them.", "correct": false, "explanation": "House rules that apply to every bid are exactly what project instructions are for. Re-pasting them into each chat means they get shortened, misremembered or forgotten, and different writers end up working to different rules."}, {"letter": "B", "text": "Upload the approved boilerplate and the last three winning tenders as knowledge.", "correct": true, "explanation": "Project knowledge is where durable background belongs: uploading the boilerplate and past winning tenders makes that material available across every chat in the project, instead of being re-pasted per bid."}, {"letter": "C", "text": "Have every writer make a personal copy of the project so nobody waits for anyone.", "correct": false, "explanation": "Personal copies destroy the shared context the project exists to provide: the separate knowledge bases drift apart, and every boilerplate update has to be repeated in each one. A shared project already lets people work at the same time."}, {"letter": "D", "text": "Ask an administrator to lift the five-project limit that applies to the bid team.", "correct": false, "explanation": "The cap of five projects is a Free-account limit. This team is on a Team plan, which has no stated project limit, so there is no cap for an administrator to lift."}, {"letter": "E", "text": "Share it with the two writers as \"Can edit\" and the four managers as \"Can view\".", "correct": true, "explanation": "and on a separate axis from knowledge: this is the access decision. Project sharing is a Team and Enterprise capability, and the creator grants specific members either \"Can view\" or \"Can edit\" — those are the two levels that exist."}], "correct": ["B", "E"], "select": 2}, {"id": "a5-023", "domain": 5, "task_id": "5.1", "objective": "Configure Claude Projects with instructions and knowledge sources", "situation": "A consultant is setting up a Claude Project for a client engagement she will work on every week for months. Which two things belong in that set-up?", "question": "Select two.", "options": [{"letter": "A", "text": "Project knowledge holding the client documents that each chat in the Project should draw on.", "correct": true, "explanation": "Project knowledge is where background documents live so Claude has them in every chat in that Project."}, {"letter": "B", "text": "One chat for each client document, so every document keeps its own conversation thread.", "correct": false, "explanation": "Splitting one document per chat fragments the engagement and abandons the shared knowledge base a Project provides."}, {"letter": "C", "text": "Project instructions covering the tone, role and house rules every chat there should follow.", "correct": true, "explanation": "Project instructions are the durable, project-scoped direction that shapes tone, role and house rules across every chat there."}, {"letter": "D", "text": "A note asking Claude to remember these client details for later use in her unrelated Projects.", "correct": false, "explanation": "A Project is self-contained, so its context does not spread into other Projects, and asking for that does not create it."}, {"letter": "E", "text": "One very long opening message that every later chat is expected to scroll back through.", "correct": false, "explanation": "A long opening message is a one-off prompt in one chat, not durable configuration for the Project."}], "correct": ["A", "C"], "select": 2}, {"id": "a5-024", "domain": 5, "task_id": "5.2", "objective": "Manage uploaded knowledge and connectors (e.g., Google Drive, Gmail)", "situation": "A programme manager connects her Gmail account to a private Claude project so she can draft client status updates using recent correspondence. She writes a draft inside that project and then sends it in an email to an external partner agency the programme works with.", "question": "What is the greatest risk in this workflow?", "options": [{"letter": "A", "text": "The partner agency gains the same access to her mailbox the moment she sends the draft, because the connector automatically reaches everyone the email goes to.", "correct": false, "explanation": "Wrong. A connector is authorised for her own account and enforced at the user level; sending an email to the partner agency does not extend that authorisation to them, and the project she is working in was never shared with the partner because connectors are only available in private projects."}, {"letter": "B", "text": "Details from messages unrelated to the update can surface in the draft, and she carries them outside the organisation when she sends it.", "correct": true, "explanation": "Claude can surface anything it finds relevant while searching her mailbox, so material from messages that have nothing to do with the update can end up woven into the draft. Once she emails that draft to the partner agency, whatever slipped in leaves the organisation with it."}, {"letter": "C", "text": "Connecting her mailbox pulls every message into the project's knowledge base as if each one had been uploaded.", "correct": false, "explanation": "Wrong. Claude draws on a connected mailbox only when something in it is relevant to the request being handled, rather than importing every message wholesale into the project's knowledge base."}, {"letter": "D", "text": "The mailbox connector stops working the moment she copies text out of the project chat into an email.", "correct": false, "explanation": "Wrong. The connector stays authorised on her account independently of what she does with the content it surfaces; copying text out of a project chat has no effect on the connection itself."}], "correct": "B", "select": 1}, {"id": "a5-025", "domain": 5, "task_id": "5.4", "objective": "Inform, maintain, and update Claude configurations, knowledge sources, and instructions", "situation": "A weekly digest produced with help from a project turns out to have been based on a source that has since been found unreliable.", "question": "What should the person who publishes the digest do?", "options": [{"letter": "A", "text": "Tell the recipients which past digests are affected and correct the source before the next one.", "correct": true, "explanation": "Maintenance includes informing the people who already acted on the earlier output."}, {"letter": "B", "text": "Add a standing disclaimer to all future digests saying that any figure in them may be inaccurate.", "correct": false, "explanation": "A permanent disclaimer neither corrects the past nor restores confidence in the future."}, {"letter": "C", "text": "Stop producing the digest, since a source problem means the workflow can never be trusted again.", "correct": false, "explanation": "One bad source is a maintenance failure, not a reason to abandon a working process."}, {"letter": "D", "text": "Quietly swap the source and let the correction show up in future digests without further comment.", "correct": false, "explanation": "Recipients keep relying on earlier digests they have no reason to doubt."}], "correct": "A", "select": 1}, {"id": "a5-026", "domain": 5, "task_id": "5.2", "objective": "Manage uploaded knowledge and connectors (e.g., Google Drive, Gmail)", "situation": "A team is about to connect a shared Google Drive to Claude for the first time. Which two considerations matter most before they do it?", "question": "Select two.", "options": [{"letter": "A", "text": "Whether organisational policy requires an approval before any data source is connected.", "correct": true, "explanation": "Most organisations require approval before a data source is connected, and that governance step precedes the technical one."}, {"letter": "B", "text": "Whether connecting a drive turns the files it holds into Project instructions as well.", "correct": false, "explanation": "Connected files are a knowledge source; Project instructions are separate configuration that a person writes."}, {"letter": "C", "text": "What the connection actually lets Claude see, so that the scope can be narrowed beforehand.", "correct": true, "explanation": "A connector is a permission-and-scope decision, so knowing and narrowing what Claude can reach comes first."}, {"letter": "D", "text": "Whether connecting up the drive will permanently increase the size of the context window used.", "correct": false, "explanation": "A connector changes what Claude can reach, not the size of the context window, which is a property of the model."}, {"letter": "E", "text": "Whether the connector removes the need to review the answers drawn from those files.", "correct": false, "explanation": "Material reaching Claude through a connector still needs the same scrutiny as any other source before it is relied on."}], "correct": ["A", "C"], "select": 2}, {"id": "a5-027", "domain": 5, "task_id": "5.2", "objective": "Manage uploaded knowledge and connectors (e.g., Google Drive, Gmail)", "situation": "A user has decided they no longer want Claude to be able to reach a connected service.", "question": "What should they do first?", "options": [{"letter": "A", "text": "Change the password on the service, which removes it from the connector list automatically.", "correct": false, "explanation": "A password change is not how a connection is managed, and it disrupts other tools too."}, {"letter": "B", "text": "Disconnect the service in the connector settings so that Claude can no longer reach it.", "correct": true, "explanation": "Removing the connection is the direct control, and it takes effect for every conversation."}, {"letter": "C", "text": "Delete the projects that used the service so that the connection has nothing left to read.", "correct": false, "explanation": "The connection remains live and reachable from any new conversation."}, {"letter": "D", "text": "Ask Claude in each new chat to avoid using that service when it is answering questions.", "correct": false, "explanation": "A per-chat request depends on being remembered every time and leaves the access in place."}], "correct": "B", "select": 1}, {"id": "a5-028", "domain": 5, "task_id": "5.1", "objective": "Configure Claude Projects with instructions and knowledge sources", "situation": "An HR coordinator uploads a policy handbook to Project A's knowledge base, so the document resides only in that project. Later, the coordinator starts a new, unrelated 'Vendor Contracts' Project B without any further action.", "question": "What is the handbook's availability in Project B?", "options": [{"letter": "A", "text": "The handbook is not available in Project B, since each project's knowledge base is scoped to itself alone", "correct": true, "explanation": "Anthropic's Claude platform offers Projects as self-contained workspaces, each with its own knowledge base and chat history. Uploaded files like a handbook are confined to the project they were added to and are not automatically shared across other projects. This scoping ensures that Project B's knowledge base remains separate and only contains files explicitly added to it."}, {"letter": "B", "text": "The handbook automatically appears in every new project the coordinator creates from that point on", "correct": false, "explanation": "There is no automatic propagation of files across projects. Each project's knowledge base is independent; files must be uploaded or shared explicitly if needed. This behavior prevents unintended data exposure and maintains clean separation between unrelated workstreams."}, {"letter": "C", "text": "The handbook is deleted from Project A's knowledge base the moment a second project gets created", "correct": false, "explanation": "Creating a new project has no effect on existing projects or their knowledge bases. Files remain intact in Project A and are not automatically removed or transferred. Project B starts with an empty knowledge base until files are added to it directly."}, {"letter": "D", "text": "The handbook becomes available only within projects sharing the same billing plan tier as Project A", "correct": false, "explanation": "Knowledge base scoping is determined by the project container, not by billing plan tier. While different plans may offer features like higher file limits or longer context windows, they do not alter the fundamental rule that each project has its own isolated set of reference files."}], "correct": "A", "select": 1}, {"id": "a5-029", "domain": 5, "task_id": "5.4", "objective": "Inform, maintain, and update Claude configurations, knowledge sources, and instructions", "situation": "A regional team keeps the manual for a discontinued product in a project's knowledge base, on the grounds that someone may still ask about it one day.", "question": "What is the greatest concern?", "options": [{"letter": "A", "text": "Older files are read first, so any newer document uploaded after them is skipped in every answer.", "correct": false, "explanation": "Upload order does not determine which document is used to answer a question."}, {"letter": "B", "text": "The manual counts twice against the project's limit, halving the space left for current documents.", "correct": false, "explanation": "Files do not count twice, and space is not the concern being raised."}, {"letter": "C", "text": "Claude may answer current questions from material about a product the company no longer sells.", "correct": true, "explanation": "Retired material that is still loaded competes with current sources and produces wrong current answers."}, {"letter": "D", "text": "Discontinued manuals cannot be stored in a project, so the whole knowledge base becomes unavailable.", "correct": false, "explanation": "There is no restriction on the age or status of an uploaded document."}], "correct": "C", "select": 1}, {"id": "a5-030", "domain": 5, "task_id": "5.3", "objective": "Create effective system-level instructions", "situation": "A researcher pastes a forty-page methodology handbook into a project's instructions field instead of uploading it as a document.", "question": "What is the greatest concern with doing this?", "options": [{"letter": "A", "text": "Uploading is the only supported route, so pasting the handbook prevents the project from being saved at all.", "correct": false, "explanation": "Text can be typed into the instructions field; the problem is judgment, not a hard block."}, {"letter": "B", "text": "Pasted text in the instructions is treated as a one-off prompt and is discarded when that chat ends.", "correct": false, "explanation": "Project instructions persist across chats; they are not discarded at the end of a conversation."}, {"letter": "C", "text": "The instructions become a document rather than direction, so the real house rules are buried and applied unevenly.", "correct": true, "explanation": "Instructions are for concise standing direction; burying a handful of rules inside long reference text weakens them."}, {"letter": "D", "text": "Project instructions can be read only by their author, so teammates in the project receive a different set of rules.", "correct": false, "explanation": "Instructions belong to the project, and members with access see the same ones."}], "correct": "C", "select": 1}, {"id": "a5-031", "domain": 5, "task_id": "5.4", "objective": "Inform, maintain, and update Claude configurations, knowledge sources, and instructions", "situation": "A project's knowledge base has grown to dozens of overlapping documents added by different people over time, and answer quality has fallen.", "question": "What is the best way to improve it?", "options": [{"letter": "A", "text": "Add an instruction telling Claude to prefer whichever of the documents happens to be the longest.", "correct": false, "explanation": "Length is no indicator of currency or correctness."}, {"letter": "B", "text": "Consolidate the material into a smaller set of current documents and remove the duplicates.", "correct": true, "explanation": "Curating down to current, non-overlapping sources removes the conflicts that degrade answers."}, {"letter": "C", "text": "Add more documents covering the same topics so that the majority of the sources agree with one another.", "correct": false, "explanation": "Adding more overlapping material makes the conflict worse, not better."}, {"letter": "D", "text": "Split the documents across several new projects so that each workspace holds only a handful of files.", "correct": false, "explanation": "Scattering stale duplicates across workspaces spreads the problem instead of resolving it."}], "correct": "B", "select": 1}, {"id": "a5-032", "domain": 5, "task_id": "5.2", "objective": "Manage uploaded knowledge and connectors (e.g., Google Drive, Gmail)", "situation": "A user tries to upload an XLSX spreadsheet into a chat and the upload is rejected.", "question": "What must be enabled first for the upload to succeed?", "options": [{"letter": "A", "text": "The Google Drive connector must be connected before spreadsheets can be uploaded.", "correct": false, "explanation": "The Drive connector is a separate feature and is not required to upload a local XLSX file in chat."}, {"letter": "B", "text": "Code execution must be turned on in account settings before XLSX files upload.", "correct": true, "explanation": "XLSX uploads require code execution to be enabled in account settings; without it, spreadsheets cannot be processed."}, {"letter": "C", "text": "The account must be upgraded from Free to Pro before spreadsheets can upload.", "correct": false, "explanation": "Plan tier is not the blocker here; the specific requirement is enabling code execution."}, {"letter": "D", "text": "The project knowledge base feature must be turned on for the account first.", "correct": false, "explanation": "Project knowledge being enabled is unrelated; XLSX support is gated by the code execution setting specifically."}], "correct": "B", "select": 1}, {"id": "a5-033", "domain": 5, "task_id": "5.1", "objective": "Configure Claude Projects with instructions and knowledge sources", "situation": "A team is writing the instructions for a shared project that will be used for all of its customer communications.", "question": "Which direction belongs in the project instructions rather than in an individual chat?", "options": [{"letter": "A", "text": "Summarise the complaint from the Ashworth account that came in during yesterday afternoon.", "correct": false, "explanation": "This is a one-off task tied to a single conversation, not standing direction for the project."}, {"letter": "B", "text": "Write at a reading level suited to customers, and never quote internal ticket references.", "correct": true, "explanation": "Project instructions carry durable, project-wide direction such as tone, audience and house rules, so they apply consistently to every chat in the project."}, {"letter": "C", "text": "Draft this week's outage notice, keeping it to one paragraph for the status page.", "correct": false, "explanation": "A specific deliverable for this week belongs in the chat where it is being produced."}, {"letter": "D", "text": "Compare the two apology drafts below and say which reads better for this customer.", "correct": false, "explanation": "A request about material pasted into one conversation cannot serve as project-wide direction."}], "correct": "B", "select": 1}, {"id": "a5-034", "domain": 5, "task_id": "5.4", "objective": "Inform, maintain, and update Claude configurations, knowledge sources, and instructions", "situation": "A benefits administrator keeps last year's policy handbook in a project's knowledge base. Claude keeps quoting the old deductible, and the administrator has now rewritten the question five different ways.", "question": "What is the greatest concern here?", "options": [{"letter": "A", "text": "The answers are confidently wrong from an out-of-date source, and no amount of rephrasing will correct them.", "correct": true, "explanation": "When the uploaded source is stale, the output is wrong at the source and reprompting cannot repair it."}, {"letter": "B", "text": "The project has run out of usage allowance, which causes older documents to be summarised inaccurately.", "correct": false, "explanation": "Usage allowance affects how much you can do, not the accuracy of a quoted figure."}, {"letter": "C", "text": "The question is too short, so Claude fills in the gaps with figures drawn from general knowledge of benefits.", "correct": false, "explanation": "The wrong figure matches the superseded handbook, which points at the source rather than the prompt."}, {"letter": "D", "text": "The handbook is too long for the project, so only the first pages are read and the remainder is dropped.", "correct": false, "explanation": "Length is not the issue; the figure being quoted is present in the document and simply out of date."}], "correct": "A", "select": 1}, {"id": "a5-035", "domain": 5, "task_id": "5.4", "objective": "Inform, maintain, and update Claude configurations, knowledge sources, and instructions", "situation": "A project's instructions state that refunds are approved within five working days, while an uploaded policy document says ten.", "question": "What should the project owner do first?", "options": [{"letter": "A", "text": "Establish which of the two figures is current, then correct whichever of the sources is wrong.", "correct": true, "explanation": "You cannot fix a contradiction without first determining which statement reflects the real process."}, {"letter": "B", "text": "Add a line telling Claude to average the two figures whenever it answers questions about refunds.", "correct": false, "explanation": "Averaging two conflicting policy statements produces a figure that is in neither source."}, {"letter": "C", "text": "Ask Claude which of the two figures it prefers, and then keep whichever source it says it uses.", "correct": false, "explanation": "Claude has no way to know which figure reflects the organisation's current process."}, {"letter": "D", "text": "Delete the project instructions and rely on the uploaded policy document from now on instead.", "correct": false, "explanation": "This assumes the document is right, which is exactly what has not been established."}], "correct": "A", "select": 1}, {"id": "a5-036", "domain": 5, "task_id": "5.3", "objective": "Create effective system-level instructions", "situation": "A project's instructions say \"write for our executive committee\". The same project is also used to produce customer-facing FAQ answers.", "question": "What is the greatest risk here?", "options": [{"letter": "A", "text": "The project can no longer hold uploaded documents, because an audience rule takes the place of the knowledge base.", "correct": false, "explanation": "Audience direction and uploaded knowledge are independent and coexist without issue."}, {"letter": "B", "text": "Customer material inherits an executive register that is wrong for its readers, and the mismatch is easy to miss.", "correct": true, "explanation": "One durable audience rule silently mis-shapes every piece of work aimed at a different audience."}, {"letter": "C", "text": "Every FAQ answer is saved as a separate project, since a second audience needs its own workspace to exist.", "correct": false, "explanation": "Nothing creates projects automatically; the workspace is unchanged by what is asked in it."}, {"letter": "D", "text": "The instruction blocks any request that is not addressed to the executive committee, so the FAQ chats fail.", "correct": false, "explanation": "An audience instruction shapes style; it does not reject requests."}], "correct": "B", "select": 1}, {"id": "a5-037", "domain": 5, "task_id": "5.3", "objective": "Create effective system-level instructions", "situation": "A writer wants one rule to apply everywhere she uses Claude, and a second rule to apply only inside one client Project. Which two placements are correct?", "question": "Select two.", "options": [{"letter": "A", "text": "Put both of the rules into the client Project's instructions, since those apply account-wide.", "correct": false, "explanation": "Project instructions apply inside that Project only, so putting the global rule there would not apply it everywhere."}, {"letter": "B", "text": "Put the client-specific rule into that Project's instructions so it applies only inside it.", "correct": true, "explanation": "Project instructions are scoped to their Project, which is exactly the boundary the second rule needs."}, {"letter": "C", "text": "Put the everywhere rule into her account-wide personalization settings so it always applies.", "correct": true, "explanation": "Account-wide personalization is the right home for direction that should apply across all of her Claude use."}, {"letter": "D", "text": "Put both rules at the top of each new chat, since durable settings are for administrators.", "correct": false, "explanation": "Durable instruction settings are available to ordinary users, and re-typing rules per chat is what they replace."}, {"letter": "E", "text": "Put the client rule into her account-wide settings too, so that it is never once missed.", "correct": false, "explanation": "Account-wide settings apply to all her work, so a client-specific rule placed there would leak into unrelated chats."}], "correct": ["B", "C"], "select": 2}, {"id": "a5-038", "domain": 5, "task_id": "5.4", "objective": "Inform, maintain, and update Claude configurations, knowledge sources, and instructions", "situation": "A team uploads the new travel policy to a project but leaves the previous year's version in the knowledge base as well.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The older document is archived automatically, so the team loses the record of what the old rules were.", "correct": false, "explanation": "Nothing removes or archives an uploaded file on its own; that is the maintainer's job."}, {"letter": "B", "text": "The project immediately rejects the newer file, because two documents cannot cover the same subject area at once.", "correct": false, "explanation": "There is no restriction preventing two documents on the same topic."}, {"letter": "C", "text": "Both versions are available, so Claude may cite the superseded rules without anyone noticing the conflict.", "correct": true, "explanation": "Two live versions of the same policy make wrong answers look exactly as authoritative as right ones."}, {"letter": "D", "text": "The project switches to retrieval mode, which stops Claude from quoting either policy document directly.", "correct": false, "explanation": "Retrieval relates to the size of project knowledge and still allows documents to be quoted."}], "correct": "C", "select": 1}, {"id": "a5-039", "domain": 5, "task_id": "5.3", "objective": "Create effective system-level instructions", "situation": "A support team and a marketing team share one project. Support wants terse troubleshooting answers; marketing wants warm promotional copy. The project owner wants consistent quality for both.", "question": "What should the owner do first?", "options": [{"letter": "A", "text": "Keep one project and ask each person to restate their preferred tone at the start of every chat.", "correct": false, "explanation": "This abandons durable direction entirely and reintroduces the inconsistency being complained about."}, {"letter": "B", "text": "Split the work into two projects, each with instructions written for that team's audience and purpose.", "correct": true, "explanation": "Durable direction only works when one workspace serves one audience and purpose."}, {"letter": "C", "text": "Ask Claude to work out from each question whether the person is in support or marketing, and adapt.", "correct": false, "explanation": "Guessing the requester's team from the wording is unreliable and leaves no explicit standard."}, {"letter": "D", "text": "Write one set of instructions listing both styles and let each person say which one they want in each chat.", "correct": false, "explanation": "This is a conflicting rule set that pushes the choice back into every individual conversation."}], "correct": "B", "select": 1}, {"id": "a5-040", "domain": 5, "task_id": "5.2", "objective": "Manage uploaded knowledge and connectors (e.g., Google Drive, Gmail)", "situation": "A team member is worried that connecting Google Drive will let colleagues read files they are not allowed to open.", "question": "What is accurate?", "options": [{"letter": "A", "text": "A connector requires every file to be made public in Drive before Claude is able to read any of them.", "correct": false, "explanation": "No change to a file's sharing settings is required to use a connector."}, {"letter": "B", "text": "A connector gives everyone in the workspace the same view, so the file permissions no longer apply inside Claude.", "correct": false, "explanation": "Connectors do not flatten or override the permissions of the connected service."}, {"letter": "C", "text": "A connector follows each person's existing permissions, so files they cannot open in Drive stay out of reach.", "correct": true, "explanation": "If someone cannot reach a file in the source system, the connector cannot reach it for them either."}, {"letter": "D", "text": "A connector copies the files into a shared store, so anyone in the organisation can search them later.", "correct": false, "explanation": "Connectors read from the source system rather than duplicating files into a shared pool."}], "correct": "C", "select": 1}, {"id": "a5-041", "domain": 5, "task_id": "5.4", "objective": "Inform, maintain, and update Claude configurations, knowledge sources, and instructions", "situation": "A campaign project has finished, but the team may want to look back at its material later in the year.", "question": "What is the most appropriate action?", "options": [{"letter": "A", "text": "Archive the project so it stays available for reference without sitting among the active workspaces.", "correct": true, "explanation": "Archiving preserves the material and its context while keeping the active list clean."}, {"letter": "B", "text": "Share the project with the whole organisation so that anyone can look back at the campaign material.", "correct": false, "explanation": "Broad sharing is not a substitute for tidying up, and it widens access without a reason."}, {"letter": "C", "text": "Strip out the instructions but keep the documents, so the project cannot produce any new answers.", "correct": false, "explanation": "Removing the direction leaves a half-configured workspace that will mislead whoever opens it."}, {"letter": "D", "text": "Delete the project, since finished work should not be left in the account under any circumstances.", "correct": false, "explanation": "Deletion destroys material the team has said it may still need."}], "correct": "A", "select": 1}, {"id": "a5-042", "domain": 5, "task_id": "5.3", "objective": "Create effective system-level instructions", "situation": "A quality lead wants Claude to stop producing figures that do not appear anywhere in the reports uploaded to a project.", "question": "Which addition to the project instructions best supports this?", "options": [{"letter": "A", "text": "Tell Claude to add a disclaimer to every answer noting that figures may not have been verified.", "correct": false, "explanation": "A blanket disclaimer shifts responsibility to the reader instead of preventing the problem."}, {"letter": "B", "text": "Tell Claude to answer only from the uploaded reports and to say plainly when a figure is not in them.", "correct": true, "explanation": "This scopes the source and gives an explicit alternative to inventing a number."}, {"letter": "C", "text": "Tell Claude to always give a numeric answer, choosing the closest available figure when none matches.", "correct": false, "explanation": "This actively instructs the behaviour the quality lead is trying to prevent."}, {"letter": "D", "text": "Tell Claude that accuracy is extremely important and that mistakes are unacceptable in this project.", "correct": false, "explanation": "Emphasis without a rule gives no actionable behaviour to follow when a figure is missing."}], "correct": "B", "select": 1}, {"id": "a5-043", "domain": 5, "task_id": "5.3", "objective": "Create effective system-level instructions", "situation": "A project's instructions tell Claude to \"respond as our compliance officer and give final approval on vendor contracts\".", "question": "What is the greatest risk in writing the instructions this way?", "options": [{"letter": "A", "text": "It assigns Claude a decision role a person must own, so drafts may be treated as approvals nobody reviewed.", "correct": true, "explanation": "Framing durable direction as decision authority invites colleagues to treat output as a sign-off."}, {"letter": "B", "text": "It uses a job title that Claude cannot recognise, so the instruction is skipped and the default tone applies.", "correct": false, "explanation": "Role framing is understood; the danger is the authority it implies, not a failure to parse the title."}, {"letter": "C", "text": "It prevents the project knowledge base from being used, since role rules take priority over uploaded files.", "correct": false, "explanation": "Instructions and uploaded knowledge work together; one does not switch the other off."}, {"letter": "D", "text": "It makes the project visible to everyone in the organisation, because role rules imply shared ownership.", "correct": false, "explanation": "Instruction wording has no effect on who the project is shared with."}], "correct": "A", "select": 1}, {"id": "a5-044", "domain": 5, "task_id": "5.4", "objective": "Inform, maintain, and update Claude configurations, knowledge sources, and instructions", "situation": "", "question": "After replacing three documents in a project's knowledge base, what is the most reliable way to confirm the refresh has worked?", "options": [{"letter": "A", "text": "Compare the total size of the knowledge base before and after the three files were replaced.", "correct": false, "explanation": "Size tells you nothing about which figures the answers now come from."}, {"letter": "B", "text": "Confirm that the file names of the three new documents appear correctly in the project's knowledge list.", "correct": false, "explanation": "A file being listed does not prove the old content has stopped shaping answers."}, {"letter": "C", "text": "Ask questions whose correct answers changed, and check the responses against the new documents.", "correct": true, "explanation": "Only a question with a changed answer distinguishes the new source from the old one."}, {"letter": "D", "text": "Ask Claude to confirm that it has read all of the newly uploaded documents in the project.", "correct": false, "explanation": "A self-report about reading files is not evidence that answers now reflect them."}], "correct": "C", "select": 1}, {"id": "a5-045", "domain": 5, "task_id": "5.4", "objective": "Inform, maintain, and update Claude configurations, knowledge sources, and instructions", "situation": "A Project that a department has relied on for a year has never had its uploaded knowledge reviewed. Which two maintenance actions are appropriate?", "question": "Select two.", "options": [{"letter": "A", "text": "Schedule a regular review so outdated documents are found before they cause wrong answers.", "correct": true, "explanation": "Stale knowledge is a documented failure mode: Claude Academy's Projects guidance says outdated documents can lead to outdated responses, so a scheduled review is what catches that before it produces wrong answers."}, {"letter": "B", "text": "Update or remove documents that no longer reflect current facts before they mislead Claude.", "correct": true, "explanation": "Claude Academy's Projects guidance says to review and update project knowledge periodically because outdated documents can lead to outdated responses - updating or removing the stale ones is what turns a review into an actual fix rather than just a diagnosis."}, {"letter": "C", "text": "Leave every old document in place, since more background material always improves the answers.", "correct": false, "explanation": "Claude Academy's Projects guidance is explicit that outdated documents can lead to outdated responses, so accumulating background material without pruning it works against accuracy, not for it."}, {"letter": "D", "text": "Replace the Project with a fresh one as soon as any single answer turns out to be wrong.", "correct": false, "explanation": "Rebuilding the whole workspace after one bad answer discards good context instead of curating the source at fault."}, {"letter": "E", "text": "Ask Claude at the start of each chat to work out which uploaded documents are outdated.", "correct": false, "explanation": "Claude cannot tell which uploaded document has been superseded; that judgment depends on knowledge it does not hold."}], "correct": ["A", "B"], "select": 2}, {"id": "a5-046", "domain": 5, "task_id": "5.3", "objective": "Create effective system-level instructions", "situation": "A finance team's project instructions include a standing rule that every figure must be reported in euros, plus several directions that were only ever meant for individual past deliverables, and Claude's answers in the project have become inconsistent. Which two revisions would make those instructions work better as standing direction?", "question": "Select two.", "options": [{"letter": "A", "text": "Rewrite each rule as a longer, more formal paragraph so none of them is overlooked.", "correct": false, "explanation": "Length and formality do not resolve which rules still apply. A long list of stale directions becomes a longer list of stale directions, and dense instructions are harder for the team to review."}, {"letter": "B", "text": "Add a line at the top of the list telling Claude to follow every rule carefully.", "correct": false, "explanation": "A generic exhortation to follow the rules adds no information about which directions still apply or which take priority. Emphasis is not specification, and the outdated entries keep pulling answers in their own direction."}, {"letter": "C", "text": "Move the list out of the project and paste it into the first message of each chat.", "correct": false, "explanation": "This gives up the main advantage of project instructions, which is that durable direction applies automatically to every chat. Pasted instructions have to be maintained by hand and will vary between chats and between people."}, {"letter": "D", "text": "Delete directions that were only ever meant for one particular past deliverable.", "correct": true, "explanation": "Standing instructions should hold for every chat in the project. One-off directions kept past their deliverable pull answers toward a job nobody is doing any more, which is a direct source of the inconsistency."}, {"letter": "E", "text": "Give the reason behind the euro-only rule so Claude can apply it to new cases.", "correct": true, "explanation": "and for an unrelated reason: it improves a rule that stays rather than removing one that should go. Stating why figures must be in euros lets Claude extend the intent sensibly to situations the rule never named."}], "correct": ["D", "E"], "select": 2}, {"id": "a5-047", "domain": 5, "task_id": "5.1", "objective": "Configure Claude Projects with instructions and knowledge sources", "situation": "A marketer discusses a campaign brief with Claude in Chat A inside a project, but Claude does not recall those details when the marketer opens a new Chat B in the same project.", "question": "What most likely explains this?", "options": [{"letter": "A", "text": "The brief was only discussed in Chat A and was never added to the project's shared knowledge base", "correct": true, "explanation": "Context is not shared across chats in a project unless it has been added to the project knowledge base; discussion in one chat alone stays local to that chat."}, {"letter": "B", "text": "The project has already reached its maximum allowed number of chats for the current billing cycle", "correct": false, "explanation": "There is no documented per-billing-cycle cap on the number of chats within a project that would explain this behavior."}, {"letter": "C", "text": "Project instructions must be re-saved before any new chat can access prior chat conversations", "correct": false, "explanation": "Re-saving project instructions has no bearing on whether one chat's conversation content is visible to another chat."}, {"letter": "D", "text": "The marketer's current account plan does not permit more than one chat inside each project", "correct": false, "explanation": "Plans do not restrict a project to a single chat; multiple chats within one project are expected and supported."}], "correct": "A", "select": 1}, {"id": "a5-048", "domain": 5, "task_id": "5.4", "objective": "Inform, maintain, and update Claude configurations, knowledge sources, and instructions", "situation": "A consultant set account-wide instructions two years ago describing a role they no longer hold, and answers now carry the wrong framing.", "question": "What is the most appropriate action?", "options": [{"letter": "A", "text": "Mention the correct role at the start of each conversation and accept the extra step as normal.", "correct": false, "explanation": "Repeating a correction forever is the manual overhead durable configuration exists to remove."}, {"letter": "B", "text": "Update the account-wide instructions so they describe the consultant's current role and the work they do now.", "correct": true, "explanation": "Personal configuration needs the same periodic refresh as any other durable direction."}, {"letter": "C", "text": "Create a brand-new account so that the outdated preferences are all left behind with the previous configuration.", "correct": false, "explanation": "Abandoning an account to escape a stale setting loses all the history and other configuration with it."}, {"letter": "D", "text": "Add a line to every project's instructions telling Claude to disregard the account-wide instructions.", "correct": false, "explanation": "This is a workaround that leaves the outdated configuration in place for all work outside projects."}], "correct": "B", "select": 1}, {"id": "a5-049", "domain": 5, "task_id": "5.2", "objective": "Manage uploaded knowledge and connectors (e.g., Google Drive, Gmail)", "situation": "Using the Calendar connector, a user asks Claude to set up a team sync that repeats every Monday.", "question": "What can Claude do here?", "options": [{"letter": "A", "text": "Claude can draft the details, but an admin must go configure the recurrence separately.", "correct": false, "explanation": "No admin intervention is needed; the connector lets Claude set up recurrence directly."}, {"letter": "B", "text": "Claude can create the event directly, since the connector supports recurring events.", "correct": true, "explanation": "The Calendar connector explicitly supports establishing recurring meetings, so a weekly Monday sync can be created directly."}, {"letter": "C", "text": "Claude can view existing recurring events but cannot create new ones itself.", "correct": false, "explanation": "The connector supports both viewing and creating events, including recurring ones."}, {"letter": "D", "text": "Claude can only create a single event with no way to set up recurrence.", "correct": false, "explanation": "Recurrence configuration is a documented capability, not a missing feature requiring a workaround."}], "correct": "B", "select": 1}, {"id": "a5-050", "domain": 5, "task_id": "5.4", "objective": "Inform, maintain, and update Claude configurations, knowledge sources, and instructions", "situation": "A colleague reports that Claude gave them an incorrect answer inside a shared project.", "question": "What should the project owner do first?", "options": [{"letter": "A", "text": "Remove every document from the knowledge base and rebuild the project from current files.", "correct": false, "explanation": "Rebuilding everything is disproportionate before the cause is even known."}, {"letter": "B", "text": "Rewrite the project instructions to add a rule that all answers must be double-checked by hand.", "correct": false, "explanation": "A blanket caution does not diagnose or fix whatever produced the wrong answer."}, {"letter": "C", "text": "Advise the colleague to try the same question again in a new chat within the same project.", "correct": false, "explanation": "The same project knowledge is read again, so a retry rarely changes a source-driven error."}, {"letter": "D", "text": "Ask which question was asked, then check the uploaded documents for the answer Claude gave.", "correct": true, "explanation": "Tracing the answer back to a source separates a stale document from a prompting or model problem."}], "correct": "D", "select": 1}, {"id": "a5-051", "domain": 5, "task_id": "5.4", "objective": "Inform, maintain, and update Claude configurations, knowledge sources, and instructions", "situation": "A team repeatedly corrects Claude in chat because it keeps using the wrong report template.", "question": "What is the most durable fix?", "options": [{"letter": "A", "text": "Keep correcting Claude in chat, since each correction improves the answers for the whole team.", "correct": false, "explanation": "A correction inside one conversation does not change the behaviour for anyone else."}, {"letter": "B", "text": "Put the template requirement into the project instructions so every chat starts with the right rule.", "correct": true, "explanation": "Making a repeated correction part of the configuration stops the team re-solving it every time."}, {"letter": "C", "text": "Ask each person to paste the template into the first message of every conversation they start.", "correct": false, "explanation": "This keeps the manual step and simply moves it to the top of each chat."}, {"letter": "D", "text": "Create a new project each month so that the wrong template does not carry over from any earlier work.", "correct": false, "explanation": "New workspaces without the rule will still default to the wrong template."}], "correct": "B", "select": 1}, {"id": "a5-052", "domain": 5, "task_id": "5.3", "objective": "Create effective system-level instructions", "situation": "A project's instructions say \"always write in plain language for non-specialists\" and, further down, \"use the full technical terminology our engineers expect\". Team members report that outputs vary widely.", "question": "What is the greatest concern with these instructions as written?", "options": [{"letter": "A", "text": "Claude will merge the two rules into one style that always sits exactly halfway between the two audiences.", "correct": false, "explanation": "A predictable midpoint is not what happens; the observed symptom is inconsistency, not a stable compromise."}, {"letter": "B", "text": "Instructions longer than a few lines are ignored, so only the very first rule in the list will ever be applied.", "correct": false, "explanation": "There is no such rule that only the opening line of project instructions is read."}, {"letter": "C", "text": "The two rules conflict, so Claude has no reliable basis for choosing a register and results vary between chats.", "correct": true, "explanation": "Contradictory house rules are the direct cause: neither rule can be satisfied without breaking the other."}, {"letter": "D", "text": "Conflicting rules cause the project knowledge base to be reindexed, which slows down every answer given.", "correct": false, "explanation": "Instruction wording does not trigger any reprocessing of uploaded documents."}], "correct": "C", "select": 1}, {"id": "a5-053", "domain": 5, "task_id": "5.3", "objective": "Create effective system-level instructions", "situation": "A legal operations manager writes project instructions containing forty numbered rules intended to cover every request the team might ever make.", "question": "What is the most likely consequence?", "options": [{"letter": "A", "text": "Each numbered rule creates a separate chat history, so the project's conversations are split across rules.", "correct": false, "explanation": "Instructions do not create or partition chat histories in any way."}, {"letter": "B", "text": "Numbered lists are not accepted in the instructions field, so only the first several rules are stored at all.", "correct": false, "explanation": "There is no restriction on numbered lists in project instructions."}, {"letter": "C", "text": "Long instructions force the project into retrieval mode, which removes uploaded documents from answers.", "correct": false, "explanation": "Retrieval behaviour relates to the size of project knowledge, and it does not remove documents from answers."}, {"letter": "D", "text": "Rules that rarely apply crowd out the few that always matter, so the key ones are followed less reliably.", "correct": true, "explanation": "Long, undifferentiated rule lists dilute the direction that actually matters for most requests."}], "correct": "D", "select": 1}, {"id": "a5-054", "domain": 5, "task_id": "5.1", "objective": "Configure Claude Projects with instructions and knowledge sources", "situation": "A team lead grants \"Can edit\" access on a shared project to everyone who might ever open it, rather than giving most people \"Can view\".", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The project's chat history becomes visible to the whole organisation regardless of the access levels granted.", "correct": false, "explanation": "Granting edit access to named people does not publish the project organisation-wide."}, {"letter": "B", "text": "Only one editor can open the project at a time, so members are blocked whenever a colleague is in it.", "correct": false, "explanation": "There is no single-editor lock on a shared project."}, {"letter": "C", "text": "Editors are billed separately for the project, so the department's costs rise with each person added.", "correct": false, "explanation": "Access levels within a project are not billed individually."}, {"letter": "D", "text": "Anyone can change the instructions and the knowledge, so the shared setup can drift without anyone noticing.", "correct": true, "explanation": "Broad edit rights let the agreed direction and sources be altered silently by any of the members."}], "correct": "D", "select": 1}, {"id": "a5-055", "domain": 5, "task_id": "5.4", "objective": "Inform, maintain, and update Claude configurations, knowledge sources, and instructions", "situation": "A Project's knowledge base has grown until it approaches the limit of what fits in the context window. Which two statements are accurate?", "question": "Select two.", "options": [{"letter": "A", "text": "The Project stays in retrieval mode permanently, even if its knowledge is reduced later.", "correct": false, "explanation": "The Project reverts to standard processing if the knowledge later drops back below the threshold."}, {"letter": "B", "text": "The user has to switch retrieval on in the settings once the knowledge passes that threshold.", "correct": false, "explanation": "There is no setting to turn on; treating it as manual is the most common misunderstanding of this behaviour."}, {"letter": "C", "text": "Claude switches over to retrieval automatically, with no configuration needed from the user.", "correct": true, "explanation": "Retrieval mode for Projects turns on by itself as the knowledge approaches the limit; there is nothing to configure."}, {"letter": "D", "text": "Retrieval disables web search and the other features for every chat inside that Project.", "correct": false, "explanation": "Retrieval works alongside other Claude features, including web search, extended thinking and Research."}, {"letter": "E", "text": "Retrieval lets the Project work with roughly ten times more content than it held before.", "correct": true, "explanation": "In retrieval mode a Project can hold up to about ten times more content than standard processing allows."}], "correct": ["C", "E"], "select": 2}, {"id": "a5-056", "domain": 5, "task_id": "5.3", "objective": "Create effective system-level instructions", "situation": "A grant writer wants every draft in a project written from the point of view of a programme officer assessing applications, so that weak arguments surface early.", "question": "What is the best way to set this up?", "options": [{"letter": "A", "text": "Upload sample funded applications to project knowledge and expect Claude to adopt a reviewer stance.", "correct": false, "explanation": "Examples of successful applications do not communicate an assessing point of view."}, {"letter": "B", "text": "Name the perspective in the project title, since the title is what Claude reads first in each new chat.", "correct": false, "explanation": "A project title is a label; it is not where behavioural direction is configured."}, {"letter": "C", "text": "Mention the reviewer's stance in whichever chats feel relevant, so the writer keeps control of when it applies.", "correct": false, "explanation": "This makes an intended default optional and easy to forget on the drafts that need it most."}, {"letter": "D", "text": "State the perspective and its purpose in the project instructions so every chat adopts that stance.", "correct": true, "explanation": "A durable stance that should hold for all work in the workspace belongs in project instructions."}], "correct": "D", "select": 1}, {"id": "a5-057", "domain": 5, "task_id": "5.1", "objective": "Configure Claude Projects with instructions and knowledge sources", "situation": "A consultant sets project instructions telling Claude to always respond as a financial advisor, then separately uploads quarterly earnings reports to the same project's knowledge base.", "question": "When the consultant opens a brand-new chat in that project, what should they expect?", "options": [{"letter": "A", "text": "Claude requires the consultant to manually re-enable the instructions and knowledge base for each chat", "correct": false, "explanation": "Neither instructions nor knowledge base content require manual re-enabling per chat; both are automatically active project-wide."}, {"letter": "B", "text": "Claude applies the advisor persona but cannot access the reports until they are re-uploaded in that chat", "correct": false, "explanation": "Knowledge base files are available project-wide and do not need to be re-uploaded inside each individual chat."}, {"letter": "C", "text": "Claude ignores the persona instructions in new chats but can still access the earnings reports fully", "correct": false, "explanation": "Project instructions apply to all chats in the project, not selectively to some and not others."}, {"letter": "D", "text": "Claude applies the advisor persona and references the earnings reports, since both persist project-wide", "correct": true, "explanation": "Both project instructions and knowledge base uploads are designed to persist and apply automatically across every chat within the same project, including new ones."}], "correct": "D", "select": 1}, {"id": "a5-058", "domain": 5, "task_id": "5.2", "objective": "Manage uploaded knowledge and connectors (e.g., Google Drive, Gmail)", "situation": "A policy team keeps a handbook in a Drive folder that changes weekly, and also has one annual report that never changes. Which two set-ups fit best?", "question": "Select two.(Select 2)", "options": [{"letter": "A", "text": "Connect the Drive folder, so the weekly handbook changes are picked up from the source.", "correct": true, "explanation": "A connector reaches the live source, which is what a document that changes every week needs."}, {"letter": "B", "text": "Connect the annual report and delete it from Drive, so that only one copy ever exists.", "correct": false, "explanation": "Deleting the original to leave a single copy destroys the source of truth for no benefit."}, {"letter": "C", "text": "Upload a copy of the handbook just once and then rely on that copy for the rest of this year.", "correct": false, "explanation": "A one-time copy of a weekly-changing handbook goes stale fast and produces confidently outdated answers."}, {"letter": "D", "text": "Paste the whole handbook into every new chat, so the newest version is always present.", "correct": false, "explanation": "Re-pasting a whole handbook per chat is the manual work Project knowledge and connectors exist to remove."}, {"letter": "E", "text": "Upload the fixed annual report into Project knowledge, since its content will not change.", "correct": true, "explanation": "A stable document is well suited to a one-time upload into Project knowledge, where it stays available to every chat."}], "correct": ["A", "E"], "select": 2}, {"id": "a5-059", "domain": 5, "task_id": "5.2", "objective": "Manage uploaded knowledge and connectors (e.g., Google Drive, Gmail)", "situation": "A policy team keeps a handbook in a Drive folder that changes weekly, and also has one annual report that never changes. Which two set-ups fit best?", "question": "Select two.", "options": [{"letter": "A", "text": "Connect the annual report and delete it from Drive, so that only one copy ever exists.", "correct": false, "explanation": "Deleting the original to leave a single copy destroys the source of truth for no benefit."}, {"letter": "B", "text": "Upload the fixed annual report into Project knowledge, since its content will not change.", "correct": true, "explanation": "A stable document is well suited to a one-time upload into Project knowledge, where it stays available to every chat."}, {"letter": "C", "text": "Paste the whole handbook into every new chat, so the newest version is always present.", "correct": false, "explanation": "Re-pasting a whole handbook per chat is the manual work Project knowledge and connectors exist to remove."}, {"letter": "D", "text": "Connect the Drive folder, so the weekly handbook changes are picked up from the source.", "correct": true, "explanation": "A connector reaches the live source, which is what a document that changes every week needs."}, {"letter": "E", "text": "Upload a copy of the handbook just once and then rely on that copy for the rest of this year.", "correct": false, "explanation": "A one-time copy of a weekly-changing handbook goes stale fast and produces confidently outdated answers."}], "correct": ["B", "D"], "select": 2}, {"id": "a5-060", "domain": 5, "task_id": "5.1", "objective": "Configure Claude Projects with instructions and knowledge sources", "situation": "An operations lead creates a project on an Enterprise plan and shares it with the whole organisation. To save time she uploads a complete team folder into the project knowledge, and that folder includes a spreadsheet of individual salary bands.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "Anyone who opens the shared project can draw the salary details out through ordinary questions.", "correct": true, "explanation": "Project knowledge is background for every chat in the project, so sharing the project shares everything uploaded to it with everyone who can open it."}, {"letter": "B", "text": "The extra documents will slow down every answer in the project until the folder has been trimmed back.", "correct": false, "explanation": "A larger knowledge base is a performance and tidiness concern, not the serious exposure here."}, {"letter": "C", "text": "The salary spreadsheet can no longer be removed once other members have opened the project.", "correct": false, "explanation": "Documents can be removed from project knowledge at any time; the exposure is the real problem."}, {"letter": "D", "text": "Claude will decline every request in the project while a salary document remains uploaded.", "correct": false, "explanation": "Claude does not refuse to work in a project because a sensitive file has been uploaded to it."}], "correct": "A", "select": 1}, {"id": "a5-061", "domain": 5, "task_id": "5.1", "objective": "Configure Claude Projects with instructions and knowledge sources", "situation": "A developer wants Claude to have awareness of a project's coding conventions when answering questions in a dedicated 'Backend Refactor' project.", "question": "Which action correctly places that context so it applies across all chats in the project?", "options": [{"letter": "A", "text": "Star the coding convention files in a personal favorites list to make them available across project chats.", "correct": false, "explanation": "A personal favorites list is a user-specific bookmarking feature and does not inject content into Claude's context. Project knowledge base is the proper mechanism for providing shared documents across chats."}, {"letter": "B", "text": "Ask Claude to memorize the coding conventions in a single chat to persist across project chats.", "correct": false, "explanation": "Memorization within a single chat does not persist to other chats. To share context across all project chats, you must upload the material to the project's knowledge base."}, {"letter": "C", "text": "Rename each new chat to include the word 'conventions' so Claude infers the context across chats.", "correct": false, "explanation": "Chat names are only labels and do not convey document content or give Claude awareness of conventions. Only the project knowledge base can provide persistent context across chats."}, {"letter": "D", "text": "Upload the convention docs to the project's knowledge base with the '+' button on the main page.", "correct": true, "explanation": "Uploading documents to the project's knowledge base with the '+' button injects them as context across all chats. This ensures Claude always has access to the coding conventions when answering in the 'Backend Refactor' project."}], "correct": "D", "select": 1}, {"id": "a5-062", "domain": 5, "task_id": "5.2", "objective": "Manage uploaded knowledge and connectors (e.g., Google Drive, Gmail)", "situation": "At an Enterprise organization, no one has enabled connectors yet, and an employee tries to connect Google Drive.", "question": "What must happen before the employee can authenticate?", "options": [{"letter": "A", "text": "The employee must switch to a personal, non-organizational Claude account first.", "correct": false, "explanation": "Switching account types is irrelevant; the organization admin simply needs to enable the connector."}, {"letter": "B", "text": "An Owner or Primary Owner must first enable the connector organization-wide.", "correct": true, "explanation": "On Team and Enterprise plans, an Owner or Primary Owner must enable connectors org-wide before members can authenticate their own accounts."}, {"letter": "C", "text": "The employee must install a separate Workspace marketplace add-on before this.", "correct": false, "explanation": "No separate marketplace add-on installation is a prerequisite for the Drive connector."}, {"letter": "D", "text": "The employee must contact Anthropic support to whitelist their Google account first.", "correct": false, "explanation": "There is no support-ticket whitelisting process; the control point is an org-level toggle set by an Owner or Primary Owner."}], "correct": "B", "select": 1}, {"id": "a5-063", "domain": 5, "task_id": "5.3", "objective": "Create effective system-level instructions", "situation": "A marketing lead is writing the Project instructions for a Project her team will use every day. Which two choices make those instructions effective?", "question": "Select two.", "options": [{"letter": "A", "text": "Listing an answer to each specific question the team happened to ask Claude during the last month.", "correct": false, "explanation": "A list of past answers is a transcript, not direction, and it dates quickly as the team's questions change."}, {"letter": "B", "text": "Stating the audience and the tone that every response inside that Project should be written for.", "correct": true, "explanation": "Audience and tone are exactly the durable direction Project instructions exist to hold, rather than being restated per chat."}, {"letter": "C", "text": "Keeping the instructions vague, so that they never conflict with any individual chat request.", "correct": false, "explanation": "Vague instructions give no usable direction, so responses drift back to whatever each individual chat happens to say."}, {"letter": "D", "text": "Repeating one rule several times in capital letters so it carries more weight than the others.", "correct": false, "explanation": "Emphasis and repetition are not what makes an instruction effective; clarity and stated intent are."}, {"letter": "E", "text": "Giving the reason behind a house rule, so that Claude can apply it to unforeseen situations.", "correct": true, "explanation": "Giving the reason behind a rule lets Claude generalise it to cases the instructions never anticipated."}], "correct": ["B", "E"], "select": 2}, {"id": "a5-064", "domain": 5, "task_id": "5.1", "objective": "Configure Claude Projects with instructions and knowledge sources", "situation": "A user wants to know if they must upgrade from a free account just to try out the basic project feature with instructions and a small knowledge base.", "question": "What is true about project availability on free accounts?", "options": [{"letter": "A", "text": "Projects are available to free accounts too, though free users are capped at a maximum of five projects", "correct": true, "explanation": "Projects are available to all users including free accounts, though free users can create a maximum of five projects."}, {"letter": "B", "text": "Free accounts can create projects but are never permitted to set project instructions at all", "correct": false, "explanation": "Setting project instructions is part of the core project feature available to free accounts, not blocked entirely."}, {"letter": "C", "text": "Projects are an exclusive Team and Enterprise feature and are unavailable on any individual account", "correct": false, "explanation": "Projects are not restricted to Team and Enterprise; free and Pro individual accounts can use them too."}, {"letter": "D", "text": "Free accounts can create projects but the knowledge base feature is entirely disabled for them", "correct": false, "explanation": "The knowledge base is part of the core project experience available on free accounts, subject to the five-project cap rather than being disabled."}], "correct": "A", "select": 1}, {"id": "a5-065", "domain": 5, "task_id": "5.1", "objective": "Configure Claude Projects with instructions and knowledge sources", "situation": "A consultant works in a free Claude account and keeps a separate project for each client. She already has five projects and has just signed a sixth client whose material she wants to keep separate too.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Ask an administrator to raise the project allowance on her account before she starts the work.", "correct": false, "explanation": "A personal free account has no administrator who can lift the cap; the plan defines the limit."}, {"letter": "B", "text": "Review her five projects and retire or repurpose one, since a free account is capped at five projects.", "correct": true, "explanation": "Projects are available to everyone, but a free account is limited to five of them, so she has to free one up before a sixth client workspace can exist."}, {"letter": "C", "text": "Create the sixth project as normal, since the cap on the number of projects applies only to shared ones.", "correct": false, "explanation": "The five-project cap on free accounts is not limited to shared projects; it covers her projects."}, {"letter": "D", "text": "Keep this client in ordinary chats instead, since projects are only offered on the paid plans.", "correct": false, "explanation": "Projects are available to all users, including free accounts, so this misreads the constraint."}], "correct": "B", "select": 1}, {"id": "a5-066", "domain": 5, "task_id": "5.3", "objective": "Create effective system-level instructions", "situation": "A new joiner inherits a shared project whose outputs do not match what their manager describes as the team standard.", "question": "What should the new joiner do first?", "options": [{"letter": "A", "text": "Ask Claude in a chat to explain which instructions it received and treat that answer as the record.", "correct": false, "explanation": "The configuration itself is the record; a conversational account of it is not a reliable substitute."}, {"letter": "B", "text": "Start a fresh project of their own so the existing team workspace is left exactly as it was found.", "correct": false, "explanation": "This leaves the team's shared problem untouched and fragments where the work is done."}, {"letter": "C", "text": "Read the project instructions already in place to understand the current direction before changing anything.", "correct": true, "explanation": "You cannot tell what is missing or wrong until you know what direction the project currently carries."}, {"letter": "D", "text": "Clear the existing instructions and rewrite them from scratch to match the manager's description of good output.", "correct": false, "explanation": "Discarding direction the team relies on, before understanding it, risks losing rules that were deliberate."}], "correct": "C", "select": 1}, {"id": "a5-067", "domain": 5, "task_id": "5.1", "objective": "Configure Claude Projects with instructions and knowledge sources", "situation": "A legal analyst wants Claude to reference a set of contract templates in every chat within a 'Contracts Review' project without re-uploading them each time.", "question": "What is the correct way to make these templates persistently available across all chats in that project?", "options": [{"letter": "A", "text": "Upload the contract templates once to the project's knowledge base so all chats can reference them", "correct": true, "explanation": "Files added to the project knowledge base are available as context across all chats within the project, avoiding repeated uploads."}, {"letter": "B", "text": "Paste the full text of every contract template directly into the project instructions field", "correct": false, "explanation": "Project instructions are meant for behavioral guidance like tone and role, not for storing full reference documents; the knowledge base is the appropriate location for that."}, {"letter": "C", "text": "Email the templates to Claude support so an agent can manually link them to the analyst's account", "correct": false, "explanation": "There is no support workflow for linking documents to an account this way; uploads go directly into the project's knowledge base."}, {"letter": "D", "text": "Attach the templates only to the first chat and mark that chat as the project's default conversation", "correct": false, "explanation": "Attachments made within an individual chat are not automatically shared with other chats; only knowledge base uploads persist project-wide."}], "correct": "A", "select": 1}, {"id": "a5-068", "domain": 5, "task_id": "5.2", "objective": "Manage uploaded knowledge and connectors (e.g., Google Drive, Gmail)", "situation": "Using the Gmail connector, a user asks Claude to summarize the contents of a PDF attached to an email.", "question": "What limitation should the user expect?", "options": [{"letter": "A", "text": "Claude can read the email metadata and body but cannot open the attachment's content.", "correct": true, "explanation": "The Gmail connector exposes metadata and body content but does not process attachment content, so the PDF's contents remain inaccessible."}, {"letter": "B", "text": "Claude cannot access the email at all until the attachment is first removed.", "correct": false, "explanation": "Claude can still read the surrounding email even with an attachment present; only the attachment content is off-limits."}, {"letter": "C", "text": "Claude can read the attachment content, but only when it stays under 10MB in size.", "correct": false, "explanation": "There is no documented size threshold that unlocks attachment reading; content stays inaccessible regardless of size."}, {"letter": "D", "text": "Claude can read the attachment content, but only when the attachment is a PDF file type.", "correct": false, "explanation": "Attachment content is inaccessible for any file type, not just non-PDF formats."}], "correct": "A", "select": 1}, {"id": "a5-069", "domain": 5, "task_id": "5.1", "objective": "Configure Claude Projects with instructions and knowledge sources", "situation": "A freelancer on a Pro plan has built a project with detailed instructions and reference documents, and now wants two collaborators to work inside the same project.", "question": "What should they expect?", "options": [{"letter": "A", "text": "Sharing becomes available once the project contains uploaded documents rather than instructions alone.", "correct": false, "explanation": "What a project contains has no bearing on whether it can be shared."}, {"letter": "B", "text": "Sharing works on every plan, so the collaborators can simply be invited with view or edit access.", "correct": false, "explanation": "Personal plans do not offer project sharing with other people."}, {"letter": "C", "text": "Sharing works only after the project has been archived, which makes it visible to other accounts.", "correct": false, "explanation": "Archiving tidies a project away; it does not publish it to anyone."}, {"letter": "D", "text": "Project sharing is a Team and Enterprise capability, so on a Pro plan the project stays personal.", "correct": true, "explanation": "Sharing projects with named colleagues is available on Team and Enterprise plans only."}], "correct": "D", "select": 1}, {"id": "a5-070", "domain": 5, "task_id": "5.2", "objective": "Manage uploaded knowledge and connectors (e.g., Google Drive, Gmail)", "situation": "A Team account with fifteen members has just had the Gmail connector enabled by the Primary Owner.", "question": "How does each member gain access through the connector?", "options": [{"letter": "A", "text": "One member authenticates once, granting Gmail access to the whole team.", "correct": false, "explanation": "Authentication is per-user; one member's OAuth grant does not extend to teammates' inboxes."}, {"letter": "B", "text": "The Primary Owner authenticates for everyone through a single shared service account.", "correct": false, "explanation": "There is no shared service account model; each user connects their own Google account individually."}, {"letter": "C", "text": "Members share one OAuth token that IT distributes to the whole team.", "correct": false, "explanation": "Shared or distributed tokens are not how the connector works; per-user authentication keeps data separate."}, {"letter": "D", "text": "Each member individually authenticates via OAuth with their own Google credentials.", "correct": true, "explanation": "Even after org-level enablement, each user authenticates through standard OAuth with their own credentials, keeping access per-person."}], "correct": "D", "select": 1}, {"id": "a5-071", "domain": 5, "task_id": "5.4", "objective": "Inform, maintain, and update Claude configurations, knowledge sources, and instructions", "situation": "A project shared with a Team plan workspace has just had its escalation matrix replaced with a new version.", "question": "What else should the project owner do?", "options": [{"letter": "A", "text": "Tell the members who use the project what changed, so they know earlier answers may no longer hold.", "correct": true, "explanation": "Maintaining shared configuration includes informing the people whose past work relied on the old source."}, {"letter": "B", "text": "Ask each member to create their own copy of the project so everyone controls their own knowledge base.", "correct": false, "explanation": "Copies multiply the maintenance problem and guarantee that some members keep using the old matrix."}, {"letter": "C", "text": "Delete the project's chat history, because past conversations would otherwise override the new material.", "correct": false, "explanation": "Old conversations do not override project knowledge, and deleting them destroys a useful record."}, {"letter": "D", "text": "Leave a note in the project title, since members are prompted to read the title before each new chat.", "correct": false, "explanation": "A title is not a change log, and members are not prompted to read it before chatting."}], "correct": "A", "select": 1}, {"id": "a5-072", "domain": 5, "task_id": "5.4", "objective": "Inform, maintain, and update Claude configurations, knowledge sources, and instructions", "situation": "A pricing analyst notices Claude quoting a discount tier that was retired at the last price review.", "question": "What should the analyst do first?", "options": [{"letter": "A", "text": "Check the project's uploaded documents to see whether the retired price list is still in the knowledge base.", "correct": true, "explanation": "Confirming whether a superseded source is still loaded is the cheapest way to explain the wrong answer."}, {"letter": "B", "text": "Rewrite the project instructions to say that all of the pricing answers must be treated as provisional estimates.", "correct": false, "explanation": "This hides the symptom behind a disclaimer instead of removing the outdated source."}, {"letter": "C", "text": "Ask Claude to search the web for the current price list and to use that instead of the uploaded material.", "correct": false, "explanation": "Public search is not a reliable source for internal pricing and leaves the stale file in place."}, {"letter": "D", "text": "Start a new conversation, since long chats gradually drift away from the documents they started with.", "correct": false, "explanation": "A fresh chat still reads the same project knowledge, so the retired figures come back."}], "correct": "A", "select": 1}, {"id": "a5-073", "domain": 5, "task_id": "5.2", "objective": "Manage uploaded knowledge and connectors (e.g., Google Drive, Gmail)", "situation": "A user pastes a Google Doc URL into a chat with the Drive connector enabled and asks about the document's contents.", "question": "What does Claude do?", "options": [{"letter": "A", "text": "Claude downloads the document as a PDF and stores a copy in project knowledge.", "correct": false, "explanation": "Accessing a document via the connector does not create a permanent copy in project knowledge."}, {"letter": "B", "text": "Claude ignores the URL and asks the user to paste the document's text instead.", "correct": false, "explanation": "The connector is designed to retrieve the document automatically, so manual pasting is not required."}, {"letter": "C", "text": "Claude requires the document to be shared publicly before it can read the link.", "correct": false, "explanation": "Public sharing is not required, since the connector authenticates to the user's own private Google account."}, {"letter": "D", "text": "Claude accesses and processes the referenced document once the message is sent.", "correct": true, "explanation": "With the Drive connector, a pasted URL is accessed and processed when the message is sent, informing the response."}], "correct": "D", "select": 1}, {"id": "a5-074", "domain": 5, "task_id": "5.2", "objective": "Manage uploaded knowledge and connectors (e.g., Google Drive, Gmail)", "situation": "An Enterprise organisation wants its members to use the Google Drive connector, but with read access only.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Members authenticate individually; the owner enables the connector; the team confirms the limits; the owner restricts it to read actions.", "correct": false, "explanation": "Members cannot authenticate before the owner has enabled the connector."}, {"letter": "B", "text": "The owner enables the connector; the owner restricts it to read actions; members authenticate individually; the team confirms the limits.", "correct": true, "explanation": "Enablement and action restrictions are owner-level and come before individuals connect their own accounts."}, {"letter": "C", "text": "The owner restricts it to read actions; the team confirms the limits; members authenticate individually; the owner enables the connector.", "correct": false, "explanation": "Restrictions are configured for a connector that has not been enabled yet."}, {"letter": "D", "text": "The team confirms the limits; members authenticate individually; the owner restricts it to read actions; the owner enables the connector.", "correct": false, "explanation": "The limits are confirmed first and the connector is enabled last, which reverses the setup."}], "correct": "B", "select": 1}, {"id": "a5-075", "domain": 5, "task_id": "5.4", "objective": "Inform, maintain, and update Claude configurations, knowledge sources, and instructions", "situation": "A finance approvals process has changed: purchases above a threshold now need two signatures instead of one. A project's instructions still describe the old single-signature rule.", "question": "What should the project owner do first?", "options": [{"letter": "A", "text": "Upload the email announcing the change so that Claude can work out the new rule from that announcement itself.", "correct": false, "explanation": "This leaves the contradicting instruction in place and asks Claude to reconcile the conflict."}, {"letter": "B", "text": "Archive the project and rebuild it later once the finance team has finished documenting the change.", "correct": false, "explanation": "Removing a working project is disproportionate when a single rule needs rewriting."}, {"letter": "C", "text": "Update the project instructions to describe the new two-signature rule before the project is used again.", "correct": true, "explanation": "Durable direction that contradicts the current process must be corrected at the source, and quickly."}, {"letter": "D", "text": "Tell colleagues to mention the new rule in their prompts whenever they ask about purchase approvals.", "correct": false, "explanation": "This relies on every person remembering, while the wrong rule stays configured for everyone else."}], "correct": "C", "select": 1}, {"id": "a5-076", "domain": 5, "task_id": "5.3", "objective": "Create effective system-level instructions", "situation": "A finance team's project instructions state: \"never give a figure without naming the report it came from\".", "question": "Which addition would most improve this instruction?", "options": [{"letter": "A", "text": "Adding the rule a second time at the end of the instructions, so it is less likely to be overlooked.", "correct": false, "explanation": "Duplication does not clarify the rule and lengthens the instructions for no benefit."}, {"letter": "B", "text": "Adding why the rule exists, so Claude can apply the same care to cases the rule did not anticipate.", "correct": true, "explanation": "Stating the reason lets the intent generalise to situations the literal wording never covered."}, {"letter": "C", "text": "Adding an instruction to repeat the rule at the top of every answer, so readers can see it was followed.", "correct": false, "explanation": "Restating the rule in the output adds noise without improving whether it is actually followed."}, {"letter": "D", "text": "Adding the names of the three people who approved the rule, so Claude knows the rule is authoritative.", "correct": false, "explanation": "Approval provenance changes nothing about how the rule is applied to an answer."}], "correct": "B", "select": 1}, {"id": "a5-077", "domain": 5, "task_id": "5.3", "objective": "Create effective system-level instructions", "situation": "An operations manager has two things to add to a project: the team's two-page approved abbreviation list, and the rule that answers must never use abbreviations outside that list.", "question": "What is the best placement?", "options": [{"letter": "A", "text": "Put both the reference list and the rule into the project instructions so that the two are always read together.", "correct": false, "explanation": "Two pages of reference text in the instructions field buries the rule that is meant to stand out."}, {"letter": "B", "text": "Upload both the reference list and the rule as documents in the project knowledge base, for consistency.", "correct": false, "explanation": "A behaviour rule stored as background material is far weaker than one stated as project direction."}, {"letter": "C", "text": "Keep the reference list on the team wiki and paste the rule at the start of each chat that needs it.", "correct": false, "explanation": "This puts the reference out of reach and turns a standing rule back into manual repetition."}, {"letter": "D", "text": "Upload the reference to project knowledge and put the rule about unapproved abbreviations in the instructions.", "correct": true, "explanation": "Reference material belongs in knowledge; the standing behaviour rule belongs in the instructions."}], "correct": "D", "select": 1}, {"id": "a5-078", "domain": 5, "task_id": "5.4", "objective": "Inform, maintain, and update Claude configurations, knowledge sources, and instructions", "situation": "An internal FAQ drafted with help from a project is about to be published to all staff.", "question": "Which maintenance step matters most before it goes out?", "options": [{"letter": "A", "text": "Confirm the FAQ is long enough to cover every question the team has ever received from staff.", "correct": false, "explanation": "Coverage is not the risk here; answers drawn from outdated sources are."}, {"letter": "B", "text": "Check each answer against the current source documents and confirm none rely on superseded material.", "correct": true, "explanation": "Publication is the point at which a stale source turns into a broadly circulated wrong answer."}, {"letter": "C", "text": "Ask Claude to review its own FAQ and to confirm that all of the answers it has written are accurate.", "correct": false, "explanation": "A self-review cannot detect that the underlying uploaded source was out of date."}, {"letter": "D", "text": "Republish the FAQ every month so that errors in the current version are eventually replaced.", "correct": false, "explanation": "Scheduled republication does not stop the first version going out with wrong answers."}], "correct": "B", "select": 1}, {"id": "a5-079", "domain": 5, "task_id": "5.1", "objective": "Configure Claude Projects with instructions and knowledge sources", "situation": "A team lead is setting up a new project that will be used for recurring quarterly business reviews.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Write the instructions; run a typical request; upload the reference documents; create the project and name its purpose.", "correct": false, "explanation": "The instructions and the test both precede the project they belong to."}, {"letter": "B", "text": "Create the project and name its purpose; write the instructions; upload the reference documents; run a typical request.", "correct": true, "explanation": "Establish the workspace and its purpose, give it durable direction and background, then check it on real work."}, {"letter": "C", "text": "Upload the reference documents; run a typical request; create the project and name its purpose; write the instructions.", "correct": false, "explanation": "Documents cannot be uploaded into a project that has not been created yet."}, {"letter": "D", "text": "Run a typical request; upload the reference documents; write the instructions; create the project and name its purpose.", "correct": false, "explanation": "The setup is tested before any part of it exists."}], "correct": "B", "select": 1}, {"id": "a5-080", "domain": 5, "task_id": "5.3", "objective": "Create effective system-level instructions", "situation": "A marketing associate's project instructions say \"keep it short\", yet the drafts Claude produces are still far longer than the team wants.", "question": "What should the associate do first?", "options": [{"letter": "A", "text": "Replace the vague rule with a specific limit, such as a stated word count and a section structure.", "correct": true, "explanation": "Vague direction is the root cause; the fix is to make the requirement concrete enough to follow."}, {"letter": "B", "text": "Switch to a more capable model, since shorter writing needs stronger reasoning about relevance.", "correct": false, "explanation": "A stronger model does not repair an under-specified instruction."}, {"letter": "C", "text": "Add capital letters and the word \"must\" to the existing rule so the instruction carries more weight.", "correct": false, "explanation": "Emphasis does not add the missing information about how short is short enough."}, {"letter": "D", "text": "Move the rule into the project knowledge base, where longer background material is normally stored.", "correct": false, "explanation": "Moving a behaviour rule into reference material makes it less likely to shape every answer, not more."}], "correct": "A", "select": 1}, {"id": "a5-081", "domain": 5, "task_id": "5.1", "objective": "Configure Claude Projects with instructions and knowledge sources", "situation": "A product manager is deciding whether onboarding guides, style docs, and roadmap notes for three unrelated products should live in one shared project or in three separate projects.", "question": "Which approach best matches how Claude Projects are intended to be used?", "options": [{"letter": "A", "text": "Store the documents outside any project and paste the relevant ones into each new chat manually", "correct": false, "explanation": "Manually pasting documents into every chat ignores the knowledge base feature that projects provide for persistent context."}, {"letter": "B", "text": "Create one project per document type instead of per product, mixing all three products' content", "correct": false, "explanation": "Organizing by document type while mixing three unrelated products still blends unrelated context, the opposite of the intended focused workspace."}, {"letter": "C", "text": "Combine everything into a single project so all three products share one unified context in every chat", "correct": false, "explanation": "Merging unrelated products into one project would mix contexts, working against the purpose of keeping conversations focused."}, {"letter": "D", "text": "Create a separate project per product so each self-contained workspace keeps a focused knowledge base", "correct": true, "explanation": "Projects are designed as self-contained workspaces with their own chat histories and knowledge bases, so organizing by unrelated product keeps context focused and avoids cross-contamination."}], "correct": "D", "select": 1}, {"id": "a5-082", "domain": 5, "task_id": "5.2", "objective": "Manage uploaded knowledge and connectors (e.g., Google Drive, Gmail)", "situation": "A user edits a Google Doc directly in Drive after Claude referenced it earlier in a conversation, then asks a follow-up question about the same document.", "question": "What should the user expect?", "options": [{"letter": "A", "text": "The user must re-upload the document as a new file for changes to appear.", "correct": false, "explanation": "Re-uploading is unnecessary because the connector already reads the live document from Drive."}, {"letter": "B", "text": "Claude uses a cached snapshot from the first access and misses the new edits entirely.", "correct": false, "explanation": "There is no static cached snapshot behavior; the connector reflects the current Drive version instead."}, {"letter": "C", "text": "Claude's view of the document reflects the latest Drive version, since it auto-syncs.", "correct": true, "explanation": "Documents accessed through the Drive connector sync automatically with their Drive versions, so the follow-up reflects the latest edits."}, {"letter": "D", "text": "Claude only sees updates when the document is re-shared through a new link.", "correct": false, "explanation": "Generating a new share link is not required, since syncing happens automatically for the existing reference."}], "correct": "C", "select": 1}, {"id": "a5-083", "domain": 5, "task_id": "5.1", "objective": "Configure Claude Projects with instructions and knowledge sources", "situation": "", "question": "Within a single Claude Project, what is the functional difference between project instructions and the project knowledge base?", "options": [{"letter": "A", "text": "Project instructions define behavioral guidance like tone or role, while the knowledge base supplies reference files", "correct": true, "explanation": "Project instructions establish behavioral guidelines such as tone or role applied to all chats, while the knowledge base holds uploaded documents, text, and code used as background context."}, {"letter": "B", "text": "Project instructions and the knowledge base are simply two names for the same underlying feature", "correct": false, "explanation": "The two are distinct features serving different purposes: behavioral guidance versus reference content."}, {"letter": "C", "text": "Project instructions apply to a single chat at a time, while the knowledge base applies to every project a user owns", "correct": false, "explanation": "Project instructions apply to every chat within the project, not just one, and the knowledge base is scoped to that project, not to all of a user's projects."}, {"letter": "D", "text": "Project instructions store uploaded files, while the knowledge base only accepts text typed into a settings field", "correct": false, "explanation": "This reverses the roles; file uploads belong to the knowledge base, not the instructions field."}], "correct": "A", "select": 1}, {"id": "a5-084", "domain": 5, "task_id": "5.3", "objective": "Create effective system-level instructions", "situation": "A team lead on a Team plan wants a shared project whose instructions the whole team relies on.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Save them as project instructions; share the project; agree the rules with the team; collect colleague feedback.", "correct": false, "explanation": "Writing the rules before the team has agreed them means the shared setup encodes one person's view."}, {"letter": "B", "text": "Agree the rules with the team; save them as project instructions; share the project; collect colleague feedback.", "correct": true, "explanation": "Agree, encode, distribute, then learn from real use by the people who now depend on it."}, {"letter": "C", "text": "Collect colleague feedback; agree the rules with the team; share the project; save them as project instructions.", "correct": false, "explanation": "Feedback is gathered before anything exists, and the instructions are saved only after sharing."}, {"letter": "D", "text": "Share the project; collect colleague feedback; agree the rules with the team; save them as project instructions.", "correct": false, "explanation": "Colleagues are asked to review a project that has no agreed direction in it yet."}], "correct": "B", "select": 1}, {"id": "a5-085", "domain": 5, "task_id": "5.2", "objective": "Manage uploaded knowledge and connectors (e.g., Google Drive, Gmail)", "situation": "A user wants Claude to find a spreadsheet they opened yesterday in Google Drive but does not remember its exact URL.", "question": "What can the Drive connector do?", "options": [{"letter": "A", "text": "Only retrieve a file when the exact URL is provided by the user first.", "correct": false, "explanation": "A URL is one option, not a requirement; recent-document search is also supported."}, {"letter": "B", "text": "List every file in the Drive account without offering any search tool.", "correct": false, "explanation": "Search capability is part of the connector, not just a flat unsearchable file list."}, {"letter": "C", "text": "Only access files explicitly shared with a separate Claude service account first.", "correct": false, "explanation": "There is no separate service account requirement; the connector uses the user's own authenticated Drive access."}, {"letter": "D", "text": "Search through recently accessed documents in Drive to help find the file.", "correct": true, "explanation": "The Drive connector lets users search recently accessed documents, or paste a URL, so the file can be found without the exact link."}], "correct": "D", "select": 1}, {"id": "a5-086", "domain": 5, "task_id": "5.3", "objective": "Create effective system-level instructions", "situation": "A communications team wants every press-release draft produced in a shared project to come out in the company's house voice, without anyone having to explain the voice again in each chat.", "question": "What should the team do first?", "options": [{"letter": "A", "text": "Ask Claude in each new chat to match the house voice, and then correct the tone in the follow-up messages each time.", "correct": false, "explanation": "This is the repeated one-off prompting the team is trying to escape, and it produces different results per person."}, {"letter": "B", "text": "Write down the house voice rules and audience the team already applies, then put them in the project instructions.", "correct": true, "explanation": "The durable direction has to exist as agreed wording before it can be configured; defining what good looks like comes first."}, {"letter": "C", "text": "Upload every press release published last year and let Claude infer the house voice on its own from them.", "correct": false, "explanation": "Samples are useful background, but they do not state the rules, and inference from mixed past work is unreliable."}, {"letter": "D", "text": "Create a separate project for each writer so that every person can keep their own preferred tone settings.", "correct": false, "explanation": "This guarantees inconsistency, which is the opposite of the shared house voice the team asked for."}], "correct": "B", "select": 1}, {"id": "a5-087", "domain": 5, "task_id": "5.2", "objective": "Manage uploaded knowledge and connectors (e.g., Google Drive, Gmail)", "situation": "A user is hesitant to connect Gmail to Claude over privacy concerns.", "question": "Which statement accurately describes how Anthropic handles connector data?", "options": [{"letter": "A", "text": "Anthropic does not train models on connector data, and stores it securely.", "correct": true, "explanation": "Anthropic does not train on Gmail, Drive, or Calendar connector data, and retrieved data is stored on Anthropic servers with standard security."}, {"letter": "B", "text": "Anthropic trains its models on connector data by default, unless the user opts out.", "correct": false, "explanation": "Model training on connector data does not occur, so no opt-out is needed in the first place."}, {"letter": "C", "text": "Connector data is processed only in memory and never stored on any server.", "correct": false, "explanation": "Retrieved connector data is stored on Anthropic's servers rather than existing only transiently."}, {"letter": "D", "text": "Connector data is shared with the user's Workspace admin for compliance review.", "correct": false, "explanation": "There is no described mechanism routing a user's connector data to their Workspace admin."}], "correct": "A", "select": 1}, {"id": "a5-088", "domain": 5, "task_id": "5.4", "objective": "Inform, maintain, and update Claude configurations, knowledge sources, and instructions", "situation": "A project owner is adding a newly issued reference document to a project's knowledge base.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Confirm it is the approved current version; upload it to project knowledge; record a review date; ask a test question.", "correct": true, "explanation": "Verify currency before it becomes a source, then schedule the next review and confirm it answers correctly."}, {"letter": "B", "text": "Upload it to project knowledge; confirm it is the approved current version; ask a test question; record a review date.", "correct": false, "explanation": "The document becomes a live source before anyone has checked that it is the current version."}, {"letter": "C", "text": "Ask a test question; upload it to project knowledge; record a review date; confirm it is the approved current version.", "correct": false, "explanation": "The test happens before the document exists in the project, and currency is checked last."}, {"letter": "D", "text": "Record a review date; ask a test question; confirm it is the approved current version; upload it to project knowledge.", "correct": false, "explanation": "A review date is set for a document that has not been added, and the upload comes last."}], "correct": "A", "select": 1}, {"id": "a5-089", "domain": 5, "task_id": "5.3", "objective": "Create effective system-level instructions", "situation": "A consultant wants Claude to use British spelling and metric units in every conversation, no matter which piece of work it belongs to.", "question": "Where does that direction best belong?", "options": [{"letter": "A", "text": "In the account-wide instructions, since the preference applies to all conversations rather than one workspace.", "correct": true, "explanation": "Account-wide instructions are the place for direction that should hold across every conversation."}, {"letter": "B", "text": "In the instructions of each project, repeated so that every workspace carries the same spelling rule.", "correct": false, "explanation": "This works only inside projects that were remembered, and every new project has to be updated by hand."}, {"letter": "C", "text": "In the opening message of each chat, restated whenever the consultant begins any new piece of work.", "correct": false, "explanation": "Repeating it forever is exactly the manual overhead durable configuration is meant to remove."}, {"letter": "D", "text": "In a document uploaded to each project's knowledge base, so that Claude reads the rule as general background material.", "correct": false, "explanation": "Knowledge is reference material for a project, not a standing behaviour rule, and it misses plain chats."}], "correct": "A", "select": 1}, {"id": "a5-090", "domain": 5, "task_id": "5.1", "objective": "Configure Claude Projects with instructions and knowledge sources", "situation": "A research team keeps adding reports to a project's knowledge base and wants to know what happens as that knowledge grows past what fits in a single conversation.", "question": "What should they expect?", "options": [{"letter": "A", "text": "Claude moves automatically to a retrieval mode that can draw on up to ten times more project content.", "correct": true, "explanation": "As project knowledge approaches the limit, Claude switches to retrieval automatically and can then hold up to ten times more content, reverting if the knowledge later drops back."}, {"letter": "B", "text": "Someone with edit rights has to switch retrieval on in the project's settings before it can take effect.", "correct": false, "explanation": "Retrieval for projects is automatic; there is no setting for a member to turn it on or off."}, {"letter": "C", "text": "Claude keeps only the documents added most recently and quietly disregards the older uploads.", "correct": false, "explanation": "Older documents are not silently dropped; retrieval is what lets the larger knowledge base work."}, {"letter": "D", "text": "The knowledge base has to be split across two projects by hand once it passes the limit.", "correct": false, "explanation": "Splitting by hand is unnecessary, because the switch to retrieval happens without any action."}], "correct": "A", "select": 1}, {"id": "a5-091", "domain": 5, "task_id": "5.1", "objective": "Configure Claude Projects with instructions and knowledge sources", "situation": "", "question": "What is the primary purpose of the knowledge base within a Claude Project?", "options": [{"letter": "A", "text": "It gives Claude uploaded documents, text, and code as shared context for every chat inside the project", "correct": true, "explanation": "The project knowledge base holds documents, text, code, and other files that Claude uses as background context across all chats within that project."}, {"letter": "B", "text": "It automatically publishes any uploaded files to other users outside of the project boundary", "correct": false, "explanation": "Files uploaded to a project's knowledge base are not automatically published outside the project; access depends on project sharing settings."}, {"letter": "C", "text": "It converts each uploaded file into a separate standalone chatbot scoped to that project alone", "correct": false, "explanation": "The knowledge base supplies context to Claude within the project; it does not spin up a separate chatbot."}, {"letter": "D", "text": "It stores a running log of every prompt a user has typed across all of their separate projects", "correct": false, "explanation": "The knowledge base is not a prompt history log; it is a repository of uploaded reference material scoped to one project."}], "correct": "A", "select": 1}, {"id": "a6-001", "domain": 6, "task_id": "6.1", "objective": "Identify appropriate and inappropriate use cases", "situation": "A university biosecurity lab asks Claude to help draft detailed synthesis instructions for a novel pathogen with pandemic potential, describing the request as an internal red-team exercise.", "question": "Is this an appropriate use of Claude?", "options": [{"letter": "A", "text": "No, but only because the lab has not signed a separate use-case exception agreement with Anthropic sales", "correct": false, "explanation": "No sales exception can authorize weapons-development content; this category is a universal prohibition, not a negotiable licensing term."}, {"letter": "B", "text": "Yes, because the request comes from an accredited academic institution conducting legitimate research", "correct": false, "explanation": "Institutional accreditation does not exempt a request from the Usage Policy's ban on assisting with the development of biological, chemical, or nuclear weapons."}, {"letter": "C", "text": "No, because the Usage Policy bans assisting biological weapons development regardless of affiliation or intent", "correct": true, "explanation": "The Usage Policy prohibits helping design or synthesize biological, chemical, radiological, or nuclear weapons, and this restriction applies regardless of who is asking or why."}, {"letter": "D", "text": "Yes, because internal red-team security exercises are exempt from the Usage Policy's weapons restrictions", "correct": false, "explanation": "Framing a request as a red-team exercise does not create an exemption; the prohibition applies to the substance of the request, not its stated purpose."}], "correct": "C", "select": 1}, {"id": "a6-002", "domain": 6, "task_id": "6.4", "objective": "Understand the ethical implications of AI usage", "situation": "A university instructor plans to use Claude to help generate and score final exam questions that determine students' course grades, without any faculty review of the generated content or scores.", "question": "Which requirement of the Usage Policy is this instructor at risk of violating?", "options": [{"letter": "A", "text": "The requirement that a qualified professional review AI-assisted academic content before it affects grades", "correct": true, "explanation": "Academic testing is listed as a high-risk category, so a qualified professional must review AI-generated or AI-scored content before it is used to affect real student outcomes."}, {"letter": "B", "text": "The requirement that exam questions be limited to multiple-choice format whenever AI is involved", "correct": false, "explanation": "There is no format restriction limiting AI-assisted exams to multiple-choice questions."}, {"letter": "C", "text": "The requirement that all exam content be generated exclusively by Claude Opus rather than Claude Sonnet models", "correct": false, "explanation": "The Usage Policy does not mandate a specific Claude model for academic use cases."}, {"letter": "D", "text": "The requirement that students be offered an opt-out from any form of AI-assisted grading", "correct": false, "explanation": "The policy's safeguard is human review before dissemination, not a student opt-out mechanism."}], "correct": "A", "select": 1}, {"id": "a6-003", "domain": 6, "task_id": "6.4", "objective": "Understand the ethical implications of AI usage", "situation": "An HR team wants to use Claude to help draft candidate rejection emails after final hiring decisions have already been made by human recruiters.", "question": "Why does this still count as a high-risk use case under Anthropic's Usage Policy?", "options": [{"letter": "A", "text": "Because rejection emails are legally binding contracts that require notarization before being sent", "correct": false, "explanation": "Rejection emails are not legally binding contracts, and notarization is not part of the Usage Policy's requirements."}, {"letter": "B", "text": "Because the emails are stored for more than 30 days, which automatically triggers high-risk status", "correct": false, "explanation": "Storage duration is not the factor that determines high-risk status; the use case category itself is what triggers the requirement."}, {"letter": "C", "text": "Because Claude cannot generate any employment-related text under the Usage Policy in any circumstance", "correct": false, "explanation": "Claude can be used for employment-related text; the policy requires safeguards like review and disclosure, not an outright ban."}, {"letter": "D", "text": "Because employment is a designated high-risk category, so review and disclosure still apply here", "correct": true, "explanation": "Employment is one of the categories the Usage Policy designates as high-risk due to public welfare implications, so human review and disclosure requirements apply even when a human made the underlying decision."}], "correct": "D", "select": 1}, {"id": "a6-004", "domain": 6, "task_id": "6.1", "objective": "Identify appropriate and inappropriate use cases", "situation": "A charity's volunteering team is deciding whether to hand its volunteer reference-checking process over to Claude.", "question": "In which order should these steps be taken?", "options": [{"letter": "A", "text": "Work out which parts of the process a mistake would make hard to undo; run the resulting arrangement on a small batch of references; keep those parts with staff and give Claude only the rest; widen it to the full caseload once that batch holds up.", "correct": false, "explanation": "The risky steps are identified but the pilot begins before they have been kept with staff, so the small batch is run through exactly the arrangement the assessment warned against."}, {"letter": "B", "text": "Keep those parts with staff and give Claude only the rest; work out which parts of the process a mistake would make hard to undo; run the resulting arrangement on a small batch of references; widen it to the full caseload once that batch holds up.", "correct": false, "explanation": "The split between staff and Claude is made before anyone has judged where an error would be irreversible, so the assessment that should have produced the split arrives after it."}, {"letter": "C", "text": "Work out which parts of the process a mistake would make hard to undo; keep those parts with staff and give Claude only the rest; run the resulting arrangement on a small batch of references; widen it to the full caseload once that batch holds up.", "correct": true, "explanation": "Reversibility is what decides which steps may be delegated, so it is assessed before anything is handed over; the split follows from that assessment; the arrangement is then proved on a small batch where a mistake is contained; and only evidence from that batch justifies the full caseload."}, {"letter": "D", "text": "Run the resulting arrangement on a small batch of references; work out which parts of the process a mistake would make hard to undo; keep those parts with staff and give Claude only the rest; widen it to the full caseload once that batch holds up.", "correct": false, "explanation": "The pilot runs before the process has been divided at all, which means the batch tests an undefined arrangement and real volunteer references are handled with no safeguard in place."}], "correct": "C", "select": 1}, {"id": "a6-005", "domain": 6, "task_id": "6.2", "objective": "Apply data sensitivity, regulatory, and privacy considerations", "situation": "An IT service desk is updating the onboarding slide it uses to explain how staff conversations relate to model training. Which two statements are accurate to put on the slide?", "question": "Select two.", "options": [{"letter": "A", "text": "Switching model improvement on applies it to every conversation already in the user's account.", "correct": false, "explanation": "Enabling the setting applies only to new or resumed chats after it is turned on; it does not reach back over untouched past conversations."}, {"letter": "B", "text": "Work done in the company's Claude for Work account is not used to train models by default.", "correct": true, "explanation": "On commercial offerings, including Claude for Work, inputs and outputs are not used for model training by default."}, {"letter": "C", "text": "Personal Free and Pro accounts are also excluded from model training by default, as Work is too.", "correct": false, "explanation": "Consumer plans work the other way round: the user chooses, so training is a setting rather than an automatic exclusion."}, {"letter": "D", "text": "Deleting a conversation immediately removes every copy of it from all of Anthropic's systems.", "correct": false, "explanation": "Deletion is not instantaneous everywhere, and some material is retained for safety and legal purposes, so this would mislead staff."}, {"letter": "E", "text": "Incognito chats are never used to improve Claude, whatever the account's other settings say.", "correct": true, "explanation": "and for a separate reason: the incognito exclusion is absolute and does not depend on how the model improvement setting is configured."}], "correct": ["B", "E"], "select": 2}, {"id": "a6-006", "domain": 6, "task_id": "6.4", "objective": "Understand the ethical implications of AI usage", "situation": "A senior analyst now has Claude draft her weekly market commentary and forwards it after only a quick skim. Colleagues read the commentary as her expert judgment and make purchasing decisions on it.", "question": "What is the greatest ethical concern?", "options": [{"letter": "A", "text": "The commentary may read in a slightly different tone from the weekly notes she wrote herself.", "correct": false, "explanation": "A change of tone is noticeable but harmless compared with decisions being made on unchecked reasoning."}, {"letter": "B", "text": "Claude may return a longer document than colleagues have time to read ahead of their meetings.", "correct": false, "explanation": "Length affects convenience; it does not change whether the content has been verified by anyone."}, {"letter": "C", "text": "Colleagues act on reasoning nobody has checked, while crediting it to her expert judgment.", "correct": true, "explanation": "The expertise colleagues believe they are relying on has quietly been withdrawn: nobody has examined the reasoning, yet the analyst's name still carries the authority behind the decisions."}, {"letter": "D", "text": "Producing the commentary this way each week may use more of her plan's allowance than before.", "correct": false, "explanation": "Allowance consumption is a cost question, not an ethical one, and does not affect who is accountable."}], "correct": "C", "select": 1}, {"id": "a6-007", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "A journalist uses Claude to help draft portions of an article that will be published under the news outlet's byline.", "question": "Under Anthropic's Usage Policy guidance for high-risk journalistic content, what is required?", "options": [{"letter": "A", "text": "The journalist must obtain a signed release from Anthropic for each article that includes any AI-assisted content.", "correct": false, "explanation": "Anthropic does not require a per-article signed release; the obligation is editorial review and disclosure by the outlet."}, {"letter": "B", "text": "The outlet must apply appropriate human editorial review and disclosure practices regarding AI involvement before publication.", "correct": true, "explanation": "Journalistic content requires human editorial review and appropriate disclosure of AI involvement before publication."}, {"letter": "C", "text": "The outlet must replace every AI-assisted sentence with fully original human-written text before the article can be published.", "correct": false, "explanation": "The policy requires editorial review and disclosure, not a blanket ban on any AI-assisted text appearing in the article."}, {"letter": "D", "text": "The journalist may publish the AI-drafted sections without any additional review, since journalism is exempt from the high-risk category.", "correct": false, "explanation": "Journalistic content generated for publication is treated as a high-risk use case, not an exempt category."}], "correct": "B", "select": 1}, {"id": "a6-008", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "A team lead asks whether using incognito chats on his personal plan makes it acceptable to paste regulated customer records into Claude.", "question": "What is the best response?", "options": [{"letter": "A", "text": "Incognito chats are equivalent to the company's managed workspace, so no separate approval would be required.", "correct": false, "explanation": "A personal plan in incognito mode is not the managed workspace and carries none of its contractual or administrative controls."}, {"letter": "B", "text": "Incognito chats are never used to improve Claude, but the policy's rules for regulated data still apply.", "correct": true, "explanation": "Incognito chats are not used to improve Claude, but the policy question is whether regulated data may be processed in that environment at all, and a privacy toggle does not answer it."}, {"letter": "C", "text": "Incognito chats satisfy the policy, because nothing from the conversation is retained anywhere after it ends.", "correct": false, "explanation": "Overstating what incognito does is the trap here; the policy governs where regulated data may go, not only whether it trains a model."}, {"letter": "D", "text": "Incognito chats are acceptable for regulated data provided customer names are replaced with reference codes.", "correct": false, "explanation": "Pseudonymising fields does not make regulated records unregulated, and the approval question remains open."}], "correct": "B", "select": 1}, {"id": "a6-009", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "A communications specialist has drafted an industry report with Claude's help, to be published under the company's name. The AI policy requires both disclosure of AI assistance and named human sign-off.", "question": "Which sequence is correct?", "options": [{"letter": "A", "text": "Add the disclosure the policy requires; publish the report; verify the facts and figures against the original sources; obtain the named reviewer's sign-off.", "correct": false, "explanation": "A disclosure added before anyone has checked the content states only how it was made, not that it is fit to publish."}, {"letter": "B", "text": "Publish the report; verify the facts and figures against the original sources; obtain the named reviewer's sign-off; add the disclosure the policy requires.", "correct": false, "explanation": "Publishing first makes every later step a remediation of material that is already public."}, {"letter": "C", "text": "Obtain the named reviewer's sign-off; publish the report; verify the facts and figures against the original sources; add the disclosure the policy requires.", "correct": false, "explanation": "Sign-off given before any verification asks the reviewer to vouch for figures nobody has checked."}, {"letter": "D", "text": "Verify the facts and figures against the original sources; obtain the named reviewer's sign-off; add the disclosure the policy requires; publish the report.", "correct": true, "explanation": "Verification supports the reviewer's sign-off, the disclosure is attached to what was signed off, and publication comes last because it is the irreversible step."}], "correct": "D", "select": 1}, {"id": "a6-010", "domain": 6, "task_id": "6.1", "objective": "Identify appropriate and inappropriate use cases", "situation": "A DevOps team configures Claude with tool access to restart non-critical staging servers and open low-priority tickets automatically, while any action affecting production systems requires explicit human approval before execution.", "question": "Is this an appropriate use of Claude?", "options": [{"letter": "A", "text": "No, because opening tickets automatically counts as a high-risk action needing professional review", "correct": false, "explanation": "Opening low-priority tickets in a staging environment is not the kind of high-risk domain, such as legal or healthcare, that triggers professional-review-and-disclosure requirements."}, {"letter": "B", "text": "No, because the Usage Policy prohibits granting Claude any tool access to servers under all circumstances", "correct": false, "explanation": "The Usage Policy does not prohibit tool access to servers outright; agentic tool use is a supported capability when scoped and overseen appropriately."}, {"letter": "C", "text": "Yes, but only after Claude is also granted full, unreviewed access to production systems directly", "correct": false, "explanation": "Removing human approval from production changes would increase risk rather than being a requirement; the described scenario is already appropriate as configured."}, {"letter": "D", "text": "Yes, because low-stakes automated actions are allowed while production changes need human approval", "correct": true, "explanation": "Allowing autonomous action on low-stakes staging tasks while requiring human approval for production changes reflects an appropriate, risk-proportionate use of agentic tool access."}], "correct": "D", "select": 1}, {"id": "a6-011", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "A communications team wants to start using a Claude capability on client-facing work, and believes the company's written AI standard may not address it. The standard itself says that anything it does not cover needs the policy owner's position before anyone uses it, even on internal material.", "question": "In which order should these steps be taken?", "options": [{"letter": "A", "text": "Check whether the standard covers the capability at all; ask the policy owner to give a position on the capability; pilot the capability on internal material only; roll it out to client work and record the agreed position in the standard.", "correct": true, "explanation": "Reading the standard establishes whether there is a gap at all; only a real gap justifies asking the policy owner for a position, which the standard requires before any use; an internal pilot then tests the capability where a mistake stays inside the company; and updating the standard makes the decision durable for everyone else."}, {"letter": "B", "text": "Ask the policy owner to give a position on the capability; check whether the standard covers the capability at all; pilot the capability on internal material only; roll it out to client work and record the agreed position in the standard.", "correct": false, "explanation": "The policy owner is asked to rule before anyone has checked whether the existing standard already covers the capability, which wastes an escalation and risks a new position that contradicts the written rules. Asking is only needed for something the standard leaves uncovered."}, {"letter": "C", "text": "Pilot the capability on internal material only; check whether the standard covers the capability at all; ask the policy owner to give a position on the capability; roll it out to client work and record the agreed position in the standard.", "correct": false, "explanation": "Use begins before anyone has even opened the standard, and before the policy owner has said anything, which breaks the standard's own rule and is the sequence that produces shadow adoption and leaves governance to catch up afterwards."}, {"letter": "D", "text": "Check whether the standard covers the capability at all; pilot the capability on internal material only; ask the policy owner to give a position on the capability; roll it out to client work and record the agreed position in the standard.", "correct": false, "explanation": "The pilot starts before the policy owner has given a position, but the standard requires that position before anyone uses an uncovered capability, even on internal material. Keeping the pilot on lower-stakes material does not make it sanctioned."}], "correct": "A", "select": 1}, {"id": "a6-012", "domain": 6, "task_id": "6.2", "objective": "Apply data sensitivity, regulatory, and privacy considerations", "situation": "A recruiter wants Claude to compare a set of candidate CVs against a job description.", "question": "What is the most appropriate handling of the data?", "options": [{"letter": "A", "text": "Include everything and rely on the plan's privacy settings to protect the candidates' personal data.", "correct": false, "explanation": "Plan settings are not a substitute for deciding what personal data needs to be there at all."}, {"letter": "B", "text": "Include the names but remove the skills, because a name on its own cannot identify anyone in practice.", "correct": false, "explanation": "Names are identifiers, and removing the skills destroys the point of the comparison."}, {"letter": "C", "text": "Remove the names and contact details, keeping only the skills and experience the comparison needs.", "correct": true, "explanation": "Data minimisation keeps the analysis useful while removing identifiers it does not require."}, {"letter": "D", "text": "Include the full CVs, since removing any details would make the comparison less accurate for hiring.", "correct": false, "explanation": "The comparison is against skills and experience, which survives removing identifiers."}], "correct": "C", "select": 1}, {"id": "a6-013", "domain": 6, "task_id": "6.4", "objective": "Understand the ethical implications of AI usage", "situation": "A marketing team asks Claude to generate a series of realistic-looking news articles attributed to fictional journalists, intending to distribute them to sway public opinion on an upcoming ballot measure.", "question": "Which ethical and policy concern does this scenario raise most directly?", "options": [{"letter": "A", "text": "It raises no concern at all, since the ballot measure itself is a matter of legitimate public interest", "correct": false, "explanation": "Public interest in the topic does not exempt content that relies on fabricated authorship and deception to manipulate opinion."}, {"letter": "B", "text": "It becomes a violation only if the articles name a specific candidate rather than a ballot measure", "correct": false, "explanation": "The prohibition covers deceptive election-related content broadly, including ballot measures, not only candidate-specific content."}, {"letter": "C", "text": "It is acceptable in this case because marketing teams are exempt from Usage Policy content rules", "correct": false, "explanation": "The Usage Policy applies to all users and use cases; there is no blanket exemption for marketing teams."}, {"letter": "D", "text": "It violates the policy's ban on deceptive content and fake personas used to manipulate civic discourse", "correct": true, "explanation": "Creating fake personas and deceptive political content to manipulate civic processes is explicitly prohibited under the Usage Policy's misinformation and election-integrity provisions."}], "correct": "D", "select": 1}, {"id": "a6-014", "domain": 6, "task_id": "6.4", "objective": "Understand the ethical implications of AI usage", "situation": "A company deploying Claude in a customer support agent notices that the model occasionally produces confident-sounding but factually incorrect answers about product specifications.", "question": "From an ethical usage standpoint, what is the most appropriate response?", "options": [{"letter": "A", "text": "Ground responses in verified documentation and route uncertain answers to human agents", "correct": true, "explanation": "Grounding responses in verified sources and escalating uncertain or high-stakes cases to humans directly addresses the risk of confident hallucinations while keeping the tool useful."}, {"letter": "B", "text": "Instruct the model through the system prompt to always sound highly confident to build trust", "correct": false, "explanation": "Instructing the model to sound more confident would worsen the ethical problem by making incorrect answers even more convincing."}, {"letter": "C", "text": "Accept the current error rate as a normal and unavoidable cost of generative AI in support", "correct": false, "explanation": "Treating hallucination risk as unavoidable ignores practical mitigations like retrieval grounding and human escalation that responsible deployments should use."}, {"letter": "D", "text": "Disable the agent's ability to mention any product specifications so it cannot be factually wrong", "correct": false, "explanation": "Disabling all specification discussion removes the tool's core value rather than addressing the accuracy problem responsibly."}], "correct": "A", "select": 1}, {"id": "a6-015", "domain": 6, "task_id": "6.1", "objective": "Identify appropriate and inappropriate use cases", "situation": "A researcher asks Claude to identify specific misconfigurations in a regional power grid's control software so that substations can be remotely shut down during a labor dispute to pressure the utility.", "question": "Is this an appropriate use of Claude?", "options": [{"letter": "A", "text": "No, because the Usage Policy bans using Claude to help disrupt critical infrastructure like power grids", "correct": true, "explanation": "The Usage Policy prohibits assisting with attacks on critical infrastructure, including power grids, healthcare systems, and voting systems, regardless of the stated motive."}, {"letter": "B", "text": "Yes, because analyzing control-software misconfigurations is a standard security-audit task", "correct": false, "explanation": "The framing as a security audit does not change that the stated goal is to enable an unauthorized shutdown of live infrastructure, which the policy prohibits."}, {"letter": "C", "text": "No, but only because the researcher lacks authorization from the utility company IT department", "correct": false, "explanation": "Even with utility authorization, deliberately shutting down substations to pressure the organization during a dispute is a disruptive act the policy is designed to prevent."}, {"letter": "D", "text": "Yes, because the goal here is short-term negotiation leverage rather than permanent physical damage", "correct": false, "explanation": "Intent to use disruption as leverage rather than cause permanent damage does not make the request acceptable; disrupting critical infrastructure operations is itself prohibited."}], "correct": "A", "select": 1}, {"id": "a6-016", "domain": 6, "task_id": "6.4", "objective": "Understand the ethical implications of AI usage", "situation": "A charity communications officer is preparing a Claude-assisted case study about a named service user for the charity's public annual review.", "question": "In which order should these steps be taken?", "options": [{"letter": "A", "text": "Publish it and keep a note of what was checked; confirm that the service user consents to her story being published; confirm that the draft says nothing about her that the case records do not support; state in the piece that Claude helped write it.", "correct": false, "explanation": "The case study goes public before consent, accuracy and disclosure have been dealt with, so every safeguard is applied to something the service user has already been exposed by."}, {"letter": "B", "text": "Confirm that the draft says nothing about her that the case records do not support; confirm that the service user consents to her story being published; state in the piece that Claude helped write it; publish it and keep a note of what was checked.", "correct": false, "explanation": "Effort goes into verifying a story the subject may not have agreed to tell, and consent is sought only after the charity has already worked her personal circumstances into a publishable draft."}, {"letter": "C", "text": "Confirm that the service user consents to her story being published; confirm that the draft says nothing about her that the case records do not support; state in the piece that Claude helped write it; publish it and keep a note of what was checked.", "correct": true, "explanation": "Consent from the person whose story it is comes first, because without it the piece should not exist; the draft is then checked against the records so that nothing invented is attributed to her; the disclosure of AI assistance is added before publication rather than after; and the record of what was checked makes the whole thing accountable later."}, {"letter": "D", "text": "State in the piece that Claude helped write it; confirm that the draft says nothing about her that the case records do not support; confirm that the service user consents to her story being published; publish it and keep a note of what was checked.", "correct": false, "explanation": "The disclosure is settled first, which is the least urgent of the four, while the two questions that protect the service user are left until later in the sequence."}], "correct": "C", "select": 1}, {"id": "a6-017", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "", "question": "A developer at a company is testing techniques to bypass Claude's safety guidelines so the model will produce restricted content, and shares these prompts internally as \"productivity tips.\" Under Anthropic's Usage Policy, this activity is:", "options": [{"letter": "A", "text": "Only a violation of the Usage Policy if the company later publishes the jailbreak techniques externally, as internal use for productivity tips is allowed.", "correct": false, "explanation": "The violation occurs when an attempt is made to circumvent Claude's safety features, irrespective of whether the techniques are later published externally. Internal use as 'productivity tips' does not make the activity permissible under the policy."}, {"letter": "B", "text": "Permitted because the Usage Policy only restricts external-facing production use, so internal jailbreak experimentation is not prohibited.", "correct": false, "explanation": "The Usage Policy does not limit its prohibitions to external-facing production use; it covers any use of Claude, including internal experimentation. Attempting to bypass safety guidelines internally is therefore still a violation."}, {"letter": "C", "text": "Acceptable internal research, provided the resulting content is kept strictly internal and is never disclosed to customers or the public.", "correct": false, "explanation": "Anthropic's Usage Policy applies to all uses of Claude, including internal research and testing, so jailbreaking attempts are not permissible even if kept strictly internal. The restriction targets the act of circumventing safety measures itself, not merely the external sharing of results."}, {"letter": "D", "text": "A violation of the Universal Usage Standards, which prohibit jailbreaking and other attempts to circumvent Claude's safety measures.", "correct": true, "explanation": "The Universal Usage Standards explicitly prohibit jailbreaking and any attempts to bypass Claude's safety mechanisms, regardless of the internal or external context. Labeling the prompts as 'productivity tips' does not exempt this activity from the policy."}], "correct": "D", "select": 1}, {"id": "a6-018", "domain": 6, "task_id": "6.4", "objective": "Understand the ethical implications of AI usage", "situation": "A consumer-facing startup builds a chatbot on Claude for its e-commerce site and wants to follow Anthropic's guidance on transparency.", "question": "What should the company implement at the start of a customer's chat session?", "options": [{"letter": "A", "text": "No disclosure at all, since chatbots are exempt from AI disclosure requirements under the policy", "correct": false, "explanation": "Consumer chatbots are explicitly called out as needing AI disclosure at the start of the session, not exempted from it."}, {"letter": "B", "text": "A one-time disclosure buried in the site's general terms of service that customers rarely read", "correct": false, "explanation": "A buried terms-of-service mention does not meet the requirement for clear, session-level disclosure that a customer would actually see."}, {"letter": "C", "text": "A clear notice at the start of the session disclosing that the customer is talking to an AI system", "correct": true, "explanation": "Anthropic's guidance for consumer-facing chatbot applications requires disclosing AI involvement at the start of the session so users know they are talking to an AI."}, {"letter": "D", "text": "A hidden instruction telling Claude to claim it is a human agent whenever a customer asks directly", "correct": false, "explanation": "Instructing the model to impersonate a human and deny being AI directly violates the Usage Policy's prohibition on deceptive impersonation."}], "correct": "C", "select": 1}, {"id": "a6-019", "domain": 6, "task_id": "6.2", "objective": "Apply data sensitivity, regulatory, and privacy considerations", "situation": "A Claude.ai Pro user's conversation is flagged by automated safety classifiers, which generate safety classification scores in addition to the conversation content.", "question": "How long may Anthropic retain those safety classification scores compared to the flagged inputs and outputs?", "options": [{"letter": "A", "text": "Safety classification scores may be retained for up to 7 years, longer than the 2-year window for flagged inputs and outputs", "correct": true, "explanation": "Under the consumer retention policy, flagged inputs and outputs may be retained up to 2 years, while safety classification scores may be retained for up to 7 years."}, {"letter": "B", "text": "Safety classification scores are deleted immediately, while flagged inputs and outputs are retained for up to 7 years", "correct": false, "explanation": "This reverses the actual durations; safety classification scores are the ones retained longer, up to 7 years, not deleted immediately."}, {"letter": "C", "text": "Both safety classification scores and flagged inputs and outputs are retained for the same 5-year window", "correct": false, "explanation": "The 5-year figure applies to the optional model improvement and feedback retention windows, not to trust and safety flagged content or its scores."}, {"letter": "D", "text": "Safety classification scores are retained for 30 days, matching the standard back-end deletion window", "correct": false, "explanation": "The 30-day window applies to standard back-end deletion of routine deleted conversations, not to safety classification scores from flagged content."}], "correct": "A", "select": 1}, {"id": "a6-020", "domain": 6, "task_id": "6.1", "objective": "Identify appropriate and inappropriate use cases", "situation": "A parenting-app startup wants to build a chatbot for teenagers using Claude, marketed as a virtual \"friend\" that engages in romantic and sexual role-play with users who state they are 15 years old.", "question": "Is this an appropriate use of Claude?", "options": [{"letter": "A", "text": "Yes, because these interactions are fictional role-play rather than real human contact or communication", "correct": false, "explanation": "Fictional framing does not exempt content from the Usage Policy's prohibition on sexualized interactions with users identified as minors."}, {"letter": "B", "text": "No, because the Usage Policy bans sexual role-play with minors regardless of consent or fictional framing", "correct": true, "explanation": "The Usage Policy prohibits sexual content involving minors outright, and framing the interaction as fictional role-play or obtaining consent does not exempt it from this rule."}, {"letter": "C", "text": "Yes, as long as a parental consent form is added to the app before this feature launches publicly", "correct": false, "explanation": "Parental consent does not override the Usage Policy's prohibition on sexual content involving minors; this category of harm is not waivable through consent forms."}, {"letter": "D", "text": "No, but only because the app has not integrated a proper third-party age-verification API service", "correct": false, "explanation": "Adding age verification does not resolve the issue, since the underlying content itself is prohibited regardless of how age is confirmed."}], "correct": "B", "select": 1}, {"id": "a6-021", "domain": 6, "task_id": "6.1", "objective": "Identify appropriate and inappropriate use cases", "situation": "An edtech company builds a Claude-powered tutoring app aimed at children aged 8 to 12, using the same content settings and data handling as its adult-facing tutoring product, with no additional protections added.", "question": "Is this an appropriate use of Claude?", "options": [{"letter": "A", "text": "Yes, because educational tutoring content is automatically exempt from minor-protection requirements", "correct": false, "explanation": "Being educational in nature does not exempt an application from the enhanced protections required when the audience includes children."}, {"letter": "B", "text": "No, but only because the company failed to obtain a separate certification before this launch", "correct": false, "explanation": "The issue is the absence of enhanced protections built into the product itself, not a missing paperwork certification."}, {"letter": "C", "text": "No, because the Usage Policy requires enhanced protections for products serving minors specifically", "correct": true, "explanation": "The Usage Policy calls for additional protections when a product serves minors, and reusing adult-facing settings unchanged fails to meet that requirement."}, {"letter": "D", "text": "Yes, because the Usage Policy treats all age groups identically once parental consent is obtained", "correct": false, "explanation": "General parental consent does not substitute for the enhanced, age-appropriate protections the Usage Policy requires for products aimed at minors."}], "correct": "C", "select": 1}, {"id": "a6-022", "domain": 6, "task_id": "6.4", "objective": "Understand the ethical implications of AI usage", "situation": "A team is deciding whether to use Claude to fully automate final housing loan approval decisions with no human involvement at any stage, in order to reduce processing time.", "question": "What is the primary ethical risk of this design choice?", "options": [{"letter": "A", "text": "The risk is purely a data storage concern about how long applicant records are retained", "correct": false, "explanation": "Data retention is a separate privacy topic and does not capture the core risk of unreviewed, automated high-stakes decisions."}, {"letter": "B", "text": "The only real risk is a minor increase in latency compared to a human-reviewed workflow", "correct": false, "explanation": "The primary concern is harm from unreviewed erroneous decisions, not processing speed, which full automation would actually improve."}, {"letter": "C", "text": "There is no meaningful risk, since housing loans are not a high-risk category under the policy", "correct": false, "explanation": "Housing decisions are explicitly treated as high-risk given their public welfare and social equity implications."}, {"letter": "D", "text": "Removing review from a high-impact decision like housing risks harming applicants unchecked", "correct": true, "explanation": "Housing is a high-risk category with significant welfare implications, so removing human review entirely means errors or biased outcomes could directly harm applicants with no safeguard to catch them."}], "correct": "D", "select": 1}, {"id": "a6-023", "domain": 6, "task_id": "6.1", "objective": "Identify appropriate and inappropriate use cases", "situation": "A cybersecurity vendor asks Claude to reverse-engineer a competitor's software and write exploit code for a previously undisclosed vulnerability, intending to sell the exploit on a private marketplace.", "question": "Is this an appropriate use of Claude?", "options": [{"letter": "A", "text": "Yes, because security research and vulnerability analysis are always permitted under the Usage Policy", "correct": false, "explanation": "Security research is not blanket-permitted; the Usage Policy specifically prohibits developing exploits and malware for unauthorized offensive use, regardless of the requester's field."}, {"letter": "B", "text": "No, but only because selling the exploit privately infringes the competitor's copyright", "correct": false, "explanation": "The core problem is the unauthorized creation of offensive exploit code, not a copyright dispute over the resale of the finished exploit."}, {"letter": "C", "text": "No, because the Usage Policy bans creating exploits and malware meant for unauthorized offensive use", "correct": true, "explanation": "Anthropic's Usage Policy prohibits using Claude to create malware, exploits, or other tools designed to exploit vulnerabilities without authorization, which is exactly what this scenario describes."}, {"letter": "D", "text": "Yes, provided the vendor labels the finished exploit code as AI-generated before selling it", "correct": false, "explanation": "Labeling content as AI-generated does not make a prohibited use case permissible; disclosure is not a substitute for complying with the underlying restriction."}], "correct": "C", "select": 1}, {"id": "a6-024", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "A consultancy's policy requires AI assistance to be disclosed on client deliverables and forbids client data on personal accounts. A consultant drafted a client report with Claude on her personal Pro account. Which two things must she now do?", "question": "Select two.", "options": [{"letter": "A", "text": "Ask the client to confirm in writing that they are comfortable receiving an AI-assisted written report.", "correct": false, "explanation": "Client comfort is not the compliance question; the policy sets the disclosure requirement, and the client does not waive it."}, {"letter": "B", "text": "Redo the work in the firm's managed Claude for Work workspace before the report goes to the client.", "correct": true, "explanation": "The account boundary is a hard policy rule, so the deliverable has to be produced where client data is permitted to sit."}, {"letter": "C", "text": "Rewrite the whole report by hand, since heavy editing removes the need to disclose the AI assistance.", "correct": false, "explanation": "Editing the draft heavily does not remove the disclosure obligation, which attaches to how the work was produced."}, {"letter": "D", "text": "Report the personal-account use through the firm's incident route so the exposure can be assessed.", "correct": true, "explanation": "on a separate axis: the breach has already happened, and disclosure of the incident is what lets the firm judge and contain the exposure."}, {"letter": "E", "text": "Delete the conversations from her personal account so that no record of the breach remains anywhere at all.", "correct": false, "explanation": "Deleting the evidence conceals the incident and prevents the firm from assessing what client data was exposed."}], "correct": ["B", "D"], "select": 2}, {"id": "a6-025", "domain": 6, "task_id": "6.2", "objective": "Apply data sensitivity, regulatory, and privacy considerations", "situation": "An employee connects their entire work Google Drive so that Claude can help with the files in one project folder.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The files are copied into training data, because connected sources bypass the account's privacy settings.", "correct": false, "explanation": "Connecting a source does not place its files into model training."}, {"letter": "B", "text": "The connection makes the files editable by anyone in the organisation who opens the same project.", "correct": false, "explanation": "A connection does not change who may edit files in the underlying service."}, {"letter": "C", "text": "Colleagues who cannot open those files in Drive will now be able to read them through the connector.", "correct": false, "explanation": "Connectors respect the source system's permissions, so they cannot widen access for other people."}, {"letter": "D", "text": "Everything they can already open in Drive becomes reachable in Claude, far beyond the folder they need.", "correct": true, "explanation": "A connector inherits the person's own access, so an unscoped connection exposes far more than intended."}], "correct": "D", "select": 1}, {"id": "a6-026", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "A user requests deletion of their personal data from conversation logs.", "question": "Which action best satisfies this request in accordance with data privacy regulations?", "options": [{"letter": "A", "text": "Anonymize the user’s data so it cannot be linked back to them but keep the logs for analysis.", "correct": false, "explanation": "Anonymizing data may reduce identifiability but does not fully comply with deletion requests under regulations like GDPR, which typically require complete removal of personal data unless a legitimate reason exists to retain it in anonymized form. Keeping the logs for analysis directly conflicts with the expectation that personal data be deleted."}, {"letter": "B", "text": "Export the conversation logs into a secure archive and then delete the active records.", "correct": false, "explanation": "Exporting logs to a secure archive before deleting active records preserves the data rather than eliminating it, which conflicts with the deletion request unless there is a legal basis for retention. This approach does not fully comply unless the archived data is also deleted or anonymized."}, {"letter": "C", "text": "Delete all conversation logs containing the user’s personal data and confirm removal promptly.", "correct": true, "explanation": "Deleting all conversation logs containing the user's personal data and confirming removal promptly directly honors the deletion request and aligns with data privacy regulations such as GDPR's right to erasure. This approach represents the most appropriate operational handling of the request."}, {"letter": "D", "text": "Disable the API key used during those conversations to permanently prevent future data retention.", "correct": false, "explanation": "Disabling the API key prevents future data collection but does not remove existing conversation logs or personal data. This action addresses access control, not data deletion, and therefore fails to satisfy the user's request."}], "correct": "C", "select": 1}, {"id": "a6-027", "domain": 6, "task_id": "6.2", "objective": "Apply data sensitivity, regulatory, and privacy considerations", "situation": "An analyst has been asked to summarise a spreadsheet of customer complaints that includes names and account numbers.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Confirm the policy for this data; remove identifiers not needed; run the summary in the approved workspace; store the output under the same rules.", "correct": true, "explanation": "Clear the use, minimise the data, do the work where it is permitted, then keep the output under the same controls."}, {"letter": "B", "text": "Run the summary in the approved workspace; confirm the policy for this data; store the output under the same rules; remove identifiers not needed.", "correct": false, "explanation": "Full identifiers are processed first and stripped only at the very end."}, {"letter": "C", "text": "Store the output under the same rules; remove identifiers not needed; confirm the policy for this data; run the summary in the approved workspace.", "correct": false, "explanation": "An output is stored before it has been produced, which reverses the whole task."}, {"letter": "D", "text": "Remove identifiers not needed; run the summary in the approved workspace; store the output under the same rules; confirm the policy for this data.", "correct": false, "explanation": "The work is completed before anyone confirms it was permitted."}], "correct": "A", "select": 1}, {"id": "a6-028", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "A project manager drafted a board memo with Claude and then edited it heavily. Her company's AI policy requires AI assistance to be disclosed on internal documents.", "question": "What is the most appropriate action?", "options": [{"letter": "A", "text": "Add a standing footer to every document she writes, so no individual disclosure decision is ever needed.", "correct": false, "explanation": "A blanket footer on everything makes the disclosure meaningless and misdescribes documents written without assistance."}, {"letter": "B", "text": "Mention the AI assistance verbally at the board meeting rather than recording it in the memo itself.", "correct": false, "explanation": "A spoken remark leaves no record on the document itself, so anyone reading it later has no disclosure."}, {"letter": "C", "text": "Treat disclosure as unnecessary here, because her edits make the final wording substantially her own work.", "correct": false, "explanation": "Editing changes the wording but not the fact that AI assistance was used, which is what the policy asks to be recorded."}, {"letter": "D", "text": "State in the memo that it was drafted with AI assistance and reviewed by her, as the policy requires.", "correct": true, "explanation": "The policy requires disclosure, and naming both the assistance and the human reviewer tells readers what was generated and who stands behind it."}], "correct": "D", "select": 1}, {"id": "a6-029", "domain": 6, "task_id": "6.2", "objective": "Apply data sensitivity, regulatory, and privacy considerations", "situation": "A Claude.ai Free plan user deletes a conversation from their chat history.", "question": "According to Anthropic's consumer data retention policy, when is that conversation removed from Anthropic's back-end storage systems?", "options": [{"letter": "A", "text": "Immediately at deletion, since consumer chats bypass backend storage entirely", "correct": false, "explanation": "Deletion is immediate only from the visible chat history; back-end systems still take up to 30 days to fully purge the data."}, {"letter": "B", "text": "Within 5 years, since deleted chats fall under the model improvement retention rule", "correct": false, "explanation": "The 5-year window applies to data retained under the optional model improvement setting, not to deleted conversations that were never opted in."}, {"letter": "C", "text": "Within 2 years, matching the retention window used for trust and safety reviews", "correct": false, "explanation": "The 2-year window applies specifically to inputs and outputs flagged for trust and safety violations, not to routine user-initiated deletions."}, {"letter": "D", "text": "Within 30 days of the deletion request, per Anthropic's standard back-end purge window", "correct": true, "explanation": "Deleted conversations are removed from chat history immediately, then purged from Anthropic's back-end storage systems within 30 days."}], "correct": "D", "select": 1}, {"id": "a6-030", "domain": 6, "task_id": "6.1", "objective": "Identify appropriate and inappropriate use cases", "situation": "", "question": "Which of the following is the strongest candidate for delegating to Claude within a marketing team?", "options": [{"letter": "A", "text": "Drafting first-pass social copy from an approved brief, with an editor reviewing it before anything is posted.", "correct": true, "explanation": "Reversible, well-specified drafting with a human reviewing before release is the ideal delegation shape."}, {"letter": "B", "text": "Sending the price change notices straight to customers as soon as they are generated, without any review at all.", "correct": false, "explanation": "This is irreversible and externally visible, so it needs review before it goes out."}, {"letter": "C", "text": "Choosing which employees are promoted, based on the performance notes gathered from their managers.", "correct": false, "explanation": "Employment decisions affect people's livelihoods and must remain a human judgment."}, {"letter": "D", "text": "Deciding which customers are refused credit, using the payment histories held in the finance system.", "correct": false, "explanation": "Credit decisions affect people's rights and finances and cannot be delegated outright."}], "correct": "A", "select": 1}, {"id": "a6-031", "domain": 6, "task_id": "6.2", "objective": "Apply data sensitivity, regulatory, and privacy considerations", "situation": "A small firm currently works on individual Pro accounts and wants to standardise how client data is handled.", "question": "What is the most relevant advantage of moving onto a business plan?", "options": [{"letter": "A", "text": "Members would no longer be able to upload any documents at all, which removes the risk of client data entering the tool.", "correct": false, "explanation": "Business plans do not remove the ability to upload documents."}, {"letter": "B", "text": "Conversations become encrypted, which is not possible on any of the individually managed accounts.", "correct": false, "explanation": "This is not the distinction between personal and business plans."}, {"letter": "C", "text": "The organisation gains owner-level control over connectors, retention and settings that individual accounts lack.", "correct": true, "explanation": "Business plans put configuration and data controls in the hands of accountable owners rather than each person."}, {"letter": "D", "text": "Individual accounts gain unlimited project storage, which removes any need to manage documents at all.", "correct": false, "explanation": "Storage is not the governance problem, and documents still need managing."}], "correct": "C", "select": 1}, {"id": "a6-032", "domain": 6, "task_id": "6.2", "objective": "Apply data sensitivity, regulatory, and privacy considerations", "situation": "A consultant wants to paste a client's confidential strategy deck into Claude in order to build a summary.", "question": "What matters most before doing so?", "options": [{"letter": "A", "text": "Whether the deck is short enough to fit into a single conversation without needing to be split.", "correct": false, "explanation": "Length is a practical detail and says nothing about whether the material may be used."}, {"letter": "B", "text": "Whether Claude's summary will be accurate enough to send to the client without any further editing.", "correct": false, "explanation": "Output quality is a separate question from whether the input was allowed to be shared."}, {"letter": "C", "text": "Whether the client's contract permits their material to be processed in a third-party tool at all.", "correct": true, "explanation": "Contractual confidentiality terms decide whether the material may be entered anywhere in the first place."}, {"letter": "D", "text": "Whether the consultant's colleagues will be able to see the summary in a shared project workspace.", "correct": false, "explanation": "Internal visibility matters later; permission to process the client's material comes first."}], "correct": "C", "select": 1}, {"id": "a6-033", "domain": 6, "task_id": "6.1", "objective": "Identify appropriate and inappropriate use cases", "situation": "A logistics firm proposes a workflow in which Claude reads supplier refund disputes and sends the final decision straight to the supplier with no review. Which two features of the proposal make it inappropriate?", "question": "Select two.", "options": [{"letter": "A", "text": "Refund thresholds are revised each quarter, so the standing instructions would have to be updated several times a year.", "correct": false, "explanation": "Keeping instructions current is ordinary maintenance and would apply to any well-run workflow, reviewed or not."}, {"letter": "B", "text": "Nobody in the firm is answerable for any individual determination, because no person sees a case before it is sent.", "correct": true, "explanation": "Accountability for each individual outcome is a separate problem from reversibility: no person is answerable for any single decision."}, {"letter": "C", "text": "Once the decision reaches the supplier it has already taken effect, so a wrong outcome cannot quietly be undone.", "correct": true, "explanation": "Sending a determination externally makes it effectively irreversible, and low reversibility is a strong signal that a human must be in the loop."}, {"letter": "D", "text": "Disputes arrive as long email threads, and Claude works far less reliably with threads than it does with structured forms.", "correct": false, "explanation": "Reading and summarising email threads is well within normal capability, so the format is not what makes the proposal unsuitable."}, {"letter": "E", "text": "The firm receives too few disputes each month for an automated workflow to save any meaningful amount of staff time.", "correct": false, "explanation": "Low volume might weaken the business case, but it is not a governance reason to reject the design."}], "correct": ["B", "C"], "select": 2}, {"id": "a6-034", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "An operations analyst wants to use a Claude capability that the company's AI policy does not mention at all.", "question": "What is the most appropriate action?", "options": [{"letter": "A", "text": "Ask Claude whether the capability would be consistent with the company's AI policy wording.", "correct": false, "explanation": "Claude cannot know the company's internal policy or authorise an exception to it."}, {"letter": "B", "text": "Ask the policy owner to decide before using it, describing the data and the task involved.", "correct": true, "explanation": "A gap in the policy is a question for whoever owns it, and describing the data and task gives them what they need to decide quickly."}, {"letter": "C", "text": "Avoid the capability altogether and assume it is banned until the policy is formally revised.", "correct": false, "explanation": "Assuming a ban is safe but needlessly forgoes value when a short question would settle it."}, {"letter": "D", "text": "Proceed with it, since anything the policy does not prohibit is permitted by default at work.", "correct": false, "explanation": "Treating silence as permission is how unreviewed data flows start, and it removes the owner's chance to decide."}], "correct": "B", "select": 1}, {"id": "a6-035", "domain": 6, "task_id": "6.2", "objective": "Apply data sensitivity, regulatory, and privacy considerations", "situation": "An analyst runs a research task that draws on connected internal mail and documents, then pastes large parts of the result into a public blog post.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "Publishing the output automatically grants the blog's readers access to the connected data sources.", "correct": false, "explanation": "Publishing text does not grant anyone access to the underlying systems."}, {"letter": "B", "text": "Research output is watermarked, so publishing any part of it identifies the analyst to everyone who reads the post.", "correct": false, "explanation": "There is no such identifying watermark on output."}, {"letter": "C", "text": "Internal material reached through the connected sources can be carried straight into text meant for the public.", "correct": true, "explanation": "Output built from internal sources may quote or paraphrase material that was never meant to leave the company."}, {"letter": "D", "text": "Research cannot draw on connected internal sources at all, so the analyst's result will be empty.", "correct": false, "explanation": "Research can use connected sources, which is precisely why the output needs checking."}], "correct": "C", "select": 1}, {"id": "a6-036", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "A finance team wants Claude to draw on the company's contract management system while it prepares supplier summaries, which means connecting that system to Claude. The company's AI standard governs how data sources are approved and recorded.", "question": "Which sequence is correct?", "options": [{"letter": "A", "text": "Confirm the AI standard covers the kind of data the contract system holds; obtain the data owner's approval to connect that system to Claude; connect it with the narrowest access the supplier summaries need; record the connection in the team's register of approved data sources.", "correct": true, "explanation": "The standard is checked first because it decides whether this data may be used at all, the data owner then approves, the connection is made no broader than the work needs, and the register is updated so the connection is visible later."}, {"letter": "B", "text": "Confirm the AI standard covers the kind of data the contract system holds; obtain the data owner's approval to connect that system to Claude; record the connection in the team's register of approved data sources; connect it with the narrowest access the supplier summaries need.", "correct": false, "explanation": "The register records a connection that does not exist yet, and the access scope is settled last."}, {"letter": "C", "text": "Connect it with the narrowest access the supplier summaries need; confirm the AI standard covers the kind of data the contract system holds; obtain the data owner's approval to connect that system to Claude; record the connection in the team's register of approved data sources.", "correct": false, "explanation": "The system is connected before the standard has been checked and before the owner has approved it."}, {"letter": "D", "text": "Obtain the data owner's approval to connect that system to Claude; confirm the AI standard covers the kind of data the contract system holds; connect it with the narrowest access the supplier summaries need; record the connection in the team's register of approved data sources.", "correct": false, "explanation": "Approval is sought before anyone has established what the standard permits for this kind of data."}], "correct": "A", "select": 1}, {"id": "a6-037", "domain": 6, "task_id": "6.1", "objective": "Identify appropriate and inappropriate use cases", "situation": "A robo-advisor platform lets Claude generate personalized investment recommendations that are sent directly to retail customers and executed automatically, with no human financial advisor reviewing any output.", "question": "Is this an appropriate use of Claude?", "options": [{"letter": "A", "text": "No, but only because automatic trade execution requires a separate brokerage license to operate", "correct": false, "explanation": "A brokerage license is a separate regulatory matter and does not address the missing professional review of the AI-generated recommendations themselves."}, {"letter": "B", "text": "No, because the Usage Policy requires a qualified professional to review high-risk financial content first", "correct": true, "explanation": "High-risk financial recommendations require review by a qualified professional before reaching customers, and this scenario skips that review entirely, violating the safeguard."}, {"letter": "C", "text": "Yes, because robo-advisory services are explicitly exempted from the Usage Policy's high-risk safeguards", "correct": false, "explanation": "Financial advice is a high-risk domain subject to the review-and-disclosure safeguards; there is no exemption for automated robo-advisory products."}, {"letter": "D", "text": "Yes, because customers implicitly consent to automated advice by choosing a robo-advisor product", "correct": false, "explanation": "Choosing to use a robo-advisor does not substitute for the required professional review of AI-generated financial recommendations before they take effect."}], "correct": "B", "select": 1}, {"id": "a6-038", "domain": 6, "task_id": "6.2", "objective": "Apply data sensitivity, regulatory, and privacy considerations", "situation": "A manager tells staff to draft sensitive client material on their own personal Claude accounts rather than on the company's Claude for Work account, because \"the Work account trains on our chats\".", "question": "What is the accurate correction?", "options": [{"letter": "A", "text": "The advice is right, but only for staff who have already switched model improvement off on their own accounts.", "correct": false, "explanation": "A personal privacy setting does not make a personal account the appropriate home for company material."}, {"letter": "B", "text": "The advice is right, because business accounts always carry much broader training permissions than personal ones do.", "correct": false, "explanation": "Business plans do not grant broader training permissions; the default runs in the opposite direction."}, {"letter": "C", "text": "The advice makes no difference, because the training rules are identical on personal and on business accounts.", "correct": false, "explanation": "The consumer and the commercial rules are different, which is precisely what the manager has confused."}, {"letter": "D", "text": "The advice has the risk backwards, since a training choice exists on personal accounts but not on the Work account.", "correct": true, "explanation": "Commercial products do not train on inputs and outputs by default, while consumer accounts carry a personal model improvement choice, so moving the work to a personal account increases exposure."}], "correct": "D", "select": 1}, {"id": "a6-039", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "An internal auditor finds that a colleague connected the company's issue-tracking system to Claude a month ago without the approval the AI policy requires.", "question": "What should the auditor do first?", "options": [{"letter": "A", "text": "Ask the colleague to carry on using it, but only for tickets that contain no customer detail.", "correct": false, "explanation": "This lets an unapproved connection continue on the auditor's own authority, substituting a personal judgement about sensitivity for the approval the policy requires."}, {"letter": "B", "text": "Wait for the annual policy review and raise it there as a general point about connectors.", "correct": false, "explanation": "Deferring a live unapproved connection to a scheduled review leaves it running for months and treats a specific breach as a general drafting point."}, {"letter": "C", "text": "Disconnect the system himself and take the matter no further, since the exposure has stopped.", "correct": false, "explanation": "Quietly removing the connection ends the access but hides the incident, so nobody assesses what was exposed during the month and the gap in the approval process goes unaddressed."}, {"letter": "D", "text": "Report the connection through the escalation route the policy sets out for exactly this case.", "correct": true, "explanation": "The policy already defines who is told about an unapproved data connection and what happens next, and a month of unreviewed access may need assessment beyond simply switching it off. Following the escalation route puts the decision with the people accountable for it."}], "correct": "D", "select": 1}, {"id": "a6-040", "domain": 6, "task_id": "6.2", "objective": "Apply data sensitivity, regulatory, and privacy considerations", "situation": "An Enterprise organisation wants a shorter data retention window applied to its chats and projects.", "question": "Who is able to make that change?", "options": [{"letter": "A", "text": "Someone holding the Primary Owner or the Owner role, in the organisation's settings.", "correct": true, "explanation": "Custom retention controls are restricted to the Primary Owner or an Owner."}, {"letter": "B", "text": "Any member of the organisation, from the privacy settings page of their own account.", "correct": false, "explanation": "Individual members cannot set organisation-level retention."}, {"letter": "C", "text": "Only Anthropic support staff, acting on a written request from the organisation's legal team.", "correct": false, "explanation": "The control is self-service for the appropriate owner roles."}, {"letter": "D", "text": "The person who created each individual project, applied to that one project only.", "correct": false, "explanation": "Retention is an organisation-level control, not a per-project setting."}], "correct": "A", "select": 1}, {"id": "a6-041", "domain": 6, "task_id": "6.2", "objective": "Apply data sensitivity, regulatory, and privacy considerations", "situation": "An employee is about to submit thumbs-down feedback on an answer that came from a conversation containing confidential figures.", "question": "What should they understand about doing so?", "options": [{"letter": "A", "text": "The related conversation goes with the feedback, so confidential content in that chat is submitted too.", "correct": true, "explanation": "Feedback carries the associated conversation, which is why rating a sensitive chat needs thought."}, {"letter": "B", "text": "Only a numeric score is sent, because the wording of a conversation is never attached to feedback.", "correct": false, "explanation": "Feedback is not reduced to a bare score."}, {"letter": "C", "text": "Only the single rated message is sent, and the rest of the conversation is excluded automatically.", "correct": false, "explanation": "The submission is not limited to the one message that was rated."}, {"letter": "D", "text": "Only the employee's own question is sent, because Claude's reply can simply be generated again later on.", "correct": false, "explanation": "The reply is not regenerated in place of being submitted."}], "correct": "A", "select": 1}, {"id": "a6-042", "domain": 6, "task_id": "6.4", "objective": "Understand the ethical implications of AI usage", "situation": "A political campaign wants to use Claude to draft policy summaries and speech outlines for a candidate, with campaign staff reviewing and editing every draft before use.", "question": "Does this use case violate the Usage Policy's provisions on political content?", "options": [{"letter": "A", "text": "Yes, because staff review does not remove the general prohibition on political content generation", "correct": false, "explanation": "Legitimate drafting assistance with human review is not prohibited; the concern is deception and manipulation, not campaign use itself."}, {"letter": "B", "text": "No, because political content restrictions only apply during the final 30 days before an election", "correct": false, "explanation": "There is no time-limited exception tied to a 30-day pre-election window in the policy's political content provisions."}, {"letter": "C", "text": "No, because the policy permits legitimate political discourse while banning deceptive tactics", "correct": true, "explanation": "The Usage Policy distinguishes legitimate political discourse and research, which is allowed, from deceptive or manipulative uses such as fake personas or targeted voter manipulation, which are prohibited."}, {"letter": "D", "text": "Yes, because any use of Claude in connection with a political campaign is prohibited outright", "correct": false, "explanation": "The policy does not ban all campaign-related use; it targets deceptive and manipulative practices specifically."}], "correct": "C", "select": 1}, {"id": "a6-043", "domain": 6, "task_id": "6.4", "objective": "Understand the ethical implications of AI usage", "situation": "A nonprofit wants to use Claude to draft outreach messages that quote scientific studies about vaccine safety, but a staff member suggests fabricating a plausible-sounding study citation when Claude cannot find a real one, in order to make the messaging more persuasive.", "question": "Why should the team reject this suggestion?", "options": [{"letter": "A", "text": "It is acceptable practice as long as the overall message matches mainstream scientific consensus", "correct": false, "explanation": "Alignment with consensus does not excuse fabricating evidence; deceptive sourcing is prohibited on its own terms."}, {"letter": "B", "text": "It should be rejected only because the fabricated study belongs to unpublished, unnamed authors", "correct": false, "explanation": "The core problem is fabrication and deception, not a copyright issue involving nonexistent, unpublished authors."}, {"letter": "C", "text": "It should be rejected only in jurisdictions that have strict health advertising laws on the books", "correct": false, "explanation": "The prohibition on deceptive health misinformation is not contingent on jurisdiction-specific advertising law."}, {"letter": "D", "text": "Fabricated citations are deceptive health misinformation, which the policy prohibits outright", "correct": true, "explanation": "Fabricating citations to bolster health-related claims is a form of deceptive misinformation, which the Usage Policy prohibits regardless of whether the underlying conclusion happens to be accurate."}], "correct": "D", "select": 1}, {"id": "a6-044", "domain": 6, "task_id": "6.2", "objective": "Apply data sensitivity, regulatory, and privacy considerations", "situation": "A Claude.ai Pro user wants their conversations excluded from any model training data, but they've enabled the optional \"model improvement\" setting and use Incognito chats occasionally.", "question": "What happens to their Incognito chats?", "options": [{"letter": "A", "text": "Incognito chats follow the 5-year retention rule instead of the standard 30-day deletion window", "correct": false, "explanation": "The 5-year rule applies to conversations retained under the model improvement setting, which does not apply to Incognito chats since they're excluded from training."}, {"letter": "B", "text": "Incognito chats are excluded from model training even when the model improvement setting is enabled", "correct": true, "explanation": "Incognito chats are specifically excluded from model training, even for users who have otherwise opted into the model improvement setting for their other conversations."}, {"letter": "C", "text": "Incognito chats are automatically flagged for trust and safety review regardless of content", "correct": false, "explanation": "Trust and safety flagging depends on detected policy violations in content, not on whether a chat was conducted in Incognito mode."}, {"letter": "D", "text": "Incognito chats are included in model training whenever the model improvement setting is turned on", "correct": false, "explanation": "The model improvement setting applies to a user's regular conversations; Incognito chats carry their own exclusion regardless of that setting."}], "correct": "B", "select": 1}, {"id": "a6-045", "domain": 6, "task_id": "6.4", "objective": "Understand the ethical implications of AI usage", "situation": "A developer integrating Claude into an internal analytics agent wants to give it broad, unsupervised authority to automatically send emails and modify production database records without any human approval step.", "question": "What ethical concern does this design most directly raise?", "options": [{"letter": "A", "text": "None, as long as the agent uses the Claude API rather than the consumer chat product", "correct": false, "explanation": "The oversight concern applies regardless of whether the integration uses the API or consumer product; both are subject to the same responsible-use principles."}, {"letter": "B", "text": "None, unless the agent is built using a third-party framework instead of Anthropic's own official tools", "correct": false, "explanation": "The choice of framework does not change the underlying risk of removing human oversight from consequential actions."}, {"letter": "C", "text": "It removes human oversight from hard-to-reverse actions, risking harm before anyone can intervene", "correct": true, "explanation": "Granting an agent unsupervised authority over irreversible, high-impact actions removes the human oversight needed to catch and correct model errors before they cause real harm."}, {"letter": "D", "text": "It is acceptable because agentic use is encouraged without oversight requirements in the policy", "correct": false, "explanation": "Agentic use must still comply with the full Usage Policy, and unsupervised high-stakes actions are not exempted from oversight expectations."}], "correct": "C", "select": 1}, {"id": "a6-046", "domain": 6, "task_id": "6.1", "objective": "Identify appropriate and inappropriate use cases", "situation": "A supplement company wants Claude to write blog posts claiming its product cures autoimmune disease, citing invented clinical trial results, in order to boost search-engine rankings.", "question": "Is this an appropriate use of Claude?", "options": [{"letter": "A", "text": "Yes, dietary supplement marketing is not subject to the Usage Policy's health-content rules.", "correct": false, "explanation": "Dietary supplement marketing is not exempt from the Usage Policy's restrictions on false medical claims; the prohibition on generating false health information applies regardless of product category. Thus, claiming a cure without evidence would still violate the policy."}, {"letter": "B", "text": "Yes, a small disclaimer stating unverified findings makes the blog posts acceptable under the Usage Policy.", "correct": false, "explanation": "Adding a disclaimer about unverified findings does not remedy the fundamental violation; fabricating clinical trial results constitutes making false medical claims, which the Usage Policy strictly prohibits."}, {"letter": "C", "text": "No, because the Usage Policy bans false or misleading medical claims, fabricating evidence is not allowed.", "correct": true, "explanation": "The Usage Policy explicitly prohibits generating false or misleading medical claims, and inventing clinical trial data to support a cure claim is a direct violation of this rule."}, {"letter": "D", "text": "No, it is not appropriate purely because this content is meant to manipulate search-engine ranking results.", "correct": false, "explanation": "While manipulating search-engine rankings is a secondary issue, the primary reason this use is inappropriate is the fabrication of false medical claims and clinical evidence, which is prohibited by the Usage Policy."}], "correct": "C", "select": 1}, {"id": "a6-047", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "A team lead worries that linking Google Drive will let Claude read everything the company keeps in Drive.", "question": "What is the accurate and policy-appropriate response?", "options": [{"letter": "A", "text": "Claude works within the permissions the person already holds, and administrators can restrict it further.", "correct": true, "explanation": "Access follows the user's existing permissions in the source system, and Team and Enterprise administrators can narrow what connections may do; narrowing never grants more than the source system allows."}, {"letter": "B", "text": "Claude reaches only files created after the connection was made, so older company material is never included.", "correct": false, "explanation": "There is no rule that limits a connection to newly created files."}, {"letter": "C", "text": "Claude gains access to the whole of Drive once linked, so the connection has to be re-approved every week.", "correct": false, "explanation": "Connecting does not grant access beyond what the person can already open, and weekly re-approval is not how the control works."}, {"letter": "D", "text": "Claude's access is decided entirely by the project it is used in, independently of the person's own account.", "correct": false, "explanation": "The project does not override the account's permissions in the connected service."}], "correct": "A", "select": 1}, {"id": "a6-048", "domain": 6, "task_id": "6.2", "objective": "Apply data sensitivity, regulatory, and privacy considerations", "situation": "A market research manager has been asked to summarise survey free-text responses that include respondents' names and employers, using the company's Claude for Work plan.", "question": "In which order should these steps be taken?", "options": [{"letter": "A", "text": "Remove the names and employers that the summary does not need; check what the participant consent and company policy allow; run the summary and review it before it is circulated; note what was removed so the result can be interpreted properly.", "correct": false, "explanation": "Data is edited before anyone has established what participants agreed to, so the amount of personal detail removed is guesswork and may be either too little or more than the analysis can afford."}, {"letter": "B", "text": "Run the summary and review it before it is circulated; check what the participant consent and company policy allow; remove the names and employers that the summary does not need; note what was removed so the result can be interpreted properly.", "correct": false, "explanation": "The personal data is processed and the output reviewed before either the consent position or the minimisation step, which is the sequence that causes the exposure rather than preventing it."}, {"letter": "C", "text": "Check what the participant consent and company policy allow; remove the names and employers that the summary does not need; run the summary and review it before it is circulated; note what was removed so the result can be interpreted properly.", "correct": true, "explanation": "Consent and policy set the boundary for what may be used at all; that boundary determines which identifiers can be stripped before anything is uploaded; the summary is then produced and reviewed within those limits; and recording what was removed lets colleagues read the findings knowing what the data no longer contains."}, {"letter": "D", "text": "Check what the participant consent and company policy allow; run the summary and review it before it is circulated; remove the names and employers that the summary does not need; note what was removed so the result can be interpreted properly.", "correct": false, "explanation": "The consent check happens first, but the full identifiable transcripts are still put into the tool before the names and employers are removed, so the minimisation step comes too late to protect anyone."}], "correct": "C", "select": 1}, {"id": "a6-049", "domain": 6, "task_id": "6.2", "objective": "Apply data sensitivity, regulatory, and privacy considerations", "situation": "An Enterprise owner decides to shorten the organisation's data retention period substantially.", "question": "What is the greatest concern before saving the change?", "options": [{"letter": "A", "text": "Shortening retention disables project sharing, because shared projects require indefinite retention.", "correct": false, "explanation": "Retention settings and project sharing are unrelated."}, {"letter": "B", "text": "Data falling outside the new window is deleted immediately, so work still in progress can be lost.", "correct": true, "explanation": "Shortening retention deletes anything already older than the new window as soon as it is saved."}, {"letter": "C", "text": "The change applies only to new chats, so nothing older is affected and the setting has no real effect.", "correct": false, "explanation": "The setting is not limited to new conversations; existing data is in scope."}, {"letter": "D", "text": "Every connector must be re-authenticated by each member before the shorter window can take effect.", "correct": false, "explanation": "Retention changes do not require connectors to be reconnected."}], "correct": "B", "select": 1}, {"id": "a6-050", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "A fintech company built a product that uses Claude to automatically approve or deny small business loan applications without human involvement.", "question": "Under Anthropic's Usage Policy, what must the company do to comply with the high-risk use case requirements for financial services?", "options": [{"letter": "A", "text": "The company should route every denial through a separate rules engine so only approvals originate from Claude, removing any human review requirement.", "correct": false, "explanation": "Routing denials away from Claude does not eliminate the requirement for disclosure and human review. High-risk financial decisions still demand human oversight regardless of which system generates the outcome."}, {"letter": "B", "text": "The company may skip disclosure to applicants as long as it retains complete logs of every underwriting input and output that Claude produces for compliance audit.", "correct": false, "explanation": "Retaining logs does not exempt the company from disclosing AI involvement or providing human review. The Usage Policy requires both disclosure and human review for high-risk financial decisions."}, {"letter": "C", "text": "The company should reduce how long loan application records are stored so AI-generated decisions cannot later be inspected during compliance audits.", "correct": false, "explanation": "Reducing how long records are kept impairs the ability to conduct compliance audits. Moreover, it does not fulfill the Usage Policy's requirements for disclosure and human review of high-risk decisions."}, {"letter": "D", "text": "The company must disclose Claude's involvement to applicants and have a qualified human review the model's output before any adverse lending decision is finalized.", "correct": true, "explanation": "The Usage Policy classifies automated lending decisions as high-risk. Therefore, the company must disclose Claude's involvement to applicants and ensure a qualified human reviews the output before any adverse lending decision is finalized."}], "correct": "D", "select": 1}, {"id": "a6-051", "domain": 6, "task_id": "6.2", "objective": "Apply data sensitivity, regulatory, and privacy considerations", "situation": "A Team-plan owner is uneasy about conversation content leaving the organisation when members rate Claude's answers.", "question": "What control is available to them?", "options": [{"letter": "A", "text": "Owners can set ratings to be stored only inside the organisation's own private workspace.", "correct": false, "explanation": "Ratings are feedback to Anthropic; they are not retained locally instead."}, {"letter": "B", "text": "Owners can have conversations redacted automatically before any rating is submitted.", "correct": false, "explanation": "No automatic redaction step is applied to submitted feedback."}, {"letter": "C", "text": "Owners can switch off the chat rating feature for the whole organisation in its settings.", "correct": true, "explanation": "Team and Enterprise owners can disable rating so that conversations are not submitted with feedback."}, {"letter": "D", "text": "Owners can require that every rating is approved by a manager before it is finally sent.", "correct": false, "explanation": "There is no approval workflow layered on top of ratings."}], "correct": "C", "select": 1}, {"id": "a6-052", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "An analyst wants to upload a partner company's forecast data, shared under a non-disclosure agreement, into a Claude project.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Rename the partner and its products throughout the file, then upload it as ordinary internal material.", "correct": false, "explanation": "Relabelling does not change who owns the data or what the agreement permits."}, {"letter": "B", "text": "Upload only part of the data at a time, so that no single conversation ever contains the full forecast.", "correct": false, "explanation": "Splitting the file spreads the same material across several conversations without addressing the contractual question."}, {"letter": "C", "text": "Upload it and delete the project as soon as the analysis is finished, so the data is not held for long.", "correct": false, "explanation": "Deleting afterwards does not cure a disclosure that the agreement may not have permitted in the first place."}, {"letter": "D", "text": "Check whether the agreement allows processing in third-party tools, and seek approval if unclear.", "correct": true, "explanation": "Third-party confidential material is governed by the contract that supplied it, so the agreement's terms, and an approval where they are ambiguous, come before any upload."}], "correct": "D", "select": 1}, {"id": "a6-053", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "A consultant is about to send a client a report she drafted with Claude. Her firm's AI policy sets out several requirements for client-facing work produced with AI, and she is unsure which of them apply to a report of this kind.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Ask the client whether they have any objection to AI assistance having been used on the report.", "correct": false, "explanation": "The client's comfort does not replace the firm's own requirements. Asking them cannot tell her what her firm obliges her to do, and their agreement would not excuse skipping it."}, {"letter": "B", "text": "Add a short line stating the report was drafted with AI assistance, and send it to the client.", "correct": false, "explanation": "This guesses at one requirement and treats it as the whole of the policy. Disclosure may be needed, but so may review, sign-off or restrictions on the client data involved."}, {"letter": "C", "text": "Ask her engagement manager to sign the report off, so responsibility for it rests with him instead.", "correct": false, "explanation": "A sign-off may well be one of the requirements, but choosing it because it shifts responsibility is not following the policy. She still does not know what the policy asks for."}, {"letter": "D", "text": "Establish what the policy requires for a deliverable of this kind, before the report goes out.", "correct": true, "explanation": "The firm has a written standard covering exactly this situation, so the first step is to read it and identify which requirements apply to this deliverable. Everything else she might do is a guess at what the policy says."}], "correct": "D", "select": 1}, {"id": "a6-054", "domain": 6, "task_id": "6.1", "objective": "Identify appropriate and inappropriate use cases", "situation": "A retail brand deploys a Claude-powered chatbot on its website. Internal testing shows many users believe they are chatting with a live human agent, and the company decides not to add any disclosure.", "question": "Is this an appropriate use of Claude?", "options": [{"letter": "A", "text": "Yes, because a personable conversational tone alone does not require any disclosure under the Usage Policy", "correct": false, "explanation": "Even a personable tone does not remove the requirement to disclose that a chatbot is AI, especially once testing shows users are being misled about who they are talking to."}, {"letter": "B", "text": "No, but only because the brand did not obtain explicit opt-in consent before the chat began", "correct": false, "explanation": "The requirement here is disclosure that the user is talking to AI, not opt-in consent before the conversation starts."}, {"letter": "C", "text": "Yes, as long as the site's terms-of-service page mentions AI use somewhere outside the chat window", "correct": false, "explanation": "A disclosure buried in an unrelated terms-of-service page does not satisfy the requirement to inform users within the interaction itself that they are talking to AI."}, {"letter": "D", "text": "No, because the Usage Policy requires consumer-facing chatbots to inform users that they are interacting with AI", "correct": true, "explanation": "The Usage Policy requires consumer-facing chatbots to clearly inform users they are interacting with AI, and this scenario intentionally omits that disclosure."}], "correct": "D", "select": 1}, {"id": "a6-055", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "A manager tells his team to 'use common sense' with Claude rather than writing anything down.", "question": "What is the most useful thing to recommend he add?", "options": [{"letter": "A", "text": "A requirement that every Claude conversation be saved and forwarded to him so he has a full record.", "correct": false, "explanation": "Collecting every conversation creates a large store of sensitive material without telling anyone what is allowed."}, {"letter": "B", "text": "A monthly report of how many messages each person sent, to show that Claude is being used enough.", "correct": false, "explanation": "Volume statistics measure activity, not whether the activity complied with anything."}, {"letter": "C", "text": "A rule that only the two most experienced people on the team may use Claude for any work purpose.", "correct": false, "explanation": "Restricting use to two people avoids the question rather than answering it, and does not guide their judgment either."}, {"letter": "D", "text": "A short written standard naming approved plans, permitted data, disclosure rules and who to ask.", "correct": true, "explanation": "A short, concrete standard is what makes a policy usable day to day: what plan, what data, what must be disclosed, and who decides the ambiguous cases."}], "correct": "D", "select": 1}, {"id": "a6-056", "domain": 6, "task_id": "6.4", "objective": "Understand the ethical implications of AI usage", "situation": "A manager plans to use Claude to help produce the written justifications that accompany her team's annual performance ratings.", "question": "What is the most appropriate way to do this?", "options": [{"letter": "A", "text": "Ask Claude to propose ratings from the evidence and change only the ones that look wrong to her.", "correct": false, "explanation": "Reviewing proposed ratings anchors her to Claude's judgment on decisions she must own herself."}, {"letter": "B", "text": "Decide each rating herself from the evidence, then use Claude to help word the reasoning behind it.", "correct": true, "explanation": "The judgment about a person's performance is hers to make and defend; Claude can help express a conclusion she has already reached and can justify herself."}, {"letter": "C", "text": "Let Claude set the ratings so that every member of the team is judged against one neutral standard.", "correct": false, "explanation": "Consistency does not make the output fair, and it hands a decision about people to a tool."}, {"letter": "D", "text": "Have Claude write the justifications, and tell each person afterwards that the wording was assisted.", "correct": false, "explanation": "Disclosing afterwards is welcome, but it does not put a person behind the judgment being made."}], "correct": "B", "select": 1}, {"id": "a6-057", "domain": 6, "task_id": "6.4", "objective": "Understand the ethical implications of AI usage", "situation": "A department head now has Claude draft her monthly operations bulletin, glances over it, and sends it to the division under her own name. Staff read it as her own assessment. Which two concerns are most significant?", "question": "Select two.", "options": [{"letter": "A", "text": "The writing will read noticeably differently from the bulletins that she used to circulate previously.", "correct": false, "explanation": "A change of voice may be noticed, but style is a cosmetic issue next to misplaced trust and unchecked claims."}, {"letter": "B", "text": "Readers are weighing the bulletin as her own considered assessment when she has not actually made it.", "correct": true, "explanation": "Staff rely on her judgment as the reason to trust the bulletin, and that reliance is now misplaced."}, {"letter": "C", "text": "Unverified claims can pass straight into business decisions because nobody has checked them at all.", "correct": true, "explanation": "for a separate reason: beyond attribution, a skim is not verification, so errors reach decisions unchallenged."}, {"letter": "D", "text": "Drafting an operations bulletin is not a task that any language model is capable of doing at all usefully.", "correct": false, "explanation": "Drafting a bulletin is a reasonable use; the problem is the absent review and the silent attribution, not the capability."}, {"letter": "E", "text": "Producing the bulletin in this way consumes rather more of her weekly allowance than writing it herself.", "correct": false, "explanation": "Allowance consumption is a cost consideration and has nothing to do with the ethical problem described."}], "correct": ["B", "C"], "select": 2}, {"id": "a6-058", "domain": 6, "task_id": "6.1", "objective": "Identify appropriate and inappropriate use cases", "situation": "An e-commerce operator wants Claude to auto-generate customer support replies containing fabricated tracking numbers and fake refund confirmation emails to delay legitimate refund requests.", "question": "Is this an appropriate use of Claude?", "options": [{"letter": "A", "text": "Yes, because automating customer support, including generating tracking and confirmation messages, is a common and encouraged use case.", "correct": false, "explanation": "Automating customer support is a common and encouraged use case, but this scenario involves generating fabricated tracking numbers and fake refund confirmations to deliberately deceive customers. The Usage Policy prohibits such deceptive content regardless of the general use case."}, {"letter": "B", "text": "Yes, as long as a human support agent reviews each fabricated tracking number and confirmation email to verify accuracy before sending.", "correct": false, "explanation": "Having a human review fabricated information does not make it legitimate; the content remains deceptive by design. The Usage Policy prohibits fraudulent content irrespective of any review process."}, {"letter": "C", "text": "No, because the Usage Policy requires outgoing emails to use approved services, and using an unapproved vendor violates the policy.", "correct": false, "explanation": "The violation stems from the fabricated content intended to deceive customers, not from using an unapproved email service. The Usage Policy focuses on the deceptive nature of the communication, not the delivery method."}, {"letter": "D", "text": "No because the Usage Policy bans generating fraudulent documents such as fake confirmations to deceive customers.", "correct": true, "explanation": "The Usage Policy explicitly bans generating fraudulent documents, including fake confirmations and deceptive communications intended to mislead customers. This scenario directly violates that prohibition."}], "correct": "D", "select": 1}, {"id": "a6-059", "domain": 6, "task_id": "6.4", "objective": "Understand the ethical implications of AI usage", "situation": "A social media analytics vendor plans to use Claude to build detailed psychological and political profiles of individual private citizens by aggregating their public posts, in order to sell these profiles to third parties for targeted persuasion campaigns.", "question": "Which concern from the Usage Policy applies most directly here?", "options": [{"letter": "A", "text": "No concern applies, since the underlying posts are already publicly available on the social platform", "correct": false, "explanation": "Public availability of source posts does not exempt the aggregation and profiling activity from policy restrictions, especially when used for targeted persuasion."}, {"letter": "B", "text": "This restriction applies only when every profiled individual happens to be a minor", "correct": false, "explanation": "The restriction is not limited to minors; it applies broadly to profiling and tracking of individuals for manipulation purposes."}, {"letter": "C", "text": "The concern is limited to biometric data and does not extend to behavioral profiling at all", "correct": false, "explanation": "Profiling restrictions extend beyond biometric data to behavioral, psychological, and political profiling used for tracking or targeting."}, {"letter": "D", "text": "It falls under the ban on profiling and tracking individuals to enable manipulation or targeting", "correct": true, "explanation": "Building individualized profiles for targeting and manipulation purposes falls under the Usage Policy's restrictions on profiling and tracking, regardless of the source data being publicly posted."}], "correct": "D", "select": 1}, {"id": "a6-060", "domain": 6, "task_id": "6.4", "objective": "Understand the ethical implications of AI usage", "situation": "A telehealth startup integrates Claude to help draft preliminary summaries of patient symptoms for physicians to review before diagnosis. During a product review, a compliance officer asks why this counts as a high-risk use case even though a licensed physician always makes the final diagnosis.", "question": "What is the best answer?", "options": [{"letter": "A", "text": "It is high-risk only if the summaries are shared with insurers for reimbursement purposes", "correct": false, "explanation": "Insurance sharing is a separate concern; the high-risk designation for healthcare applies regardless of downstream billing use."}, {"letter": "B", "text": "It is high-risk solely because telehealth companies typically operate across state lines", "correct": false, "explanation": "Cross-state operation is not the basis for the high-risk designation; the healthcare subject matter itself is."}, {"letter": "C", "text": "Healthcare's public welfare implications mean the safeguards apply to the summary itself, too", "correct": true, "explanation": "Healthcare is one of the categories explicitly designated high-risk under the Usage Policy due to public welfare implications, so review and disclosure safeguards apply to AI-assisted content that feeds into care decisions."}, {"letter": "D", "text": "It does not count as high-risk, since only the physician's final diagnosis is subject to the policy", "correct": false, "explanation": "The safeguards apply to AI-assisted content used in the healthcare workflow, not only to the final decision point."}], "correct": "C", "select": 1}, {"id": "a6-061", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "A company systematically queries Claude through the API to collect large volumes of model outputs, then uses that data to train a rival large language model that it plans to sell.", "question": "This activity is:", "options": [{"letter": "A", "text": "Prohibited under the Usage Policy's restrictions on using Claude's outputs to develop competing models.", "correct": true, "explanation": "The Usage Policy prohibits model scraping and using Claude's outputs to develop competing models."}, {"letter": "B", "text": "Prohibited only if the company also republishes Anthropic's original training data alongside the generated outputs.", "correct": false, "explanation": "The prohibition applies to the scraping and competing-model training itself, not only to republishing Anthropic's original training data."}, {"letter": "C", "text": "Permitted as long as the resulting competing model is not marketed as being built using Claude's technology.", "correct": false, "explanation": "Omitting attribution does not make the underlying prohibited activity, training a competitor on scraped outputs, acceptable."}, {"letter": "D", "text": "Permitted because outputs generated through the API belong entirely to the company that generated them without restriction.", "correct": false, "explanation": "Ownership of individual outputs does not exempt systematic scraping used to train a competing model from Usage Policy restrictions."}], "correct": "A", "select": 1}, {"id": "a6-062", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "A staff member realises she has pasted regulated customer records into her personal Claude account rather than the company's managed workspace.", "question": "Which sequence is correct?", "options": [{"letter": "A", "text": "Redo the work in the managed workspace; delete the affected conversations as the response team directs; stop using the personal account for this work; report the incident through the escalation path the policy names.", "correct": false, "explanation": "Getting the work done comes ahead of every governance step, and the incident is reported last of all. By then the conversations have already been deleted without direction and the exposure has gone unreported."}, {"letter": "B", "text": "Delete the affected conversations as the response team directs; stop using the personal account for this work; report the incident through the escalation path the policy names; redo the work in the managed workspace.", "correct": false, "explanation": "Deletion is placed before the incident has been reported, so there is nobody yet to direct it. Destroying the record ahead of the report also removes the evidence the response team needs to scope what was exposed."}, {"letter": "C", "text": "Stop using the personal account for this work; report the incident through the escalation path the policy names; delete the affected conversations as the response team directs; redo the work in the managed workspace.", "correct": true, "explanation": "Containment comes first, so no further regulated data is exposed; the incident is then reported through the path the policy names, so the response team can assess it; remediation such as deletion follows their direction rather than preceding it; and the work resumes in the approved workspace."}, {"letter": "D", "text": "Report the incident through the escalation path the policy names; redo the work in the managed workspace; delete the affected conversations as the response team directs; stop using the personal account for this work.", "correct": false, "explanation": "Reporting comes first here, but the exposure is left running: the work is redone and conversations deleted while the personal account is still in use for regulated material, and it is stopped only at the very end."}], "correct": "C", "select": 1}, {"id": "a6-063", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "A project lead is finishing a client status pack that Claude helped draft. His firm's AI policy requires AI assistance to be disclosed on external documents, and the client contract says nothing about the subject.", "question": "What should be his first step?", "options": [{"letter": "A", "text": "Leave the disclosure off, because the contract is silent and the client has not asked for it.", "correct": false, "explanation": "Treating silence in the contract as permission inverts the relationship between the two. The internal policy binds the firm's staff regardless of what a client has thought to ask about."}, {"letter": "B", "text": "Ask the client whether they would rather the pack carried no disclosure line on it at all.", "correct": false, "explanation": "Inviting the client to waive a rule the firm set for itself puts the obligation in the wrong hands, and it would be awkward to act on a preference that conflicts with the policy."}, {"letter": "C", "text": "Add the disclosure the firm's policy requires, since the contract does not override it.", "correct": true, "explanation": "The policy applies to external documents and the contract's silence does not create an exception to it. Disclosing is the required action, and it is also the honest one for a pack the client will rely on."}, {"letter": "D", "text": "Rewrite enough of Claude's wording that he can argue the disclosure requirement no longer applies.", "correct": false, "explanation": "Editing the text in order to escape the disclosure defeats the purpose of the rule and is a deliberate attempt to reach a technicality rather than a good-faith reading of the policy."}], "correct": "C", "select": 1}, {"id": "a6-064", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "A support lead wants to paste a year of customer complaint emails, which contain names and postal addresses, into Claude so it can find common themes. Her company's AI policy sets rules for handling customer personal data.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Use an incognito chat for the work, so the conversation cannot be used to improve Claude.", "correct": false, "explanation": "Incognito chats are not used to improve Claude, but that addresses only training. The policy's rules on where and how customer personal data may be processed still apply in full."}, {"letter": "B", "text": "Take the names and addresses out of the emails herself, and paste what remains of the text.", "correct": false, "explanation": "Removing identifiers by hand may be exactly what the policy asks for, or may fall short of it. Deciding the treatment herself before reading the rules is what the policy exists to prevent."}, {"letter": "C", "text": "Find out how the policy classifies this data and what it requires, before any of it is pasted.", "correct": true, "explanation": "The data is customer personal data and the company has rules for it, so the first step is to establish which category it falls into and what that category requires. That determines whether pasting is permitted at all, and under what conditions."}, {"letter": "D", "text": "Do the work inside the company's managed workspace, since company data is approved for use there.", "correct": false, "explanation": "The managed workspace is very likely part of the answer, but a general approval for company data does not settle what is allowed for personal data specifically. That is what the policy has to be checked for."}], "correct": "C", "select": 1}, {"id": "a6-065", "domain": 6, "task_id": "6.1", "objective": "Identify appropriate and inappropriate use cases", "situation": "An HR software vendor uses Claude to automatically reject job applicants based solely on resume-screening scores. No recruiter reviews borderline or rejected candidates, and applicants are not told AI was used.", "question": "Is this an appropriate use of Claude?", "options": [{"letter": "A", "text": "Yes, provided the vendor can show the model's screening accuracy exceeds that of human recruiters", "correct": false, "explanation": "Demonstrated accuracy does not replace the required human review and disclosure; the safeguards apply regardless of measured model performance."}, {"letter": "B", "text": "No, but only because rejected applicants are not offered any formal appeals process afterward", "correct": false, "explanation": "The absence of an appeals process is a separate issue from the missing professional review and AI-use disclosure required by the Usage Policy."}, {"letter": "C", "text": "Yes, because resume screening is an administrative task outside the scope of high-risk use cases", "correct": false, "explanation": "Employment decisions such as hiring rejections are treated as a high-risk domain, not an administrative task exempt from review-and-disclosure safeguards."}, {"letter": "D", "text": "No, because the Usage Policy requires human review and disclosure for high-risk employment decisions", "correct": true, "explanation": "High-risk employment use cases require a qualified human to review the AI's decisions and require disclosure to affected individuals, and this scenario has neither safeguard in place."}], "correct": "D", "select": 1}, {"id": "a6-066", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "A hospital deploys a Claude-powered chatbot on its patient portal to answer questions about medication side effects.", "question": "According to Anthropic's Usage Policy for high-risk, consumer-facing use cases, what must the hospital do before patients rely on the chatbot's answers?", "options": [{"letter": "A", "text": "The hospital should configure the chatbot to always redirect patients to speak with their pharmacist by phone instead of answering any medication-related questions.", "correct": false, "explanation": "Redirecting every question avoids the disclosure obligation rather than fulfilling it, and is not what the policy requires."}, {"letter": "B", "text": "The hospital should limit chatbot responses to verbatim excerpts from publicly available FDA drug labeling to ensure no original content is ever generated.", "correct": false, "explanation": "Restricting content sourcing does not address the requirement to disclose that the interaction is AI-driven."}, {"letter": "C", "text": "The hospital may require patients to sign a liability waiver acknowledging the chatbot's limitations before it provides any medication-related answers.", "correct": false, "explanation": "A liability waiver does not satisfy the Usage Policy's transparency requirement, which is about disclosing AI involvement, not shifting legal liability."}, {"letter": "D", "text": "The hospital must clearly disclose to patients that they are interacting with an AI system before they receive medical guidance from the chatbot.", "correct": true, "explanation": "Consumer-facing chatbots in high-risk domains like healthcare must transparently disclose that the user is interacting with an AI system before providing guidance."}], "correct": "D", "select": 1}, {"id": "a6-067", "domain": 6, "task_id": "6.1", "objective": "Identify appropriate and inappropriate use cases", "situation": "A small law firm uses Claude to draft first-pass responses to routine client contract questions. A licensed attorney reviews and finalizes every draft before it is sent, and clients are told that AI assisted in drafting.", "question": "Is this consistent with Anthropic's guidance for high-risk use cases?", "options": [{"letter": "A", "text": "No, because attorney review does not remove all liability for AI-drafted content in legal contexts", "correct": false, "explanation": "This statement is about legal liability, not about compliance with the Usage Policy's stated safeguards, which the firm has already met."}, {"letter": "B", "text": "No, because the Usage Policy prohibits legal-advice use cases entirely regardless of professional review", "correct": false, "explanation": "Legal advice is a high-risk use case that is permitted with safeguards, not a universally prohibited category like weapons development or CSAM."}, {"letter": "C", "text": "Yes, but only if the firm also purchases a separate high-risk-use license directly from Anthropic", "correct": false, "explanation": "The Usage Policy does not require a separate high-risk-use license; it requires professional review and disclosure, both of which are already present."}, {"letter": "D", "text": "Yes, because a qualified professional reviews the output and the AI's involvement is disclosed to clients", "correct": true, "explanation": "High-risk domains like legal advice require both qualified professional review of the content and disclosure that AI was used, and this scenario satisfies both safeguards."}], "correct": "D", "select": 1}, {"id": "a6-068", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "A finance team wants Claude to take actions in a connected system, such as updating records, rather than only reading from it.", "question": "Under a typical organisational AI policy, what matters most before this is enabled?", "options": [{"letter": "A", "text": "Confirmation that the connected system's data is small enough to fit inside a single conversation.", "correct": false, "explanation": "Data volume is a workload question and says nothing about the risk of unreviewed changes."}, {"letter": "B", "text": "Approval for write access, with a human checkpoint before any change is committed to the system.", "correct": true, "explanation": "Write access changes the system of record, so it needs explicit approval and a person who confirms each change before it takes effect."}, {"letter": "C", "text": "Agreement that only one nominated person may open conversations that use the connected system.", "correct": false, "explanation": "Limiting who may use it reduces exposure slightly but still leaves unreviewed writes possible."}, {"letter": "D", "text": "Assurance that Claude will describe each change in the chat so the record can be located later.", "correct": false, "explanation": "A description in the chat is a weak record and does not prevent an incorrect change from being made."}], "correct": "B", "select": 1}, {"id": "a6-069", "domain": 6, "task_id": "6.1", "objective": "Identify appropriate and inappropriate use cases", "situation": "A software team uses Claude to review pull requests, suggest bug fixes, and generate unit tests as part of its normal development workflow, with engineers reviewing all suggested code before merging.", "question": "Is this an appropriate use of Claude?", "options": [{"letter": "A", "text": "Yes, because code review and test generation are common development tasks with engineer sign-off", "correct": true, "explanation": "Code review, bug-fix suggestions, and test generation are standard, supported development use cases, and human engineers retaining final approval fits normal responsible practice."}, {"letter": "B", "text": "No, because the Usage Policy restricts AI-assisted code review to Anthropic-certified engineering teams", "correct": false, "explanation": "There is no requirement that engineering teams be specially certified by Anthropic to use Claude for code review or development assistance."}, {"letter": "C", "text": "No, because generated unit tests are excluded from the Usage Policy's permitted development activities", "correct": false, "explanation": "Generating unit tests is a routine and supported coding use case, not an excluded activity."}, {"letter": "D", "text": "Yes, but only if the team disables Claude for any file that touches production credentials", "correct": false, "explanation": "There is no blanket requirement to disable Claude on credential-related files for this use case to be appropriate; the scenario as described is already acceptable."}], "correct": "A", "select": 1}, {"id": "a6-070", "domain": 6, "task_id": "6.4", "objective": "Understand the ethical implications of AI usage", "situation": "A journalist uses Claude to help draft a news article, and later a colleague asks whether this requires the same human-in-the-loop and disclosure safeguards as other high-risk categories like healthcare or finance.", "question": "What is the correct answer?", "options": [{"letter": "A", "text": "Yes, journalism is a listed high-risk category requiring review and AI disclosure too", "correct": true, "explanation": "Journalism is one of the categories the Usage Policy designates as high-risk, requiring a qualified professional to review AI-assisted content and disclosure of AI involvement before publication."}, {"letter": "B", "text": "Yes, but only when the article specifically discusses Anthropic or its own products", "correct": false, "explanation": "The requirement applies to journalism generally, not only to articles about Anthropic or its products."}, {"letter": "C", "text": "No, because the review requirement applies only to direct financial transactions", "correct": false, "explanation": "The human-in-the-loop requirement is not limited to financial transactions; it spans several public-welfare categories including journalism."}, {"letter": "D", "text": "No, because journalism is a creative field fully exempt from the high-risk provisions", "correct": false, "explanation": "Journalism is not exempt; it is specifically named alongside legal, healthcare, and financial use cases as requiring safeguards."}], "correct": "A", "select": 1}, {"id": "a6-071", "domain": 6, "task_id": "6.2", "objective": "Apply data sensitivity, regulatory, and privacy considerations", "situation": "A European team wants to use Claude to analyse a customer database that includes personal data about EU residents.", "question": "What is the most appropriate action?", "options": [{"letter": "A", "text": "Involve the people responsible for data protection to confirm the lawful basis and required safeguards.", "correct": true, "explanation": "Regulated personal data requires a decision from the accountable data-protection function, not an individual."}, {"letter": "B", "text": "Proceed, because analysing data the company already holds does not raise any new privacy questions.", "correct": false, "explanation": "Introducing a new processing tool is itself a change that has to be assessed."}, {"letter": "C", "text": "Move the analysis onto a personal account so that it falls outside the organisation's compliance scope.", "correct": false, "explanation": "Using a personal account does not remove the obligation and removes the oversight as well."}, {"letter": "D", "text": "Ask Claude to summarise the relevant regulation and treat that summary as the compliance decision.", "correct": false, "explanation": "A summary of a regulation is not a compliance decision and carries no accountability."}], "correct": "A", "select": 1}, {"id": "a6-072", "domain": 6, "task_id": "6.2", "objective": "Apply data sensitivity, regulatory, and privacy considerations", "situation": "An employee's organisation has no written policy on AI tools, and they want to use Claude with supplier contracts.", "question": "What should they do first?", "options": [{"letter": "A", "text": "Write their own policy for the team and begin using the tool under the rules that they drafted.", "correct": false, "explanation": "An individual cannot grant the organisation's permission to themselves."}, {"letter": "B", "text": "Assume it is permitted, since the absence of a written policy means that nothing has been prohibited yet.", "correct": false, "explanation": "Silence in a policy is not permission for third-party contractual material."}, {"letter": "C", "text": "Ask their manager or legal contact for a decision before putting any supplier material into the tool.", "correct": true, "explanation": "Where no policy exists, the accountable route is to obtain a decision rather than to assume one."}, {"letter": "D", "text": "Use the tool but avoid saving anything, so that no record of the contracts is kept anywhere at all.", "correct": false, "explanation": "Not saving output does not change whether the contracts could be entered in the first place."}], "correct": "C", "select": 1}, {"id": "a6-073", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "A team lead is writing a short local practice note to sit underneath the company-wide AI standard. Which two items genuinely belong in it?", "question": "Select two.", "options": [{"letter": "A", "text": "Who to contact when a situation arises that the company-wide AI standard does not clearly cover.", "correct": true, "explanation": "for a different reason: a named escalation contact is what stops people improvising when the standard runs out."}, {"letter": "B", "text": "Which categories of team data must never be pasted into Claude under any circumstances at all.", "correct": true, "explanation": "A local note is where the general standard becomes concrete about the specific data this team actually handles."}, {"letter": "C", "text": "The exact number of days for which the vendor keeps stored conversations before removal.", "correct": false, "explanation": "Retention periods are vendor facts that change and are documented by the vendor; copying a number into a team note invites it to go stale."}, {"letter": "D", "text": "A ranked list telling each team member which model tier they should prefer for all of their own work.", "correct": false, "explanation": "Model choice is a task-fit judgment made per piece of work, not a governance rule to assign per person."}, {"letter": "E", "text": "A target for the number of working hours that every team member should save each week using Claude.", "correct": false, "explanation": "A savings target pressures people towards using Claude and belongs to performance management, not to a governance note."}], "correct": ["A", "B"], "select": 2}, {"id": "a6-074", "domain": 6, "task_id": "6.4", "objective": "Understand the ethical implications of AI usage", "situation": "A council service team plans to use a Claude-produced priority list to decide which residents receive home-adaptation grants first. The list was built from the council's historical case records.", "question": "What should the team do first?", "options": [{"letter": "A", "text": "Publish the priority list so that residents can see how the ordering was arrived at.", "correct": false, "explanation": "Transparency about the ordering is worth having, but publishing a list that may embed historical inequity makes it visible rather than fair, and residents are affected either way."}, {"letter": "B", "text": "Ask Claude to state whether its own priority list contains any unfair bias in it.", "correct": false, "explanation": "Claude cannot audit the representativeness of records it was given, and a reassuring self-assessment would give the team false confidence about exactly the risk they need to investigate."}, {"letter": "C", "text": "Use the list for this year and then compare the outcomes against last year's decisions.", "correct": false, "explanation": "Using the list and reviewing afterwards means a year of residents receive grants in an order that may be skewed, and the comparison arrives long after the decisions have been made."}, {"letter": "D", "text": "Examine whether the historical records under-represent groups who applied less often.", "correct": true, "explanation": "A priority list derived from past cases inherits whatever unevenness is in those cases, and residents who historically applied less often will be under-served by it. Because the output drives a decision about people, checking the data for that skew comes before the list is used at all."}], "correct": "D", "select": 1}, {"id": "a6-075", "domain": 6, "task_id": "6.1", "objective": "Identify appropriate and inappropriate use cases", "situation": "A city housing department lists tasks it might hand to Claude. Which two are appropriate to delegate with routine human review, rather than needing a person to own each outcome?", "question": "Select two.", "options": [{"letter": "A", "text": "Grouping a year of tenant survey comments into recurring themes for the quarterly service report.", "correct": true, "explanation": "Theme-finding across free-text feedback informs a report rather than determining anything about an individual, and the themes can be checked against the source comments."}, {"letter": "B", "text": "Setting the priority score that determines the order in which families are offered available homes.", "correct": false, "explanation": "A priority score decides who is housed first; that is a rights-affecting judgment Claude may inform but must not make."}, {"letter": "C", "text": "Rewriting the published tenancy policy into plain language for the department's public web pages.", "correct": true, "explanation": "Rewriting already-published policy is drafting work that a person reviews before it goes live, so any error is caught and reversible."}, {"letter": "D", "text": "Issuing the final written refusal to tenants whose appeal against rehousing has failed.", "correct": false, "explanation": "A final refusal of an appeal is an authoritative outcome for a named person and must be owned and issued by a human decision-maker."}, {"letter": "E", "text": "Deciding which households are removed from the housing waiting list for failing to respond in time.", "correct": false, "explanation": "Removing a household from a waiting list affects a person's access to housing and needs a named person answerable for the decision."}], "correct": ["A", "C"], "select": 2}, {"id": "a6-076", "domain": 6, "task_id": "6.1", "objective": "Identify appropriate and inappropriate use cases", "situation": "A city government wants to use Claude to build a system that cross-references live facial-recognition camera feeds with social media posts to covertly track individual protest attendees, without a warrant or their consent.", "question": "Is this an appropriate use of Claude?", "options": [{"letter": "A", "text": "No, because the Usage Policy bans covert individual tracking through facial recognition without consent", "correct": true, "explanation": "The Usage Policy prohibits using Claude to build systems that covertly track or identify specific individuals through facial recognition without their consent, which matches this scenario."}, {"letter": "B", "text": "Yes, because public safety agencies are exempt from the Usage Policy's surveillance provisions here", "correct": false, "explanation": "Government or public-safety status does not exempt an organization from the Usage Policy's restrictions on non-consensual individual tracking via facial recognition."}, {"letter": "C", "text": "Yes, because these cameras are located in public spaces where there is no expectation of privacy", "correct": false, "explanation": "Being in a public space does not authorize covert identification and tracking of specific individuals using facial recognition; the policy restriction still applies."}, {"letter": "D", "text": "No, but only because the city lacks a proper data-retention policy for the collected camera footage", "correct": false, "explanation": "A missing retention policy is not the core problem; the covert, non-consensual tracking of individuals is itself the prohibited activity."}], "correct": "A", "select": 1}, {"id": "a6-077", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "A grants officer wants to use a Claude feature that her university's written AI standard does not mention at all, on funding applications containing unpublished research plans. Which two actions follow good governance?", "question": "Select two.", "options": [{"letter": "A", "text": "Ask Claude to read the standard and confirm whether or not the university permits this particular feature.", "correct": false, "explanation": "Claude can summarise a document but cannot grant an approval, and its reading of the standard carries no authority."}, {"letter": "B", "text": "Raise it with whoever owns the AI standard and get a decision before using it on the applications.", "correct": true, "explanation": "A gap in the standard is a question for the policy owner, and the escalation happens before the confidential work begins, not after."}, {"letter": "C", "text": "Try the capability first on material that is public, so nothing confidential is involved in the trial.", "correct": true, "explanation": "for an independent reason: trying it on non-confidential material lets the officer learn what it does without putting the research plans at risk."}, {"letter": "D", "text": "Treat the standard's silence as permission, on the basis that nothing in it forbids using the capability.", "correct": false, "explanation": "Silence is not approval; policies cannot enumerate every capability, which is exactly why an escalation route exists."}, {"letter": "E", "text": "Go ahead now and record the use in the project file so that it can be reviewed by someone else afterwards.", "correct": false, "explanation": "Recording the use afterwards does not undo the disclosure of applicant data under an unapproved feature."}], "correct": ["B", "C"], "select": 2}, {"id": "a6-078", "domain": 6, "task_id": "6.2", "objective": "Apply data sensitivity, regulatory, and privacy considerations", "situation": "A team on a Team plan wants Claude to work with their shared Slack workspace.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Confirm the policy allows it; have the owner enable the connector; members authenticate individually; agree what may be shared.", "correct": true, "explanation": "Policy clearance comes first, the owner enables it for the organisation, then individuals connect their own access."}, {"letter": "B", "text": "Have the owner enable the connector; members authenticate individually; confirm the policy allows it; agree what may be shared.", "correct": false, "explanation": "The data source is live before anyone has checked whether policy permits it."}, {"letter": "C", "text": "Agree what may be shared; members authenticate individually; have the owner enable the connector; confirm the policy allows it.", "correct": false, "explanation": "Authentication precedes enablement, and the policy check is left until the end."}, {"letter": "D", "text": "Members authenticate individually; confirm the policy allows it; agree what may be shared; have the owner enable the connector.", "correct": false, "explanation": "Members cannot authenticate to a connector the organisation's owner has not enabled."}], "correct": "A", "select": 1}, {"id": "a6-079", "domain": 6, "task_id": "6.1", "objective": "Identify appropriate and inappropriate use cases", "situation": "A political consulting firm asks Claude to generate thousands of localized social media posts, each written to appear as though it came from a different concerned local resident, to make opposition to a ballot measure look larger than it actually is.", "question": "Is this an appropriate use of Claude?", "options": [{"letter": "A", "text": "No, because the Usage Policy bans fake grassroots campaigns that misrepresent the scale of public support", "correct": true, "explanation": "The Usage Policy prohibits using Claude to generate fake grassroots or astroturf campaigns that misrepresent how widespread or organic public support actually is."}, {"letter": "B", "text": "Yes, because political messaging about ballot measures is fully protected under free speech principles", "correct": false, "explanation": "The Usage Policy restricts election-related manipulation regardless of general speech protections; the issue is manufactured false consensus, not the topic itself."}, {"letter": "C", "text": "Yes, provided the firm discloses to the hosting platform that some of these posts were AI-generated", "correct": false, "explanation": "Disclosing AI involvement to the platform does not fix the deception, since the posts are still designed to misrepresent that many distinct residents independently hold the same view."}, {"letter": "D", "text": "No, but only because this astroturfing campaign targets a ballot measure rather than a candidate", "correct": false, "explanation": "The prohibition on astroturfing and fabricated grassroots support applies to both ballot measures and candidate campaigns."}], "correct": "A", "select": 1}, {"id": "a6-080", "domain": 6, "task_id": "6.2", "objective": "Apply data sensitivity, regulatory, and privacy considerations", "situation": "A clinical operations coordinator wants Claude to summarise patient case notes to speed up a backlog.", "question": "What should the coordinator do first?", "options": [{"letter": "A", "text": "Use a personal account so that the summaries do not appear in the organisation's shared workspace.", "correct": false, "explanation": "Moving regulated data to a personal account removes oversight and makes the exposure worse."}, {"letter": "B", "text": "Ask Claude to confirm that it complies with health privacy rules before pasting the case notes into a chat.", "correct": false, "explanation": "Compliance is determined by the organisation's own assessment and agreements, not by asking the tool."}, {"letter": "C", "text": "Establish whether the organisation permits patient data in the tool, and under which plan and safeguards.", "correct": true, "explanation": "Regulated health data requires an authorisation and safeguards decision before any of it is entered."}, {"letter": "D", "text": "Remove the patients' surnames while keeping their dates of birth and record numbers in the notes.", "correct": false, "explanation": "Dates of birth and record numbers still identify people, so this is not meaningful de-identification."}], "correct": "C", "select": 1}, {"id": "a6-081", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "A department has an AI policy but keeps no record of which connectors individual staff have linked to Claude.", "question": "What is the greatest risk when a security incident review takes place?", "options": [{"letter": "A", "text": "Staff may duplicate one another's connections, creating extra work whenever a shared document is updated.", "correct": false, "explanation": "Duplicated connections are an efficiency annoyance rather than a barrier to understanding exposure."}, {"letter": "B", "text": "Nobody can establish what company data Claude could reach, so the review cannot scope the possible exposure.", "correct": true, "explanation": "Without an inventory the first question of any review, what data was reachable, cannot be answered, so the scope of the incident stays unknown."}, {"letter": "C", "text": "Connectors left linked for long periods gradually slow down responses in conversations that do not use them.", "correct": false, "explanation": "Performance is not the governance concern, and it does not affect what a review is able to establish."}, {"letter": "D", "text": "The department may exceed the number of connections its plan permits and lose access without any warning.", "correct": false, "explanation": "A capacity limit would be visible immediately and is unrelated to what an incident review needs to determine."}], "correct": "B", "select": 1}, {"id": "a6-082", "domain": 6, "task_id": "6.1", "objective": "Identify appropriate and inappropriate use cases", "situation": "A charity plans to have Claude pre-screen grant applications and produce a shortlist that trustees then review. Which two safeguards matter most before the charity starts?", "question": "Select two.", "options": [{"letter": "A", "text": "Applications are gathered into a single file so that the whole set fits inside one continuous conversation.", "correct": false, "explanation": "How the files are packaged is a convenience question and has no bearing on whether the screening is fair or reviewable."}, {"letter": "B", "text": "The criteria Claude applies are written down and match the criteria the charity published to applicants.", "correct": true, "explanation": "Applicants were judged against published criteria, so the screening must apply those same criteria and they must be inspectable."}, {"letter": "C", "text": "Trustees read the applications Claude ranked lowest before any of them are set aside without funding.", "correct": true, "explanation": "For a different reason: it keeps the harmful outcome recoverable, because a wrongly down-ranked application can still be rescued by a person."}, {"letter": "D", "text": "The charity times each shortlisting round so it can report how many staff hours the new approach has saved.", "correct": false, "explanation": "Measuring time saved is useful for the business case but does nothing to protect applicants from an unfair screen."}, {"letter": "E", "text": "The shortlist is produced on the most capable model available, so the ranking is more accurate.", "correct": false, "explanation": "A stronger model does not make a rights-affecting screen safe, and model choice is not the governance safeguard here."}], "correct": ["B", "C"], "select": 2}, {"id": "a6-083", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "A marketing analyst wants Claude to summarise campaign results that live in the company's Google Drive, so she plans to link Drive to her Claude account. Her company's AI policy states that any new data source must be approved by the data owner before it is connected.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Request the data owner's approval through the process the AI policy defines, before connecting anything at all.", "correct": true, "explanation": "The policy makes approval a precondition for connecting a data source, so the approval request is the first action, not something obtained after the fact."}, {"letter": "B", "text": "Link Drive to one test folder first, then request approval once she can show the results it produces for her.", "correct": false, "explanation": "Connecting even a single folder is still connecting a data source, which is exactly what the policy says requires approval first."}, {"letter": "C", "text": "Link Drive and rely on the fact that Claude works only with files her own account already has permission to open.", "correct": false, "explanation": "Inheriting her own permissions limits exposure but does not replace the organisational approval the policy requires."}, {"letter": "D", "text": "Ask Claude to list the Drive files it is able to read, and use that answer to judge whether approval is needed.", "correct": false, "explanation": "Asking Claude what it can see does not change the fact that the connection itself needed approval before it was made."}], "correct": "A", "select": 1}, {"id": "a6-084", "domain": 6, "task_id": "6.2", "objective": "Apply data sensitivity, regulatory, and privacy considerations", "situation": "A corporate trainer uses a personal Max plan for her own course-design work. Her manager asks her whether her Claude conversations are being used to improve Anthropic's models.", "question": "What should be her first step?", "options": [{"letter": "A", "text": "Assume that they are, because conversations on personal plans always feed model improvement work.", "correct": false, "explanation": "Consumer plans are not automatically in scope for model improvement. It is a user-controlled setting, so assuming the worst gives her manager an answer that may simply be wrong."}, {"letter": "B", "text": "Ask her manager to raise it with Anthropic, as only the vendor can answer that question.", "correct": false, "explanation": "Escalating to the vendor is unnecessary for something visible in her own account settings, and it leaves the manager waiting for an answer she could give today."}, {"letter": "C", "text": "Check her own model improvement privacy setting, since on a personal plan the choice is hers.", "correct": true, "explanation": "On Free, Pro and Max the user chooses whether their conversations may be used to improve Claude, so the honest answer depends on how her own setting is configured. Checking it is a step she can take herself and it produces a factual answer rather than an assumption."}, {"letter": "D", "text": "Assume that they are not, because Anthropic never trains on customer data on any plan at all.", "correct": false, "explanation": "This states the commercial position, where inputs and outputs are not used for training by default, and applies it to a personal plan. The consumer and commercial rules are different and must not be conflated."}], "correct": "C", "select": 1}, {"id": "a6-085", "domain": 6, "task_id": "6.1", "objective": "Identify appropriate and inappropriate use cases", "situation": "A team is considering using Claude to triage inbound complaints, some of which allege safety issues.", "question": "What should they do first?", "options": [{"letter": "A", "text": "Ask Claude to judge which complaints are serious enough to need a person, and follow that judgement.", "correct": false, "explanation": "Delegating the escalation boundary itself removes the safeguard the team is trying to build."}, {"letter": "B", "text": "Decide which complaint categories must always reach a person, and route those out of the automation.", "correct": true, "explanation": "Defining the cases that require human handling has to happen before any triage is automated."}, {"letter": "C", "text": "Delay the project until every complaint category has been rewritten into a standard template form.", "correct": false, "explanation": "Standardising the intake forms does not address which complaints need a human at all."}, {"letter": "D", "text": "Automate the whole queue and review the outcomes at the end of the first quarter of operation.", "correct": false, "explanation": "A quarter of unreviewed safety complaints is exactly the harm this design needs to prevent."}], "correct": "B", "select": 1}, {"id": "a6-086", "domain": 6, "task_id": "6.2", "objective": "Apply data sensitivity, regulatory, and privacy considerations", "situation": "A market research team on the company's Claude for Work plan wants Claude to analyse interview transcripts containing participants' names, employers and health conditions. Which two steps should they take before uploading?", "question": "Select two.", "options": [{"letter": "A", "text": "Confirm the consent participants gave covers having their transcripts processed in an external tool.", "correct": true, "explanation": "Independently of minimisation, health data about identifiable people carries a consent and lawful-basis question that must be settled before processing."}, {"letter": "B", "text": "Open an incognito chat for the upload so that these health details cannot be retained anywhere at all.", "correct": false, "explanation": "Incognito chats are not used to improve Claude, but they are not a lawful basis for processing sensitive personal data and do not address consent."}, {"letter": "C", "text": "Strip out the direct identifiers that the thematic analysis does not actually need in order to work.", "correct": true, "explanation": "Data minimisation comes first: if the themes can be found without names and employers, that sensitive detail should never be pasted in."}, {"letter": "D", "text": "Divide the transcripts across several separate chats so that no single conversation ever holds them all.", "correct": false, "explanation": "Splitting the same sensitive content across chats does not reduce what was disclosed; it only makes the work harder to manage."}, {"letter": "E", "text": "Turn off the model improvement setting on the account so the transcripts cannot be used for training.", "correct": false, "explanation": "Commercial plans such as Claude for Work do not use inputs or outputs to train models by default, so this changes nothing about the real risk."}], "correct": ["A", "C"], "select": 2}, {"id": "a6-087", "domain": 6, "task_id": "6.4", "objective": "Understand the ethical implications of AI usage", "situation": "A team building an insurance claims triage tool on Claude wants to minimize the risk that the model's outputs unfairly disadvantage claimants from a particular demographic group, even though no single field in the input data directly names race or ethnicity.", "question": "What is the most ethically sound step for the team to take?", "options": [{"letter": "A", "text": "Add a single disclaimer stating outputs may be biased, in place of testing for disparate outcomes", "correct": false, "explanation": "A disclaimer does not substitute for actually evaluating and mitigating disparate outcomes in a high-risk application like insurance claims."}, {"letter": "B", "text": "Test outputs for disparate impact across groups, since proxies like zip code can encode bias", "correct": true, "explanation": "Even without explicit protected fields, proxy variables can encode demographic information, so responsible practice requires testing for disparate impact across groups using representative scenarios before deployment."}, {"letter": "C", "text": "Conclude that bias is not possible, since the input data excludes explicitly protected fields", "correct": false, "explanation": "Excluding explicit protected fields does not prevent bias, since correlated proxy variables can still produce disparate outcomes."}, {"letter": "D", "text": "Restrict testing to overall accuracy metrics, since subgroup fairness is outside model evaluation", "correct": false, "explanation": "Overall accuracy can mask unequal error rates across subgroups, so subgroup-level fairness evaluation is necessary in high-risk contexts."}], "correct": "B", "select": 1}, {"id": "a6-088", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "A compliance lead is writing an internal FAQ explaining how staff may use Claude.", "question": "Which statement about model training is accurate to include?", "options": [{"letter": "A", "text": "Inputs and outputs from commercial products are used for training unless an organisation owner opts the account out.", "correct": false, "explanation": "This reverses the commercial default; the commercial position is that data is not used for training unless the customer opts in."}, {"letter": "B", "text": "Training behaviour is the same on personal and commercial plans, so a single sentence can cover every account type.", "correct": false, "explanation": "Consumer and commercial rules differ, and conflating them is the most common error an FAQ like this can make."}, {"letter": "C", "text": "Inputs and outputs from commercial products such as Claude for Work are not used to train models by default.", "correct": true, "explanation": "Commercial products do not use customer inputs and outputs for model training by default, which is the distinction staff most often get wrong."}, {"letter": "D", "text": "Personal Free, Pro and Max accounts never contribute any conversation content to model training under any circumstance.", "correct": false, "explanation": "On consumer plans the user chooses, and content can be used for model improvement if that setting is enabled, so 'never' is wrong."}], "correct": "C", "select": 1}, {"id": "a6-089", "domain": 6, "task_id": "6.2", "objective": "Apply data sensitivity, regulatory, and privacy considerations", "situation": "A hospital foundation's fundraising officer wants to upload a donor list containing home addresses and giving histories so Claude can segment it for a campaign. The foundation is on a Claude for Work plan.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Confirm that the Claude for Work plan does not use the foundation's data to train models.", "correct": false, "explanation": "It is true that commercial plans do not use inputs and outputs for model training by default, but that is already settled and it is not the constraint here. Training is only one of the questions a donor list raises."}, {"letter": "B", "text": "Run the segmentation in an incognito chat so that none of the donor details are retained at all.", "correct": false, "explanation": "Incognito chats are not used to improve Claude, but they are not a substitute for permission to process donor data in the first place, and they do not address what the foundation promised its donors."}, {"letter": "C", "text": "Remove the home addresses but keep the giving history, which she considers less sensitive.", "correct": false, "explanation": "Stripping one field is a reasonable minimisation step, but choosing which fields to remove on her own judgement pre-empts the policy and consent questions that should decide it. Giving history is itself sensitive personal data."}, {"letter": "D", "text": "Check what the foundation's own policy and its donor commitments allow her to do with the list.", "correct": true, "explanation": "The plan already settles the training question; what is unresolved is whether the foundation's own rules and the privacy commitments made to donors permit this data to be put into the tool at all. That governs everything else she does with the list."}], "correct": "D", "select": 1}, {"id": "a6-090", "domain": 6, "task_id": "6.4", "objective": "Understand the ethical implications of AI usage", "situation": "A hiring manager asks Claude to summarise 200 cover letters and rank the candidates for a shortlist.", "question": "What is the most ethically sound way to use that output?", "options": [{"letter": "A", "text": "Treat the ranking as one input, review the applications herself, and record that she made the decision.", "correct": true, "explanation": "A ranking drawn from writing style can disadvantage groups in ways no field names directly, so it may inform a decision about people but must not make it, and a person must own the outcome."}, {"letter": "B", "text": "Accept the ranking but produce it a second time, and shortlist the candidates that appear on both lists.", "correct": false, "explanation": "Agreement between two runs shows consistency, not fairness, and can simply repeat the same bias twice."}, {"letter": "C", "text": "Use the ranking as the shortlist, then ask Claude to explain each position so the reasoning is on record.", "correct": false, "explanation": "An explanation produced after the fact describes the ordering; it does not verify that the ordering was fair."}, {"letter": "D", "text": "Remove the names from the letters first, then use the ranking directly without further human review.", "correct": false, "explanation": "Removing names leaves plenty of signals that correlate with background, and it removes the human review as well."}], "correct": "A", "select": 1}, {"id": "a6-091", "domain": 6, "task_id": "6.1", "objective": "Identify appropriate and inappropriate use cases", "situation": "A benefits team proposes letting Claude decide which employee insurance claims are denied, with staff spot-checking only a sample each month.", "question": "What is the greatest concern?", "options": [{"letter": "A", "text": "Claims processing is too repetitive a task for Claude, so the team would see no real time saving.", "correct": false, "explanation": "Repetitive, well-specified work is well suited to assistance; the problem is who owns the decision."}, {"letter": "B", "text": "Insurance wording changes often, so the team would need to rewrite all of their prompts every single month.", "correct": false, "explanation": "Keeping guidance current is routine maintenance, not the greatest risk in this proposal."}, {"letter": "C", "text": "A decision affecting people's benefits would be made with nobody accountable for each individual outcome.", "correct": true, "explanation": "Decisions about entitlements need a person answerable for every case, not a sampled review afterwards."}, {"letter": "D", "text": "Spot-checking a sample each month costs more staff time than reviewing every single claim would.", "correct": false, "explanation": "Sampling is cheaper than full review; the concern is accountability, not cost."}], "correct": "C", "select": 1}, {"id": "a6-092", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "A client asks a project lead to confirm in writing that material shared during the engagement will not be used to train Anthropic's models.", "question": "What is the appropriate first step before he answers?", "options": [{"letter": "A", "text": "Confirm which Claude plan the engagement's work is actually being carried out on.", "correct": true, "explanation": "The answer depends entirely on which plan the work sits on: commercial products such as Claude for Work do not use inputs and outputs for model training by default, whereas on personal Free, Pro and Max accounts it turns on a user-controlled setting. He cannot answer honestly until he knows which applies."}, {"letter": "B", "text": "Tell the client the team will use incognito chats throughout the whole engagement.", "correct": false, "explanation": "Incognito chats are never used to improve Claude, but reaching for that before knowing how the team actually works substitutes a workaround for an accurate answer, and would not suit an engagement's ongoing work."}, {"letter": "C", "text": "Assure the client that no Anthropic product ever trains on any customer material.", "correct": false, "explanation": "This is too broad to be true. It ignores the difference between commercial and consumer plans, and would be an inaccurate written assurance if any of the work is happening on personal accounts."}, {"letter": "D", "text": "Send the client the instructions for changing model improvement privacy settings.", "correct": false, "explanation": "Those instructions cover personal accounts only. Sending them implies the engagement runs on consumer plans, which he has not established and which may misrepresent the arrangement."}], "correct": "A", "select": 1}, {"id": "a6-093", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "An operations coordinator notices a teammate pasting spreadsheets of customer names and order histories into a personal Claude account, even though the company provides a managed Claude for Work workspace.", "question": "What should the coordinator do first?", "options": [{"letter": "A", "text": "Ask the teammate to delete those conversations, and treat the matter as closed once deletion is confirmed.", "correct": false, "explanation": "Deletion may be part of the remedy, but an individual arrangement bypasses the assessment and record the policy requires."}, {"letter": "B", "text": "Post a general reminder about approved tools in the team channel so everyone is aware of the expectation.", "correct": false, "explanation": "A general reminder does not address a specific incident that has already put customer data outside the managed workspace."}, {"letter": "C", "text": "Wait and see whether the practice continues, raising it only if customer data later appears in a shared file.", "correct": false, "explanation": "Waiting leaves an active exposure unreported and makes the eventual review harder, since the extent will be larger."}, {"letter": "D", "text": "Raise it through the escalation path the company's AI policy defines for suspected data-handling problems.", "correct": true, "explanation": "Customer data has already left the approved environment, and the policy's escalation path exists precisely so this is assessed by the people accountable for it."}], "correct": "D", "select": 1}, {"id": "a6-094", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "A team manager wants to change how his team uses Claude, in a way the team's current written standard does not allow.", "question": "In which order should these steps be taken?", "options": [{"letter": "A", "text": "Write down the proposed change and the data it would involve; update the written standard to match the decision; brief the team on the updated standard; get the policy owner's decision on it.", "correct": false, "explanation": "The standard is updated and the team briefed on the updated standard on the strength of an unapproved proposal, with the policy owner asked only at the end, once the change is already in effect."}, {"letter": "B", "text": "Update the written standard to match the decision; write down the proposed change and the data it would involve; get the policy owner's decision on it; brief the team on the updated standard.", "correct": false, "explanation": "The standard is amended to match a decision that has not been made yet. Rewriting governance ahead of approval is exactly the shortcut a policy exists to stop."}, {"letter": "C", "text": "Write down the proposed change and the data it would involve; get the policy owner's decision on it; update the written standard to match the decision; brief the team on the updated standard.", "correct": true, "explanation": "A written proposal naming the change and the data involved is what the policy owner needs in order to decide; the decision is what authorises the change; the written standard is then updated to match it; and the team is briefed on the updated standard, which is the one they will actually follow."}, {"letter": "D", "text": "Brief the team on the current standard; write down the proposed change and the data it would involve; get the policy owner's decision on it; update the written standard to match the decision.", "correct": false, "explanation": "The team is briefed on the current standard before anything has changed, so the briefing is out of date as soon as the standard is rewritten, and the update only lands after they have already been informed."}], "correct": "C", "select": 1}, {"id": "a6-095", "domain": 6, "task_id": "6.2", "objective": "Apply data sensitivity, regulatory, and privacy considerations", "situation": "A department is preparing to pilot Claude on a workflow that touches customer records.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Classify the data involved; check what policy allows for that class; choose the plan and settings; run a limited pilot.", "correct": true, "explanation": "Know what the data is, learn what is permitted for it, configure to match, then trial it on a small scope."}, {"letter": "B", "text": "Run a limited pilot; classify the data involved; choose the plan and settings; check what policy allows for that class.", "correct": false, "explanation": "Customer records are used before anyone has classified them or checked the rules."}, {"letter": "C", "text": "Check what policy allows for that class; choose the plan and settings; run a limited pilot; classify the data involved.", "correct": false, "explanation": "The policy is checked for a data class that has not yet been determined."}, {"letter": "D", "text": "Choose the plan and settings; run a limited pilot; classify the data involved; check what policy allows for that class.", "correct": false, "explanation": "Configuration is chosen with no idea what the data requires, and the policy check comes last."}], "correct": "A", "select": 1}, {"id": "a6-096", "domain": 6, "task_id": "6.1", "objective": "Identify appropriate and inappropriate use cases", "situation": "A research analyst uses Claude to summarize long industry reports and extract key data points into a spreadsheet for internal use, then verifies the figures against the source documents before including them in a client deliverable.", "question": "Is this an appropriate use of Claude?", "options": [{"letter": "A", "text": "No, because summarizing third-party reports without direct publisher permission always violates policy", "correct": false, "explanation": "Summarizing publicly or lawfully accessed reports for internal analysis does not require separate written publisher permission under the Usage Policy."}, {"letter": "B", "text": "Yes, because summarizing reports and extracting data points are core supported use cases here", "correct": true, "explanation": "Summarizing documents and extracting data points are core supported use cases, and verifying the output before it reaches a client reflects appropriate diligence."}, {"letter": "C", "text": "No, because spreadsheet data extraction falls outside Claude's supported text and code capabilities", "correct": false, "explanation": "Extracting structured data points from text, including into spreadsheet form, is well within Claude's supported text-processing capabilities."}, {"letter": "D", "text": "Yes, but only if extracted data is used solely internally and never in any client-facing document", "correct": false, "explanation": "Nothing in the Usage Policy bars verified, accurate extracted data from being used in a client deliverable; the analyst's verification step supports appropriate use."}], "correct": "B", "select": 1}, {"id": "a6-097", "domain": 6, "task_id": "6.1", "objective": "Identify appropriate and inappropriate use cases", "situation": "A nonprofit organization wants to use AI-generated content to create emotional appeals for fundraising.", "question": "Is this appropriate?", "options": [{"letter": "A", "text": "No, AI‑generated fundraising content is inherently deceptive because it fabricates beneficiary stories.", "correct": false, "explanation": "AI-generated fundraising content is not inherently deceptive; its ethical status depends on how it is created and disclosed. It can be responsible when it faithfully represents real stories and avoids misleading the audience."}, {"letter": "B", "text": "Yes, AI‑generated emotional appeals are always permissible, as the charitable purpose outweighs concerns.", "correct": false, "explanation": "Charitable purpose does not automatically justify AI-generated emotional appeals, as they can still be misleading or untruthful. The key concern is whether the content is accurate and transparent, not the nobility of the cause."}, {"letter": "C", "text": "No, because any emotional appeal in fundraising is manipulative, regardless of AI involvement.", "correct": false, "explanation": "Emotional appeals are not inherently manipulative; they are a common and ethical fundraising tool when based on true stories. The ethical issue is whether the content distorts facts or exploits, not whether it evokes emotion."}, {"letter": "D", "text": "Yes, but only if the beneficiary stories are truthful and accurately represent real events.", "correct": true, "explanation": "AI-generated fundraising content can be appropriate if the beneficiary stories are truthful and accurately represent real events. Ethical use demands that the content not mislead donors or fabricate details, ensuring authenticity and transparency."}], "correct": "D", "select": 1}, {"id": "a6-098", "domain": 6, "task_id": "6.2", "objective": "Apply data sensitivity, regulatory, and privacy considerations", "situation": "A Team-plan owner shares a project organisation-wide so that more colleagues can benefit from the documents it contains.", "question": "What is the greatest concern?", "options": [{"letter": "A", "text": "Everyone in the organisation gains access to knowledge that may include restricted or confidential material.", "correct": true, "explanation": "Broadening a project's audience broadens access to every document already uploaded into it."}, {"letter": "B", "text": "Organisation-wide sharing converts the project's documents into training data for future Claude models.", "correct": false, "explanation": "How a project is shared has no bearing on model training."}, {"letter": "C", "text": "The original owner loses the ability to edit the project once it is visible to the whole organisation.", "correct": false, "explanation": "The owner keeps their access when a project is shared more widely."}, {"letter": "D", "text": "Sharing removes the project's instructions, so the shared version behaves differently for everyone who opens it.", "correct": false, "explanation": "Sharing does not strip a project's instructions."}], "correct": "A", "select": 1}, {"id": "a6-099", "domain": 6, "task_id": "6.4", "objective": "Understand the ethical implications of AI usage", "situation": "A fintech startup wants to use Claude to help loan officers draft explanations for credit decisions before those explanations are sent to applicants.", "question": "Under Anthropic's Usage Policy, what must the company do to use Claude responsibly in this high-risk context?", "options": [{"letter": "A", "text": "Add a disclaimer to the system prompt and skip further review, since the disclaimer satisfies the policy obligations", "correct": false, "explanation": "A disclaimer alone does not satisfy the human-in-the-loop requirement; a qualified professional must still review the specific decision before it goes out."}, {"letter": "B", "text": "Keep Claude's drafts internal only, and avoid mentioning AI involvement in any customer-facing communication", "correct": false, "explanation": "Avoiding disclosure does not meet the policy's transparency requirement, and internal-only use does not remove the review obligation once content reaches applicants."}, {"letter": "C", "text": "Have a loan officer review each explanation before sending it, and disclose to applicants that AI helped draft it", "correct": true, "explanation": "Financial services is a high-risk use case under the Usage Policy, requiring a qualified professional to review AI-assisted content before dissemination and requiring disclosure to the end user."}, {"letter": "D", "text": "Log every generated explanation for six months and rely on periodic audits instead of reviewing each decision", "correct": false, "explanation": "Retroactive periodic audits are not a substitute for prior review of each decision by a qualified professional before it is disseminated."}], "correct": "C", "select": 1}, {"id": "a6-100", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "A team in a regulated industry is setting up a Claude Project to draft client onboarding letters.", "question": "Which sequence best follows a typical organisational AI policy?", "options": [{"letter": "A", "text": "Confirm the use case and data types are permitted; get the data owner's approval; upload the material and set project instructions; pilot with a named human reviewer.", "correct": true, "explanation": "Policy questions come before data moves: establish that the use and the data classes are allowed, obtain the owner's approval, then configure, then prove it works under human review."}, {"letter": "B", "text": "Upload the material and set project instructions; pilot with a named human reviewer; confirm the use case and data types are permitted; get the data owner's approval.", "correct": false, "explanation": "This puts client material into the workspace before anyone has confirmed the use case is permitted, which is the step the policy exists to prevent."}, {"letter": "C", "text": "Upload the material and set project instructions; confirm the use case and data types are permitted; get the data owner's approval; pilot with a named human reviewer.", "correct": false, "explanation": "Uploading first means the data is already in the workspace while the permission question is still open."}, {"letter": "D", "text": "Get the data owner's approval; upload the material and set project instructions; confirm the use case and data types are permitted; pilot with a named human reviewer.", "correct": false, "explanation": "Approval cannot be sought sensibly before the use case itself has been confirmed as permitted, and configuration still precedes that confirmation here."}], "correct": "A", "select": 1}, {"id": "a6-101", "domain": 6, "task_id": "6.2", "objective": "Apply data sensitivity, regulatory, and privacy considerations", "situation": "A colleague realises they pasted a confidential merger document into a chat on their personal Free-plan account.", "question": "What should they do first?", "options": [{"letter": "A", "text": "Delete the conversation and treat the matter as resolved without mentioning it to anyone else.", "correct": false, "explanation": "Deleting the chat hides the incident from the people responsible for assessing it."}, {"letter": "B", "text": "Switch on incognito mode for future chats and carry on working with the merger document there.", "correct": false, "explanation": "This does nothing about the exposure that has already happened, and continues the unapproved use."}, {"letter": "C", "text": "Ask Claude to delete the document from its memory and to confirm that nothing at all has been retained.", "correct": false, "explanation": "An assurance from within the conversation is not a record of containment."}, {"letter": "D", "text": "Report it through the organisation's incident or escalation process so the exposure can be assessed.", "correct": true, "explanation": "A confidentiality incident has to reach the people who can assess and contain it, and quickly."}], "correct": "D", "select": 1}, {"id": "a6-102", "domain": 6, "task_id": "6.4", "objective": "Understand the ethical implications of AI usage", "situation": "A marketing writer is about to publish a company blog post drafted with Claude. Its structure and its main claims follow a published industry study closely, and the post names no source at all.", "question": "What is the most appropriate thing to do?", "options": [{"letter": "A", "text": "Leave the post as it stands, since Claude wrote the words and nothing was copied out verbatim.", "correct": false, "explanation": "Reusing another party's findings and structure needs crediting even when no sentence is copied."}, {"letter": "B", "text": "Reword the passages that resemble the study most closely, so that the similarity is less obvious.", "correct": false, "explanation": "Disguising the resemblance addresses how it looks rather than the missing credit to the source."}, {"letter": "C", "text": "Trace the claims back to that study and credit it, rather than presenting its work as the company's.", "correct": true, "explanation": "Attribution is owed to whoever did the underlying work; readers are otherwise given someone else's research as though the company had produced it."}, {"letter": "D", "text": "Add a line stating the post was written with AI assistance, which covers where the material came from.", "correct": false, "explanation": "Disclosing AI assistance says how the post was drafted, not whose research the claims rest on."}], "correct": "C", "select": 1}, {"id": "a6-103", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "A contractor drafts a client's unreleased pricing strategy in his own personal Claude account, with the setting that helps improve Claude's models switched on.", "question": "What is the biggest concern?", "options": [{"letter": "A", "text": "Usage limits on a personal plan may interrupt him partway through before the pricing work is finished.", "correct": false, "explanation": "Hitting a usage limit delays the work; it does not expose the client's confidential pricing."}, {"letter": "B", "text": "Confidential client content may be retained and used for model improvement under the setting he enabled.", "correct": true, "explanation": "On consumer plans the user decides, and with model improvement enabled the conversation content can be retained and used for model improvement, which the client never agreed to."}, {"letter": "C", "text": "Material held in a personal account may be hard to produce later if the engagement is formally audited.", "correct": false, "explanation": "Producing records later is a genuine secondary problem, but it is smaller than confidential client material being used for model improvement."}, {"letter": "D", "text": "The conversation may grow too long for Claude to keep the whole pricing strategy in view within one chat.", "correct": false, "explanation": "Length is a working constraint within one chat and has nothing to do with the confidentiality of the client's material."}], "correct": "B", "select": 1}, {"id": "a6-104", "domain": 6, "task_id": "6.3", "objective": "Follow organizational AI policies and governance standards", "situation": "A consultancy holds a Claude for Work plan, but several consultants prefer to use their own personal accounts for client deliverables because it feels quicker.", "question": "What is the greatest governance risk?", "options": [{"letter": "A", "text": "Output quality may vary between accounts because personal plans offer a narrower selection of models.", "correct": false, "explanation": "Model availability affects convenience, not the firm's ability to account for where client data was processed."}, {"letter": "B", "text": "Consultants may reach their personal usage limits sooner than they would inside the managed workspace.", "correct": false, "explanation": "Running out of personal allowance is an inconvenience, not the governance exposure created by moving client material outside the managed plan."}, {"letter": "C", "text": "Client material sits outside the plan and the administrative controls the firm's policy relies on.", "correct": true, "explanation": "The firm's commitments to clients rest on the managed environment and its controls; work done in personal accounts sits outside all of them and outside any audit the firm could produce."}, {"letter": "D", "text": "Personal accounts limit how many projects a consultant can keep, which constrains larger engagements.", "correct": false, "explanation": "Project count limits are a workspace constraint and do not describe the risk to client confidentiality."}], "correct": "C", "select": 1}, {"id": "a6-105", "domain": 6, "task_id": "6.4", "objective": "Understand the ethical implications of AI usage", "situation": "An HR team uses Claude to summarise performance notes and rank internal candidates for promotion, with a panel making the final calls. Which two steps address the ethical risks most directly?", "question": "Select two.", "options": [{"letter": "A", "text": "Drop any performance note written before the current review year, using only recent evidence.", "correct": false, "explanation": "Restricting the time window changes the evidence base but says nothing about whether the ranking treats groups equally."}, {"letter": "B", "text": "Tell the candidates that AI was used to prepare the summaries the promotion panel will be reading.", "correct": true, "explanation": "on independent grounds: people affected by a decision are owed transparency about how their case was prepared."}, {"letter": "C", "text": "Examine whether the rankings systematically disadvantage any group before the panel relies on them.", "correct": true, "explanation": "Performance notes carry human bias, and a ranking built on them can reproduce that bias at scale unless it is checked."}, {"letter": "D", "text": "Have Claude state a confidence score for each ranking so that the panel can see which ones are shakiest.", "correct": false, "explanation": "A self-reported confidence score is not a reliable measure of correctness and does not reveal systematic unfairness."}, {"letter": "E", "text": "Include only those candidates who actively agree to have their own notes summarised by Claude at all.", "correct": false, "explanation": "Opting out of a promotion process is not a real choice for candidates and would itself create unequal treatment."}], "correct": ["B", "C"], "select": 2}, {"id": "a6-106", "domain": 6, "task_id": "6.4", "objective": "Understand the ethical implications of AI usage", "situation": "A training provider asks Claude to mark written assessments and intends to use the marks as they come to decide which candidates pass the course.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The mark given to one script may vary slightly if the same script is submitted a second time.", "correct": false, "explanation": "Small variation between runs is worth knowing about, but consistent unfairness is the greater harm."}, {"letter": "B", "text": "Setting the marking up may take the assessors longer than they expected when they agreed to it.", "correct": false, "explanation": "A slower start is an ordinary project cost, not a harm to the candidates being assessed."}, {"letter": "C", "text": "Assessors may gradually lose their familiarity with the assessment once the marking is done for them.", "correct": false, "explanation": "Some loss of familiarity is a real cost, but it does not change who passes and who does not."}, {"letter": "D", "text": "Candidates may be marked down for how they happen to write, with no way for them to challenge it.", "correct": true, "explanation": "This is a decision about people's qualifications; patterns in how the marking treats certain writing styles produce unfair outcomes, and nobody is positioned to notice or contest them."}], "correct": "D", "select": 1}, {"id": "a6-107", "domain": 6, "task_id": "6.2", "objective": "Apply data sensitivity, regulatory, and privacy considerations", "situation": "An operations manager wants Claude to help with the files in one shared folder and is about to connect the company's Google Drive. Which two actions best limit unnecessary exposure?", "question": "Select two.", "options": [{"letter": "A", "text": "Switch on incognito mode for any conversation that reads files from the connected company Drive.", "correct": false, "explanation": "Incognito affects whether chats are kept and used for improvement; it does not limit what a connected Drive exposes."}, {"letter": "B", "text": "Check what the connecting account can already open in Drive, because Claude inherits that same reach.", "correct": true, "explanation": "A connector is scoped by the permissions of the account that links it, so the exposure question is what that account can already reach."}, {"letter": "C", "text": "Obtain whatever approval the organisation's AI policy requires before linking a company data source.", "correct": true, "explanation": "on separate grounds: connecting a corporate data source is a governance decision, and most policies require approval before it happens."}, {"letter": "D", "text": "Copy the folder's files into the project's knowledge and then delete the originals from the shared Drive.", "correct": false, "explanation": "Deleting the originals destroys the team's copy of the files and does nothing to narrow what the connector can see."}, {"letter": "E", "text": "Tell Claude in the prompt to look only at that one folder and rely on it to leave everything else alone.", "correct": false, "explanation": "An instruction in the prompt is not an access control; it does not change what the connector is technically able to reach."}], "correct": ["B", "C"], "select": 2}, {"id": "a6-108", "domain": 6, "task_id": "6.1", "objective": "Identify appropriate and inappropriate use cases", "situation": "A school district's operations team wants Claude to work through the admissions criteria and produce the letters telling families whether their child has a place at an oversubscribed school.", "question": "What should the team do first?", "options": [{"letter": "A", "text": "Ask Claude to restate the district's admissions criteria so the team can confirm it has them right.", "correct": false, "explanation": "Confirming that Claude can repeat the criteria says nothing about whether it should be the thing applying them. A correct restatement would make the team more comfortable handing over a decision it should not hand over."}, {"letter": "B", "text": "Run the letters for one school first and see whether any family complains about the wording.", "correct": false, "explanation": "Using one school as the trial means real families receive decisions from an untested arrangement, and complaints are a late and unreliable signal of whether the allocation was right."}, {"letter": "C", "text": "Separate deciding who gets a place from writing the letters, and keep the decision with staff.", "correct": true, "explanation": "A school place materially affects a child's life and the decision is hard to reverse once families have been told, so it belongs with accountable staff. Drafting the letters afterwards is a text task where an error is recoverable, and splitting the two makes the appropriate part safe to delegate."}, {"letter": "D", "text": "Add an instruction telling Claude to apply the admissions criteria exactly as they are written.", "correct": false, "explanation": "A firmer instruction addresses how faithfully the criteria are applied, not whether an allocation affecting children's schooling should be made this way at all."}], "correct": "C", "select": 1}, {"id": "a6-109", "domain": 6, "task_id": "6.2", "objective": "Apply data sensitivity, regulatory, and privacy considerations", "situation": "A Max-plan user turns off the setting that allows their chats to be used to improve Claude's models.", "question": "What is the effect?", "options": [{"letter": "A", "text": "Chats are still used for training, but the account is excluded from the de-identification process.", "correct": false, "explanation": "Switching the setting off stops the training use; it does not change how data is de-identified."}, {"letter": "B", "text": "Every past conversation is erased from Anthropic's systems immediately once the setting is switched off.", "correct": false, "explanation": "Turning off a training setting is not the same as deleting stored data."}, {"letter": "C", "text": "Nothing changes until the account is upgraded, since the setting applies only on Team and Enterprise plans.", "correct": false, "explanation": "This is a consumer-plan setting; business plans work differently."}, {"letter": "D", "text": "New chats will not be used for future training, and previously stored chats also stop being used.", "correct": true, "explanation": "Switching it off applies to new conversations and stops previously stored ones feeding future training runs."}], "correct": "D", "select": 1}, {"id": "a7-001", "domain": 7, "task_id": "7.2", "objective": "Adjust approach based on feedback and results", "situation": "Results show that Claude handles four of the five steps in a reporting task well but consistently mishandles the fifth.", "question": "What is the best adjustment?", "options": [{"letter": "A", "text": "Rewrite the whole request from scratch so all five steps are described in a consistent new style.", "correct": false, "explanation": "A full rewrite puts four working steps at risk to fix one, and makes the cause harder to see."}, {"letter": "B", "text": "Work on that step separately with clearer direction and an example, leaving the rest untouched.", "correct": true, "explanation": "The evidence localises the fault to one step, so the change should be localised too; rewriting what already works risks losing it."}, {"letter": "C", "text": "Ask Claude to carry out all five steps twice and take the better of the two attempts each time.", "correct": false, "explanation": "Running everything twice doubles the effort without addressing why the fifth step fails."}, {"letter": "D", "text": "Split the task across five separate conversations so that each step gets a dedicated attempt.", "correct": false, "explanation": "Five conversations add coordination work and lose the context the steps share with each other."}], "correct": "B", "select": 1}, {"id": "a7-002", "domain": 7, "task_id": "7.1", "objective": "Identify, diagnose, and resolve issues with underperforming prompts or poor outputs", "situation": "A communications officer asks Claude to 'write an event invitation' and gets something generic that could describe almost any event. Which two changes are most likely to fix it?", "question": "Select two.", "options": [{"letter": "A", "text": "Paste in the actual agenda and speaker biographies rather than describing them loosely from memory.", "correct": true, "explanation": "The output is generic because nothing event-specific was supplied; the source material has to go into the prompt."}, {"letter": "B", "text": "Ask for ten different versions of the invitation and then simply pick whichever one happens to read best.", "correct": false, "explanation": "Ten generic drafts are still generic, and the extra generations consume allowance for no real gain."}, {"letter": "C", "text": "Send exactly the same request over again, but this time using the most capable model that is available.", "correct": false, "explanation": "Reaching for a larger model does not supply the missing event detail, so the answer stays generic."}, {"letter": "D", "text": "Add more emphasis to the request, telling Claude the invitation really has to be genuinely excellent.", "correct": false, "explanation": "Emphasis adds no information; insisting on quality tells Claude nothing it did not already have."}, {"letter": "E", "text": "State who the invitation is for, roughly how long it should be, and what the reader has to do.", "correct": true, "explanation": "on a separate axis: even with the details, audience, length and purpose have to be specified or the shape of the invitation is left to chance."}], "correct": ["A", "E"], "select": 2}, {"id": "a7-003", "domain": 7, "task_id": "7.1", "objective": "Identify, diagnose, and resolve issues with underperforming prompts or poor outputs", "situation": "A marketing writer asks Claude to 'write a launch email' and receives something generic that could describe almost any product.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Add a line asking Claude to be more creative and to write with noticeably more energy and detail.", "correct": false, "explanation": "Adjectives such as 'creative' add emphasis but no information about what the email should say."}, {"letter": "B", "text": "Move to the most capable model available, since generic writing suggests limited reasoning depth.", "correct": false, "explanation": "Reaching for a larger model does not tell it anything about the product or the audience."}, {"letter": "C", "text": "Produce the email several more times and choose whichever version reads best for the audience.", "correct": false, "explanation": "Regenerating merely samples different wordings of the same under-specified request."}, {"letter": "D", "text": "Restate the request with the audience, offer and length, and paste the actual product brief.", "correct": true, "explanation": "A generic answer to a generic request is a description problem: the audience, the offer and the source material were never supplied, and no other change substitutes for supplying them."}], "correct": "D", "select": 1}, {"id": "a7-004", "domain": 7, "task_id": "7.1", "objective": "Identify, diagnose, and resolve issues with underperforming prompts or poor outputs", "situation": "A marketing writer's prompt asks Claude to \"write engaging product descriptions.\" The outputs are technically correct but read as flat and generic compared to what the writer had in mind.", "question": "Which change is most likely to close that gap?", "options": [{"letter": "A", "text": "Explicitly describe the desired tone, voice, and level of creative flourish rather than relying on Claude to infer \"engaging\" from one adjective.", "correct": true, "explanation": "Anthropic's guidance is that if you want \"above and beyond\" behavior, you should explicitly request it rather than relying on the model to infer this from vague prompts; naming the specific tone and style closes the gap between what was asked and what was wanted."}, {"letter": "B", "text": "Append the word \"please\" along with several exclamation points so Claude treats the entire request with a substantially higher degree of enthusiasm.", "correct": false, "explanation": "Politeness markers or punctuation do not change how specifically the task is defined and are not a documented technique for shaping tone."}, {"letter": "C", "text": "Reduce the max_tokens limit noticeably so Claude is forced into far terser, punchier phrasing within every product description it writes overall.", "correct": false, "explanation": "Shortening the token budget constrains length, not voice or creative quality, and does not address the underlying vagueness of \"engaging.\""}, {"letter": "D", "text": "Generate ten separate variations of each description and randomly select one, assuming sheer variety alone will surface a livelier result.", "correct": false, "explanation": "Randomly sampling from multiple generic outputs does not add the missing specificity about desired tone and is not a reliable improvement technique."}], "correct": "A", "select": 1}, {"id": "a7-005", "domain": 7, "task_id": "7.3", "objective": "Optimize workflows for efficiency and effectiveness", "situation": "Every week, four members of an operations team each re-explain the same reporting conventions to Claude before asking for their report. Each of them has found wording that works.", "question": "What is the greatest risk of leaving this as it is?", "options": [{"letter": "A", "text": "Claude will begin to apply one member's conventions to everyone else's reports once it has seen them often enough", "correct": false, "explanation": "Conversations are separate; one person's wording does not carry over into a colleague's chats."}, {"letter": "B", "text": "Four private versions of the conventions will drift apart, so the reports stop matching one another over time", "correct": true, "explanation": "The waste is minor next to the drift: four separately maintained versions of the same conventions will diverge, and the reports will quietly stop matching. Moving the conventions into shared project instructions makes the fix persistent and consistent."}, {"letter": "C", "text": "Re-explaining the conventions each week will gradually train Claude on them, which the team's policy does not permit", "correct": false, "explanation": "Typing instructions into a conversation does not train the model on them."}, {"letter": "D", "text": "The repeated explanations will be stored and counted against the team's shared project knowledge capacity every week", "correct": false, "explanation": "Chat messages are not written into project knowledge, so they do not consume its capacity."}], "correct": "B", "select": 1}, {"id": "a7-006", "domain": 7, "task_id": "7.2", "objective": "Adjust approach based on feedback and results", "situation": "Several colleagues tell a project coordinator that Claude's meeting summaries 'miss the decisions'. Nobody has said which decisions or which meetings.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Read a few criticised summaries beside their transcripts to see what kind of item is missed.", "correct": true, "explanation": "'Misses the decisions' could mean several different faults, and the fix differs for each, so the first move is to see the failures against the source before changing anything."}, {"letter": "B", "text": "Move the summaries to a more capable model, since missing content suggests shallow reasoning.", "correct": false, "explanation": "Model choice is being changed before anyone knows whether the summaries were even asked for the decisions."}, {"letter": "C", "text": "Rewrite the instructions to say that decisions must always be listed at the top of every summary.", "correct": false, "explanation": "A specific rule written before the fault is understood may address something that was never the problem."}, {"letter": "D", "text": "Ask colleagues to flag any summary that is wrong, and return to the question again in a month.", "correct": false, "explanation": "Collecting more of the same vague feedback delays the diagnosis that a few examples would already allow."}], "correct": "A", "select": 1}, {"id": "a7-007", "domain": 7, "task_id": "7.3", "objective": "Optimize workflows for efficiency and effectiveness", "situation": "A manager is close to her weekly usage limit, is inside a very long conversation, and still has an unrelated task to finish.", "question": "What is the most sensible step?", "options": [{"letter": "A", "text": "Carry on in the long conversation, since keeping the history avoids explaining things twice.", "correct": false, "explanation": "The unrelated task gains nothing from the history, while the length keeps adding to what each message costs."}, {"letter": "B", "text": "Start a new conversation for the unrelated task with only the material that task needs.", "correct": true, "explanation": "Long conversations consume more of the allowance as they grow, and the new task shares none of that history, so a fresh conversation with just the relevant material is both cheaper and clearer."}, {"letter": "C", "text": "Wait until the weekly limit resets before beginning the unrelated task in any conversation.", "correct": false, "explanation": "Waiting stops the work altogether when a new conversation would let it continue now."}, {"letter": "D", "text": "Ask Claude to summarise the long conversation and then continue working beneath that summary.", "correct": false, "explanation": "Summarising to keep going still carries the old thread into work that has no use for it."}], "correct": "B", "select": 1}, {"id": "a7-008", "domain": 7, "task_id": "7.2", "objective": "Adjust approach based on feedback and results", "situation": "A team has made three successive changes to a shared instruction. The output is now clearly worse than before the changes, and nobody is sure which change did it.", "question": "What is the most appropriate response?", "options": [{"letter": "A", "text": "Ask Claude which of the three changes it thinks damaged the output the most", "correct": false, "explanation": "Claude cannot know which edit degraded results for this team's work; the answer would be speculation."}, {"letter": "B", "text": "Return to the last version that worked and reintroduce changes one at a time", "correct": true, "explanation": "Rolling back to a known-good version restores quality immediately, and reintroducing changes singly is the only way to find out which one was responsible."}, {"letter": "C", "text": "Add a fourth change intended to counteract whatever the other three caused", "correct": false, "explanation": "A compensating fourth change adds another variable to a situation that is already impossible to attribute."}, {"letter": "D", "text": "Keep the current version and ask the reviewers to correct the output by hand", "correct": false, "explanation": "Manual correction absorbs the damage indefinitely and leaves the faulty instruction in place for everyone."}], "correct": "B", "select": 1}, {"id": "a7-009", "domain": 7, "task_id": "7.1", "objective": "Identify, diagnose, and resolve issues with underperforming prompts or poor outputs", "situation": "A researcher's request for \"an analysis of the survey results\" returns a bland restatement of the numbers.", "question": "Which change is most likely to help?", "options": [{"letter": "A", "text": "Split the survey into ten parts and send each one as a separate request", "correct": false, "explanation": "Splitting an under-specified request into ten under-specified requests produces ten bland fragments."}, {"letter": "B", "text": "Send the same request again with the word \"detailed\" added to the front", "correct": false, "explanation": "Adding an emphasis word does not tell Claude what depth means here, so the answer stays shallow."}, {"letter": "C", "text": "State the questions the analysis must answer and who will read the result", "correct": true, "explanation": "\"Analyse this\" does not say what counts as an insight or for whom. Naming the questions and the audience gives the response something to be deep about."}, {"letter": "D", "text": "Ask Claude to say how confident it is in the analysis that it has just given", "correct": false, "explanation": "A statement of confidence describes the existing shallow answer rather than producing a better one."}], "correct": "C", "select": 1}, {"id": "a7-010", "domain": 7, "task_id": "7.3", "objective": "Optimize workflows for efficiency and effectiveness", "situation": "An analyst assembles one client brief by sending Claude eight short messages, each adding another fragment of the source material.", "question": "What would make this more efficient?", "options": [{"letter": "A", "text": "Ask Claude to hold off responding until all eight fragments have arrived in the conversation.", "correct": false, "explanation": "Waiting still requires eight messages, and it delays the point at which she can correct course."}, {"letter": "B", "text": "Put the request and all the material into one well-specified opening message, then refine it.", "correct": true, "explanation": "Batching the related material into a single, clearly specified request is the efficient pattern: Claude sees the whole picture at once instead of reasoning from a partial one eight times."}, {"letter": "C", "text": "Split the eight messages across two conversations so that each holds a manageable amount.", "correct": false, "explanation": "Two conversations split the material so neither one has everything the brief actually needs."}, {"letter": "D", "text": "Send the eight messages in a fixed order each time, so the assembly follows a predictable pattern.", "correct": false, "explanation": "A fixed order makes the same inefficient exchange more predictable without making it shorter."}], "correct": "B", "select": 1}, {"id": "a7-011", "domain": 7, "task_id": "7.2", "objective": "Adjust approach based on feedback and results", "situation": "A reviewer asks for far more background in every Claude-drafted briefing, but the team had already agreed that briefings must fit on one page.", "question": "What is the most appropriate response?", "options": [{"letter": "A", "text": "Take the conflict back to the team and settle which of the two now holds", "correct": true, "explanation": "The feedback contradicts the agreed definition of good. Adjusting the approach means resolving that contradiction openly, not silently letting the newest voice overwrite an agreed standard."}, {"letter": "B", "text": "Follow the reviewer's request, because feedback from a reviewer takes priority", "correct": false, "explanation": "Acting on it immediately lets one reviewer overturn a team decision without the team knowing it has changed."}, {"letter": "C", "text": "Produce both a one-page version and a long version for every briefing sent out", "correct": false, "explanation": "Producing two versions of everything doubles the work and leaves the underlying disagreement unresolved."}, {"letter": "D", "text": "Ignore the request, because the one-page rule was agreed before the feedback", "correct": false, "explanation": "Dismissing the feedback because it came later ignores the possibility that the one-page rule is now wrong."}], "correct": "A", "select": 1}, {"id": "a7-012", "domain": 7, "task_id": "7.2", "objective": "Adjust approach based on feedback and results", "situation": "Several engineers tell a site coordinator that Claude's shift-handover digests 'skip the open risks', but nobody says which risks or which shifts. Which two steps should the coordinator take first?", "question": "Select two.(Select 2)", "options": [{"letter": "A", "text": "Ask for two digests where an open risk was missed, with the missing risk actually named.", "correct": true, "explanation": "Vague feedback cannot be acted on; specific failing examples turn an impression into something diagnosable."}, {"letter": "B", "text": "Rewrite the shared project instructions straight away to stress that open risks matter most of all.", "correct": false, "explanation": "Changing shared instructions before the cause is known risks fixing the wrong thing for everyone at once."}, {"letter": "C", "text": "Compare those digests against the shift logs to see exactly what was left out of them.", "correct": true, "explanation": "on separate grounds: only the source shows whether the risk was absent from the log, buried, or genuinely dropped."}, {"letter": "D", "text": "Ask everyone to score every future digest out of ten so that a trend can be tracked over time.", "correct": false, "explanation": "A score captures sentiment but still never says which risks went missing, or why they went missing."}, {"letter": "E", "text": "Switch the digests into a table format, since open risks may then be easier to spot in a table.", "correct": false, "explanation": "A format change may help presentation, but it cannot add a risk the digest never captured."}], "correct": ["A", "C"], "select": 2}, {"id": "a7-013", "domain": 7, "task_id": "7.2", "objective": "Adjust approach based on feedback and results", "situation": "A pilot shows that Claude's drafts need almost no editing for routine status updates but heavy editing for sensitive client letters.", "question": "What is the best adjustment to the workflow?", "options": [{"letter": "A", "text": "Keep using Claude for both, since the editing effort on the letters will fall as the team practises.", "correct": false, "explanation": "Hoping the editing burden falls ignores what the pilot actually measured about sensitive letters."}, {"letter": "B", "text": "Use Claude for the client letters only, as that is where the largest time saving would be found.", "correct": false, "explanation": "Aiming at the task with the heaviest editing burden pursues the saving that the evidence contradicts."}, {"letter": "C", "text": "Stop using Claude for both, because uneven quality makes the workflow too unreliable to depend on.", "correct": false, "explanation": "Abandoning both discards a demonstrated benefit because of a separate task that did not work."}, {"letter": "D", "text": "Keep Claude on the routine updates and have people write the client letters, revisiting this later.", "correct": true, "explanation": "The results draw the delegation boundary for them: keep the work where it performs well, keep the judgment-heavy work with people, and revisit as the approach matures."}], "correct": "D", "select": 1}, {"id": "a7-014", "domain": 7, "task_id": "7.2", "objective": "Adjust approach based on feedback and results", "situation": "After one weak result, a marketing lead rewrites her prompt, gets a good answer, and immediately tells the team to adopt the new wording as the standard for all campaign briefs.", "question": "What is the biggest concern with this?", "options": [{"letter": "A", "text": "Rewriting a prompt after a weak result is the wrong response, because the first output should always be accepted as given", "correct": false, "explanation": "Iterating after a weak result is exactly the right response; the problem is the size of the evidence, not the act of revising."}, {"letter": "B", "text": "Prompt wording cannot be shared between colleagues, so the team will be unable to reuse whatever she found to work well", "correct": false, "explanation": "Prompt wording is ordinary text and is easy to share; nothing prevents the team reusing it."}, {"letter": "C", "text": "One good result on one brief is thin evidence, and the wording may suit that example rather than the task", "correct": true, "explanation": "A single success is not evidence that a change generalises. The improvement may reflect that one brief's specifics, so it needs checking across a few different cases before it becomes the team standard."}, {"letter": "D", "text": "A standard prompt will stop the team from supplying their own context, so every brief will come back equally generic", "correct": false, "explanation": "A shared starting prompt does not prevent anyone adding their own context on top of it."}], "correct": "C", "select": 1}, {"id": "a7-015", "domain": 7, "task_id": "7.3", "objective": "Optimize workflows for efficiency and effectiveness", "situation": "A support team keeps a Claude project for drafting customer replies.", "question": "Which item is best placed in the project's knowledge rather than its instructions?", "options": [{"letter": "A", "text": "The rule that every reply should end with the support team's standard closing sentence.", "correct": false, "explanation": "A standard closing sentence is a rule about how replies are written, so it is an instruction."}, {"letter": "B", "text": "The direction to write in short paragraphs and to avoid technical language with customers.", "correct": false, "explanation": "Paragraph length and vocabulary are directions about style, which belong in the instructions."}, {"letter": "C", "text": "The requirement that any refund offered must be checked by a supervisor before sending.", "correct": false, "explanation": "A supervisor check is a working rule for the team's replies rather than reference material."}, {"letter": "D", "text": "The current product manual, which changes rarely and is needed by every chat there.", "correct": true, "explanation": "Project knowledge holds reference material Claude should be able to draw on; the other three are directions about how to respond, which belong in the instructions."}], "correct": "D", "select": 1}, {"id": "a7-016", "domain": 7, "task_id": "7.2", "objective": "Adjust approach based on feedback and results", "situation": "A team lead is told by a colleague that \"the client thinks the summaries are off\". No example and no further detail came with it.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Get one specific summary the client objected to and what was wrong with it", "correct": true, "explanation": "Secondhand, unspecific feedback cannot be acted on. Getting one concrete example turns a rumour into something diagnosable, and it may turn out to be a single unusual case."}, {"letter": "B", "text": "Ask Claude to guess what a client in this sector would object to in a summary", "correct": false, "explanation": "Asking Claude to speculate about the objection produces a plausible guess that is no closer to what the client actually said."}, {"letter": "C", "text": "Rewrite the shared instructions to make the summaries shorter and more formal", "correct": false, "explanation": "Changing shared instructions on a paraphrase risks fixing something the client never raised and breaking what already worked."}, {"letter": "D", "text": "Stop using the arrangement until the client confirms the problem has been fixed", "correct": false, "explanation": "Suspending a working arrangement over an unexamined comment is a disproportionate response to feedback nobody has confirmed."}], "correct": "A", "select": 1}, {"id": "a7-017", "domain": 7, "task_id": "7.1", "objective": "Identify, diagnose, and resolve issues with underperforming prompts or poor outputs", "situation": "A support-ticket triage prompt gives Claude a single sentence: \"Categorize this ticket.\" Categories come back inconsistent between runs, and the team wants a fix that addresses the root cause rather than adding post-processing.", "question": "What should they change first?", "options": [{"letter": "A", "text": "Lower the temperature parameter to zero so sampling becomes deterministic and the same ticket always returns the same category.", "correct": false, "explanation": "Temperature affects sampling randomness, not whether the model understands what categories exist or how to choose between them; it does not address an underspecified prompt."}, {"letter": "B", "text": "Route every ticket to a larger model on the assumption that added model capacity will infer the missing category definitions.", "correct": false, "explanation": "Swapping models does not resolve ambiguity in the instructions themselves; an underspecified prompt will still under perform on a larger model."}, {"letter": "C", "text": "Rewrite the prompt to list the exact categories, the desired output format, and the criteria for choosing between them.", "correct": true, "explanation": "Claude responds well to clear, explicit instructions; a vague one-line prompt lacking the category list, output format, and selection criteria is the classic underperformance pattern, and the documented fix is to be specific about desired output and constraints."}, {"letter": "D", "text": "Have a reviewer manually relabel any ticket that looks miscategorized before the ticket reaches a customer-facing queue.", "correct": false, "explanation": "Manual relabeling is a workaround applied after generation, not a fix to the prompt causing the inconsistency."}], "correct": "C", "select": 1}, {"id": "a7-018", "domain": 7, "task_id": "7.3", "objective": "Optimize workflows for efficiency and effectiveness", "situation": "A team's Claude-assisted reporting routine includes a step in which Claude summarises a document that a colleague then reads in full anyway.", "question": "What is the most appropriate optimisation?", "options": [{"letter": "A", "text": "Drop the summarising step, because nothing downstream depends on its output", "correct": true, "explanation": "A step whose output nobody uses is pure overhead. Removing work that changes no decision is the cheapest and most durable efficiency gain available."}, {"letter": "B", "text": "Keep the step and ask for a shorter summary so it costs the team less time", "correct": false, "explanation": "Shortening a summary that nobody relies on reduces the waste slightly but keeps a step that serves no purpose."}, {"letter": "C", "text": "Keep the step and move the summary into the project so it can be reused later", "correct": false, "explanation": "Storing an unused summary preserves it for reuse that is not happening and adds material to maintain."}, {"letter": "D", "text": "Keep the step and have a second colleague read the summary as a cross-check", "correct": false, "explanation": "Adding a second reader to an unused output doubles the effort spent on something that changes nothing."}], "correct": "A", "select": 1}, {"id": "a7-019", "domain": 7, "task_id": "7.1", "objective": "Identify, diagnose, and resolve issues with underperforming prompts or poor outputs", "situation": "A bid writer asks Claude to \"check whether our answer meets the requirement\" and pastes only the answer, not the requirement. The reply is confident and unhelpful.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Paste the requirement itself, so Claude has the thing it must check against", "correct": true, "explanation": "The request asked for a comparison while supplying only one side of it. Supplying the missing material is the fix; nothing else can substitute for the requirement text."}, {"letter": "B", "text": "Move to a more capable model and send exactly the same message once again", "correct": false, "explanation": "A stronger model still cannot see a requirement that was never provided, so the answer will be confidently unhelpful again."}, {"letter": "C", "text": "Ask Claude to recall the requirement from what it knows about bids like this", "correct": false, "explanation": "Asking Claude to reconstruct the requirement invites a plausible invention that the writer would then check her answer against."}, {"letter": "D", "text": "Repeat the same request and stress that the check must be thorough this time", "correct": false, "explanation": "Repeating the request with added emphasis changes the tone, not the information, so the same gap remains."}], "correct": "A", "select": 1}, {"id": "a7-020", "domain": 7, "task_id": "7.3", "objective": "Optimize workflows for efficiency and effectiveness", "situation": "A project a team has relied on for a year now occasionally produces answers that reflect last year's process.", "question": "What routine would best prevent this recurring?", "options": [{"letter": "A", "text": "A habit of starting a fresh conversation whenever an answer looks as though it may be wrong.", "correct": false, "explanation": "A fresh conversation draws on the same project knowledge, so the same old process reappears."}, {"letter": "B", "text": "A scheduled review of the project's instructions and documents, deleting superseded ones.", "correct": true, "explanation": "Configuration decays as the business changes, so maintenance has to be a scheduled task; removing superseded material is what stops the old process being cited again."}, {"letter": "C", "text": "A rule that colleagues read every answer carefully before acting on the information in it.", "correct": false, "explanation": "Careful reading catches some errors after the fact but leaves the outdated source in place."}, {"letter": "D", "text": "A note in the instructions asking Claude to flag whenever information may be out of date.", "correct": false, "explanation": "Claude cannot reliably tell that a document it was given has since been superseded."}], "correct": "B", "select": 1}, {"id": "a7-021", "domain": 7, "task_id": "7.2", "objective": "Adjust approach based on feedback and results", "situation": "An operations lead has corrected the same fault in Claude's output for three weeks running and now wants to deal with it properly.", "question": "Which sequence is correct?", "options": [{"letter": "A", "text": "Identify the pattern they share; re-check the change on those same examples; collect the specific outputs that went wrong; change the instruction to address that pattern.", "correct": false, "explanation": "A pattern cannot be identified before the failing examples have been collected."}, {"letter": "B", "text": "Collect the specific outputs that went wrong; identify the pattern they share; change the instruction to address that pattern; re-check the change on those same examples.", "correct": true, "explanation": "Evidence first, then the pattern, then a change aimed at that pattern, then a re-check on the same cases so the improvement is measured rather than assumed."}, {"letter": "C", "text": "Change the instruction to address the pattern; collect the specific outputs that went wrong; re-check the change on those same examples; identify the pattern they share.", "correct": false, "explanation": "Changing the instruction before gathering the failures means aiming at a pattern nobody has established."}, {"letter": "D", "text": "Re-check the change on those same examples; change the instruction to address the pattern; identify the pattern they share; collect the specific outputs that went wrong.", "correct": false, "explanation": "Re-checking a change that has not been made yet inverts the whole loop."}], "correct": "B", "select": 1}, {"id": "a7-022", "domain": 7, "task_id": "7.2", "objective": "Adjust approach based on feedback and results", "situation": "Feedback on a Claude-drafted grant report is that it reads well but covers the wrong reporting period.", "question": "Which adjustment addresses the cause?", "options": [{"letter": "A", "text": "Ask Claude to check its own report for errors before it hands anything over", "correct": false, "explanation": "Self-checking against an unstated period cannot detect that the wrong period was used."}, {"letter": "B", "text": "Say which reporting period the draft must cover, in the request itself", "correct": true, "explanation": "The fault is an unstated requirement, so the adjustment belongs in the request. Stating the period is what stops the same error recurring on every future report."}, {"letter": "C", "text": "Ask for a more formal register so the report reads as more authoritative", "correct": false, "explanation": "Register was never the problem; the report already reads well and would simply be wrong more formally."}, {"letter": "D", "text": "Ask for a shorter report so that errors are easier for a reviewer to spot", "correct": false, "explanation": "A shorter report makes the error easier to catch but does nothing to stop it being made."}], "correct": "B", "select": 1}, {"id": "a7-023", "domain": 7, "task_id": "7.3", "objective": "Optimize workflows for efficiency and effectiveness", "situation": "Each person on a team works out their own way of fixing the same recurring formatting problem by correcting Claude in the moment, and nobody records the fix anywhere.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "Quality depends on who ran the task that day, and the fix is lost whenever that person is away.", "correct": true, "explanation": "An unrecorded fix is not a fix but a habit, so output varies with the person, and the knowledge leaves the team whenever they do."}, {"letter": "B", "text": "Colleagues may word their corrections differently, making conversations harder to search later.", "correct": false, "explanation": "Searchability is a minor inconvenience compared with inconsistent output reaching colleagues."}, {"letter": "C", "text": "Correcting Claude in the moment takes a little longer than getting the format right first time.", "correct": false, "explanation": "The extra minutes are real but small next to results that cannot be relied on from one person to the next."}, {"letter": "D", "text": "Claude may eventually stop responding to corrections repeated within the same conversation.", "correct": false, "explanation": "Claude does not stop accepting corrections, so this describes a behaviour that will not occur."}], "correct": "A", "select": 1}, {"id": "a7-024", "domain": 7, "task_id": "7.2", "objective": "Adjust approach based on feedback and results", "situation": "A customer-service supervisor has four weeks of mixed comments on Claude-drafted reply templates — some praise, some complaints, several contradicting each other.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Change the instruction for that one fault; group the comments to find the repeated fault; run it on several recent real cases; record why the change was made", "correct": false, "explanation": "Making the change first and analysing afterwards inverts diagnosis and treatment, so the analysis can only justify a decision already taken."}, {"letter": "B", "text": "Group the comments to find the repeated fault; change the instruction for that one fault; run it on several recent real cases; record why the change was made", "correct": true, "explanation": "Find the pattern before acting, change one thing, confirm it on a spread of real cases rather than the one that prompted the complaint, then write down why so nobody undoes it later."}, {"letter": "C", "text": "Run it on several recent real cases; group the comments to find the repeated fault; record why the change was made; change the instruction for that one fault", "correct": false, "explanation": "Running cases before knowing what fault is being investigated produces output with nothing to compare it against."}, {"letter": "D", "text": "Change the instruction for that one fault; run it on several recent real cases; group the comments to find the repeated fault; record why the change was made", "correct": false, "explanation": "Changing the instruction before the comments are grouped means acting on whichever comment was loudest rather than on the pattern."}], "correct": "B", "select": 1}, {"id": "a7-025", "domain": 7, "task_id": "7.3", "objective": "Optimize workflows for efficiency and effectiveness", "situation": "Four colleagues on personal Pro accounts want to work from one teammate's well-tuned project setup.", "question": "What is the accurate advice?", "options": [{"letter": "A", "text": "Project sharing needs a Team or Enterprise plan, so on personal plans each recreates the setup.", "correct": true, "explanation": "Sharing a project with named colleagues or an organisation is a Team and Enterprise capability; on personal plans a project is personal, so the instructions and material have to be recreated by each person."}, {"letter": "B", "text": "Personal plans do allow project sharing, but only with colleagues who are also on a paid plan.", "correct": false, "explanation": "There is no personal-plan sharing that depends on the recipient's plan; personal projects are not shareable."}, {"letter": "C", "text": "Any Claude user can share a project with named colleagues, so she should grant each of them access.", "correct": false, "explanation": "Sharing is not available on personal plans, so this describes a capability the team does not have."}, {"letter": "D", "text": "She should give the colleagues her own account details so they can work in the project directly.", "correct": false, "explanation": "Sharing account credentials breaches normal account rules and creates an untraceable mess of work."}], "correct": "A", "select": 1}, {"id": "a7-026", "domain": 7, "task_id": "7.3", "objective": "Optimize workflows for efficiency and effectiveness", "situation": "A four-person team runs out of its weekly Claude allowance every Thursday and work stalls until the limit resets.", "question": "What should they do first?", "options": [{"letter": "A", "text": "Upgrade the whole team to a higher plan so the weekly ceiling stops interrupting work.", "correct": false, "explanation": "Upgrading may be the right answer eventually, but paying to avoid a diagnosis usually hides the real cause."}, {"letter": "B", "text": "Look at which tasks are actually consuming the allowance before changing plans or habits.", "correct": true, "explanation": "Until the team knows where the allowance is actually going, every remedy is a guess; checking which work is consuming it points at the habit or task to change."}, {"letter": "C", "text": "Move all routine work to the fastest model, as it is the cheapest option for daily tasks.", "correct": false, "explanation": "Forcing the fastest model on everything trades quality away without knowing what actually consumed the allowance."}, {"letter": "D", "text": "Ask everyone to keep messages short and avoid pasting long documents into conversations.", "correct": false, "explanation": "Short messages can cost more overall if they cause long back-and-forth exchanges to reach the same result."}], "correct": "B", "select": 1}, {"id": "a7-027", "domain": 7, "task_id": "7.2", "objective": "Adjust approach based on feedback and results", "situation": "One stakeholder complains that a summary was too detailed, so the team immediately rewrites the shared project instructions to make every summary much shorter.", "question": "What is the biggest concern?", "options": [{"letter": "A", "text": "Shorter summaries are quicker to produce, which may make the change hard to justify on cost.", "correct": false, "explanation": "Production effort is not the concern; the concern is that other users' needs were never consulted."}, {"letter": "B", "text": "The team will have to explain the new instructions to colleagues who joined the project recently.", "correct": false, "explanation": "Explaining the change to new colleagues is routine and unrelated to whether the change was right."}, {"letter": "C", "text": "The stakeholder may not notice the change and could raise the same complaint again next month.", "correct": false, "explanation": "Whether the stakeholder notices is secondary to the effect on everyone else's summaries."}, {"letter": "D", "text": "One unrepresentative complaint now shapes output that colleagues rely on for other purposes.", "correct": true, "explanation": "One person's preference has been generalised into a rule for everyone, and the colleagues who needed the detail will only discover the change when their own work suffers."}], "correct": "D", "select": 1}, {"id": "a7-028", "domain": 7, "task_id": "7.3", "objective": "Optimize workflows for efficiency and effectiveness", "situation": "When an answer is not right, a coordinator's habit is to ask for another version repeatedly until one looks acceptable.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "Successive versions may differ in length, which makes the final document harder to format well.", "correct": false, "explanation": "Varying length is a formatting nuisance that is easily fixed by asking for a length."}, {"letter": "B", "text": "She spends the allowance on one unclear request and chooses by appearance rather than accuracy.", "correct": true, "explanation": "Regeneration samples the same under-specified request again and again, so the selection is made on how the answer reads rather than on whether it is right, at a cost in allowance each time."}, {"letter": "C", "text": "Repeated attempts leave a long conversation that is more difficult to search through afterwards.", "correct": false, "explanation": "A long conversation is harder to search, but that is a side effect rather than the main risk."}, {"letter": "D", "text": "She may lose an earlier version that was better than the one she eventually decides to keep.", "correct": false, "explanation": "Losing a good earlier draft is annoying, and it is avoidable by copying versions out as she goes."}], "correct": "B", "select": 1}, {"id": "a7-029", "domain": 7, "task_id": "7.1", "objective": "Identify, diagnose, and resolve issues with underperforming prompts or poor outputs", "situation": "A marketing assistant's Claude drafts have become unreliable and she wants to work out why rather than guess.", "question": "Which sequence of steps is correct?", "options": [{"letter": "A", "text": "Add the detail the request was missing; reproduce the fault in a fresh chat; try the fix on two more real cases; check whether the sources are current", "correct": false, "explanation": "Changing the request first means she is fixing before she knows what is broken, and the source check arrives after she has already concluded."}, {"letter": "B", "text": "Reproduce the fault in a fresh chat; check whether the sources are current; add the detail the request was missing; try the fix on two more real cases", "correct": true, "explanation": "Isolate the fault away from a possibly overloaded conversation, rule out an out-of-date source, then address the request itself, and confirm the fix on other real cases rather than only the one that failed."}, {"letter": "C", "text": "Add the detail the request was missing; check whether the sources are current; reproduce the fault in a fresh chat; try the fix on two more real cases", "correct": false, "explanation": "Editing the request before reproducing the fault means she may be solving a problem the fresh chat would not even have shown."}, {"letter": "D", "text": "Check whether the sources are current; try the fix on two more real cases; add the detail the request was missing; reproduce the fault in a fresh chat", "correct": false, "explanation": "Testing a fix before any fix has been made is out of order, and reproducing the fault last removes the point of reproducing it."}], "correct": "B", "select": 1}, {"id": "a7-030", "domain": 7, "task_id": "7.1", "objective": "Identify, diagnose, and resolve issues with underperforming prompts or poor outputs", "situation": "A communications lead asks Claude for a section of a report and keeps receiving three-sentence answers where she needs several paragraphs.", "question": "What is most likely to help?", "options": [{"letter": "A", "text": "Send the same request once more with added emphasis, so Claude treats the task as more important.", "correct": false, "explanation": "Emphasis alone adds no information about how long or how detailed the section should be."}, {"letter": "B", "text": "Retool the request to state the depth and format she needs, adding the previous exchange.", "correct": true, "explanation": "Brief answers are addressed by retooling the request so the wanted depth and shape are explicit, and by giving the previous prompt and response so Claude continues from where it left off."}, {"letter": "C", "text": "Ask Claude why its answers are short, and use its explanation as the basis for her next request.", "correct": false, "explanation": "Claude's account of its own behaviour is speculation and does not tell her what depth she needs."}, {"letter": "D", "text": "Break the section into ten separate questions and paste the short answers together afterwards.", "correct": false, "explanation": "Ten fragments produce a stitched-together section with repetition and gaps between the pieces."}], "correct": "B", "select": 1}, {"id": "a7-031", "domain": 7, "task_id": "7.1", "objective": "Identify, diagnose, and resolve issues with underperforming prompts or poor outputs", "situation": "A grants manager has kept one conversation running for a fortnight across two tender submissions. Claude now blends in details from the earlier tender and contradicts what it said days ago. Which two responses address the actual cause?", "question": "Select two.", "options": [{"letter": "A", "text": "Move the background that stays true across tasks into a project, so it need not be repeated.", "correct": true, "explanation": "for an independent reason: it stops the problem returning, by giving durable context a home outside any single conversation."}, {"letter": "B", "text": "Begin a fresh conversation containing only the material that the current piece of work needs.", "correct": true, "explanation": "A conversation this long has accumulated superseded material, and a clean start removes what is confusing the answers."}, {"letter": "C", "text": "Repeat the correction every single time details from the earlier tender reappear in the answers.", "correct": false, "explanation": "Fixing each recurrence by hand treats the symptom and leaves the overloaded conversation exactly as it was."}, {"letter": "D", "text": "Ask Claude to read the whole conversation again from the very top before it answers each question.", "correct": false, "explanation": "Re-reading the conversation reinforces the mixed-up history rather than clearing it, and consumes more allowance."}, {"letter": "E", "text": "Paste the previous answer back in each morning to remind Claude of what it said before.", "correct": false, "explanation": "Adding the earlier output back in makes the conversation longer still, which is the condition causing the drift."}], "correct": ["A", "B"], "select": 2}, {"id": "a7-032", "domain": 7, "task_id": "7.2", "objective": "Adjust approach based on feedback and results", "situation": "Feedback on a customer-reply project shows that Claude's replies are accurate but consistently more formal than the audience expects.", "question": "What is the most effective adjustment?", "options": [{"letter": "A", "text": "Correct the tone by hand in each reply, since tone is a matter of taste that varies with the customer.", "correct": false, "explanation": "Hand-editing every reply re-solves the same problem daily and leaves the default unchanged."}, {"letter": "B", "text": "Add the audience and wanted tone to the project instructions, with one approved reply as an example.", "correct": true, "explanation": "The fault is consistent, so it belongs in the durable configuration, and one approved example conveys the register far more precisely than an adjective would."}, {"letter": "C", "text": "Change to a faster model, which tends to produce shorter and less formal customer-facing writing.", "correct": false, "explanation": "Speed and tone are unrelated, and model choice is not how register is specified."}, {"letter": "D", "text": "Ask Claude to rewrite each reply less formally as a second step after the first version is produced.", "correct": false, "explanation": "A second rewriting step doubles the work for a fault that could be prevented at the outset."}], "correct": "B", "select": 1}, {"id": "a7-033", "domain": 7, "task_id": "7.2", "objective": "Adjust approach based on feedback and results", "situation": "An assistant notices that Claude's drafts in a shared project have become weaker since a teammate edited the project instructions last week.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Stop using the project and go back to writing each request from scratch in ordinary conversations.", "correct": false, "explanation": "Abandoning the project discards the shared context instead of fixing an edit that can be reverted."}, {"letter": "B", "text": "Ask Claude to review the project instructions and rewrite them so that the drafts improve again.", "correct": false, "explanation": "Claude cannot know which wording produced last month's better drafts, so this is guesswork."}, {"letter": "C", "text": "Look at what changed in the instructions and retry a recent task with the old wording restored.", "correct": true, "explanation": "A change in results that follows a known change in configuration is best tested by reverting that change on a real task, which confirms or rules out the cause in one step."}, {"letter": "D", "text": "Add her own extra instructions on top, so the parts she depends on are stated again more firmly.", "correct": false, "explanation": "Layering instructions on top may mask the conflict while leaving the contradictory wording in place."}], "correct": "C", "select": 1}, {"id": "a7-034", "domain": 7, "task_id": "7.1", "objective": "Identify, diagnose, and resolve issues with underperforming prompts or poor outputs", "situation": "An analyst is unhappy with a Claude output and wants to diagnose the cause properly rather than guess.", "question": "Which sequence is correct?", "options": [{"letter": "A", "text": "Compare the output with what the request actually asked for; check whether the source material was ever supplied; rewrite the request with the missing specifics; consider a different model or a fresh chat.", "correct": true, "explanation": "Diagnosis comes before action: establish what is actually missing, check whether the material was ever provided, fix the request, and only then change the model or the conversation."}, {"letter": "B", "text": "Consider a different model or a fresh chat; rewrite the request with the missing specifics; compare the output with what the request actually asked for; check whether the source material was ever supplied.", "correct": false, "explanation": "Changing the model first treats the most expensive lever as the first resort, before anyone knows what is wrong."}, {"letter": "C", "text": "Check whether the source material was ever supplied; consider a different model or a fresh chat; compare the output with what the request actually asked for; rewrite the request with the missing specifics.", "correct": false, "explanation": "Checking the source material first is reasonable, but switching models before reading the output leaves the diagnosis half done."}, {"letter": "D", "text": "Rewrite the request with the missing specifics; consider a different model or a fresh chat; check whether the source material was ever supplied; compare the output with what the request actually asked for.", "correct": false, "explanation": "Rewriting before comparing means guessing at what was missing rather than establishing it."}], "correct": "A", "select": 1}, {"id": "a7-035", "domain": 7, "task_id": "7.3", "objective": "Optimize workflows for efficiency and effectiveness", "situation": "A team wants to turn a procedure that one colleague performs well into something anyone in the team can run.", "question": "Which sequence is correct?", "options": [{"letter": "A", "text": "Run the steps on a second case to confirm they generalise; capture them as a reusable skill; write down the steps that worked on a real case; have a colleague run it unaided and fix what confused them.", "correct": false, "explanation": "The steps cannot be run on a second case before they have been written down at all."}, {"letter": "B", "text": "Capture the steps as a reusable skill; have a colleague run it unaided and fix what confused them; write down the steps that worked on a real case; run the same steps on a second case to confirm they generalise.", "correct": false, "explanation": "Packaging the procedure before it has been written down or tested captures whatever the author happened to remember."}, {"letter": "C", "text": "Have a colleague run the procedure unaided and fix what confused them; write down the steps that worked on a real case; capture them as a reusable skill; run it on a second case to confirm it generalises.", "correct": false, "explanation": "Handing an uncaptured procedure to a colleague first asks them to reconstruct it from nothing."}, {"letter": "D", "text": "Write down the steps that worked on a real case; run the same steps on a second case to confirm they generalise; capture them as a reusable skill; have a colleague run it unaided and fix what confused them.", "correct": true, "explanation": "Capture what worked, prove it generalises beyond the first case, make it reusable, then let a colleague test it without help, which is the only way to find the assumptions the author left out."}], "correct": "D", "select": 1}, {"id": "a7-036", "domain": 7, "task_id": "7.1", "objective": "Identify, diagnose, and resolve issues with underperforming prompts or poor outputs", "situation": "A team building a classification prompt has zero examples in the prompt, only a description of the categories. Outputs are inconsistent on edge cases, such as tickets that could reasonably fit two categories.", "question": "What is the most effective next step to reduce this inconsistency?", "options": [{"letter": "A", "text": "Add three to five diverse, relevant examples, including edge cases, wrapped in example tags so Claude can mirror the intended pattern.", "correct": true, "explanation": "Anthropic's guidance is that a few well-crafted, diverse examples (few-shot/multishot prompting) are one of the most reliable ways to steer output consistency, and including edge cases in examples helps Claude generalize to ambiguous inputs."}, {"letter": "B", "text": "Instruct Claude to always pick the first matching category alphabetically whenever a ticket could reasonably fit more than one category.", "correct": false, "explanation": "An arbitrary tie-breaking rule like alphabetical order does not reflect the actual decision criteria and can produce systematically wrong categorizations for ambiguous tickets."}, {"letter": "C", "text": "Increase the total number of categories so that ambiguous tickets have a narrower, more specific category to fall into.", "correct": false, "explanation": "Adding more categories increases the classification surface and is likely to increase ambiguity rather than resolve the existing edge cases."}, {"letter": "D", "text": "Remove the existing category descriptions entirely and let Claude infer the categories purely from ticket content.", "correct": false, "explanation": "Removing the category descriptions removes information Claude needs and would likely worsen consistency rather than improve it."}], "correct": "A", "select": 1}, {"id": "a7-037", "domain": 7, "task_id": "7.2", "objective": "Adjust approach based on feedback and results", "situation": "Colleagues report that Claude-assisted replies to invoice queries are accurate but that the whole process now takes longer than it did before.", "question": "What is the most appropriate response?", "options": [{"letter": "A", "text": "Accept the delay, because accuracy matters more than speed in every case", "correct": false, "explanation": "Accepting the delay without understanding it forfeits most of the benefit the change was meant to deliver."}, {"letter": "B", "text": "Ask Claude to reply faster by giving it much shorter written instructions", "correct": false, "explanation": "Shorter instructions are likely to reduce quality, and instruction length was probably not where the time went."}, {"letter": "C", "text": "Look at where the extra time goes; the fault may lie in the steps around it", "correct": true, "explanation": "The result is telling the team that the drafting improved and something else got worse. Finding which step absorbed the time is what turns that feedback into a usable adjustment."}, {"letter": "D", "text": "Remove the checking step, since accurate output no longer needs to be reviewed", "correct": false, "explanation": "Dropping the check on the strength of early accuracy removes the safeguard just as volume grows."}], "correct": "C", "select": 1}, {"id": "a7-038", "domain": 7, "task_id": "7.3", "objective": "Optimize workflows for efficiency and effectiveness", "situation": "An HR adviser pastes the same forty-page benefits manual into a new chat for every query, then retypes the same house-style preamble. Which two changes give the biggest gain?", "question": "Select two.", "options": [{"letter": "A", "text": "Put the house-style preamble into that project's instructions instead of retyping it.", "correct": true, "explanation": "for an independent reason: standing direction belongs in instructions, which is a different mechanism from knowledge."}, {"letter": "B", "text": "Keep one very long conversation permanently open so that the manual stays within view.", "correct": false, "explanation": "One endless conversation eventually loses the thread and mixes unrelated queries together."}, {"letter": "C", "text": "Move the queries to a faster model tier so that each answer takes less time to arrive.", "correct": false, "explanation": "A faster tier shortens the wait but leaves the repeated pasting and retyping completely untouched."}, {"letter": "D", "text": "Have Claude condense the manual and then paste that shorter summary in each time instead.", "correct": false, "explanation": "A condensed manual still gets pasted every time and quietly drops detail some queries will need."}, {"letter": "E", "text": "Upload the manual once as the knowledge of a project used for all of these queries.", "correct": true, "explanation": "Uploading the manual once as project knowledge removes the repeated paste entirely, for every chat in that project."}], "correct": ["A", "E"], "select": 2}, {"id": "a7-039", "domain": 7, "task_id": "7.1", "objective": "Identify, diagnose, and resolve issues with underperforming prompts or poor outputs", "situation": "A facilities project still has last year's contractor safety standard uploaded as project knowledge, and Claude keeps quoting its old inspection interval with complete confidence. Which two actions actually resolve this?", "question": "Select two.", "options": [{"letter": "A", "text": "Move the project onto a more capable model tier, which will state the inspection interval more accurately.", "correct": false, "explanation": "A stronger model reads the same outdated file and repeats the same wrong interval just as confidently."}, {"letter": "B", "text": "Replace the uploaded file with the current standard so the outdated interval cannot be retrieved again.", "correct": true, "explanation": "The knowledge source is wrong, not the model, so nothing improves until the superseded document is replaced."}, {"letter": "C", "text": "Add an instruction telling Claude to take particular care whenever any inspection interval is mentioned.", "correct": false, "explanation": "An instruction to be careful cannot compensate for a source that is authoritative-looking and out of date."}, {"letter": "D", "text": "Open a fresh conversation and put the same question to Claude once again, this time much more firmly worded.", "correct": false, "explanation": "Rewording or insisting changes nothing when the retrieved document itself contains the wrong number."}, {"letter": "E", "text": "Re-check the inspection dates already issued on the strength of answers quoting the superseded figure.", "correct": true, "explanation": "for a separate reason: fixing the source does not undo the decisions already taken on bad information, which still need remediation."}], "correct": ["B", "E"], "select": 2}, {"id": "a7-040", "domain": 7, "task_id": "7.3", "objective": "Optimize workflows for efficiency and effectiveness", "situation": "An analyst begins every chat in her weekly reporting project by typing the same six lines about house style, the sections required and how figures should be rounded.", "question": "What is the most effective first change?", "options": [{"letter": "A", "text": "Ask Claude at the start of each chat to recall the style used in her previous conversations.", "correct": false, "explanation": "Asking Claude to recall earlier style is unreliable and does not make the requirement durable."}, {"letter": "B", "text": "Keep the six lines in a note and paste them at the start of each chat so nothing gets forgotten.", "correct": false, "explanation": "A note reduces typing but still relies on her pasting it correctly every single time."}, {"letter": "C", "text": "Upload a document holding the six lines and point Claude at it whenever she opens a new chat.", "correct": false, "explanation": "Uploaded documents are background knowledge; a standing rule about how to work belongs in the instructions."}, {"letter": "D", "text": "Move those standing lines into the project's instructions so every chat there starts with them.", "correct": true, "explanation": "A requirement repeated in every chat of the same project belongs in the project instructions, where it applies automatically and stops depending on her remembering it."}], "correct": "D", "select": 1}, {"id": "a7-041", "domain": 7, "task_id": "7.2", "objective": "Adjust approach based on feedback and results", "situation": "A team reviews its Claude-assisted arrangement using only the straightforward cases, because the difficult ones take much longer to check.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The straightforward cases will be over-checked and the team will lose time", "correct": false, "explanation": "Redundant checking of easy cases costs time; it does not create a false picture of how the arrangement performs."}, {"letter": "B", "text": "The difficult cases will take even longer to check at the following review", "correct": false, "explanation": "Checking effort may grow, but that is a workload issue rather than the reason the conclusion is unsafe."}, {"letter": "C", "text": "The reviewers will disagree about which cases counted as straightforward", "correct": false, "explanation": "Disagreement about classification is a process wrinkle that would surface and be settled quickly."}, {"letter": "D", "text": "The review will report success on exactly the cases least likely to fail", "correct": true, "explanation": "Results gathered only from easy cases confirm what was never in doubt. The arrangement is then trusted across work it has never been shown to handle."}], "correct": "D", "select": 1}, {"id": "a7-042", "domain": 7, "task_id": "7.2", "objective": "Adjust approach based on feedback and results", "situation": "A team judges its Claude-assisted research notes purely by whether colleagues say they enjoy reading them.", "question": "What is the greatest risk in using this as the measure?", "options": [{"letter": "A", "text": "Positive ratings may encourage the team to produce more notes than colleagues have time to read.", "correct": false, "explanation": "Producing more notes is a workload effect rather than a risk to the reliability of the notes."}, {"letter": "B", "text": "Fluent, confident notes can be well liked while carrying errors nobody checked against sources.", "correct": true, "explanation": "Reader satisfaction measures how the notes read, not whether they are right, so a confident and wrong note scores as well as a correct one and the error is adopted."}, {"letter": "C", "text": "Colleagues who read quickly may rate the notes higher than colleagues who read them thoroughly.", "correct": false, "explanation": "Variation between readers adds noise to the measure, but the deeper problem is what it measures at all."}, {"letter": "D", "text": "Preferences drift over time, which makes this quarter's ratings hard to compare with last quarter's.", "correct": false, "explanation": "Drift makes comparison harder over time; it does not explain why accuracy goes unmeasured."}], "correct": "B", "select": 1}, {"id": "a7-043", "domain": 7, "task_id": "7.3", "objective": "Optimize workflows for efficiency and effectiveness", "situation": "An analyst wants to get more done inside her weekly allowance without the quality of her work dropping.", "question": "Which sequence of working habits is correct?", "options": [{"letter": "A", "text": "Put the request and the relevant material in the opening message; work through follow-ups that refer back to it; open a new chat for the next unrelated task; decide what she needs before starting.", "correct": false, "explanation": "Everything here is done before she has decided what she needs, so the opening message is assembled blind."}, {"letter": "B", "text": "Open a new chat for the next unrelated task; decide what she needs before starting; work through follow-ups that refer back to it; put the request and the relevant material in the opening message.", "correct": false, "explanation": "Opening a chat for the next task before the current one has been started puts the last step first."}, {"letter": "C", "text": "Decide what she needs before starting; put the request and the relevant material in the opening message; work through follow-ups that refer back to it; open a new chat for the next unrelated task.", "correct": true, "explanation": "Planning before starting, front-loading the material in one well-specified message, then building on it, is what keeps the exchange short; unrelated work starts fresh rather than dragging old context along."}, {"letter": "D", "text": "Work through follow-ups that refer back to it; open a new chat for the next unrelated task; decide what she needs before starting; put the request and the relevant material in the opening message.", "correct": false, "explanation": "Following up before the request has even been made, and deciding what is needed last, inverts the whole habit."}], "correct": "C", "select": 1}, {"id": "a7-044", "domain": 7, "task_id": "7.1", "objective": "Identify, diagnose, and resolve issues with underperforming prompts or poor outputs", "situation": "A coding assistant prompt says \"Can you suggest some changes to improve this function?\" The team wants Claude to actually implement the changes in the file rather than describe them in prose, but Claude keeps responding with suggestions only.", "question": "What is the most direct fix?", "options": [{"letter": "A", "text": "Rephrase the request as a direct instruction to make the edit, such as telling Claude to change the function to improve its performance.", "correct": true, "explanation": "Anthropic's guidance notes that phrasing like \"can you suggest changes\" leads Claude to provide suggestions rather than implement them; being explicit with a direct instruction, such as telling Claude to make the change, is the documented fix for getting Claude to take action."}, {"letter": "B", "text": "Move the question into the system prompt instead of the user message, while keeping the exact same original suggestive wording as before.", "correct": false, "explanation": "Moving the same suggestive wording to the system prompt does not change its meaning; the ambiguity about suggesting versus acting remains."}, {"letter": "C", "text": "Send a follow-up message that just says \"good suggestions\" so that Claude infers approval to proceed with implementing them.", "correct": false, "explanation": "A vague follow-up acknowledgment doesn't constitute an explicit instruction to implement changes and is not the documented technique for triggering action."}, {"letter": "D", "text": "Increase the effort or thinking configuration considerably so Claude reasons for longer before answering the exact same phrasing.", "correct": false, "explanation": "More reasoning effort affects how thoroughly Claude thinks through a response, not whether ambiguous suggestive phrasing gets interpreted as a request for action."}], "correct": "A", "select": 1}, {"id": "a7-045", "domain": 7, "task_id": "7.2", "objective": "Adjust approach based on feedback and results", "situation": "After one complaint about a single report, a team changes the shared instruction, re-runs that same report, is satisfied with the result, and rolls the change out to everyone.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The changed instruction cannot be reversed once other people begin using it", "correct": false, "explanation": "Instructions can be changed back at any time, so reversibility is not the exposure here."}, {"letter": "B", "text": "The person who complained will not be told the change was made for them", "correct": false, "explanation": "Failing to close the loop with the complainant is a courtesy issue, not the reason the rollout is unsafe."}, {"letter": "C", "text": "The change may fix that one report and quietly damage every other kind", "correct": true, "explanation": "Testing a fix only on the case that failed shows that the case now passes; it says nothing about the reports that were already fine. Trying it on a few different real cases is what would reveal the damage."}, {"letter": "D", "text": "Re-running one report will consume more of the team's allowance than usual", "correct": false, "explanation": "The cost of re-running one report is trivial and unrelated to whether the change was sound."}], "correct": "C", "select": 1}, {"id": "a7-046", "domain": 7, "task_id": "7.1", "objective": "Identify, diagnose, and resolve issues with underperforming prompts or poor outputs", "situation": "A researcher asks about a regulatory change announced last month. The answer is fluent and confident but wrong on the specifics, and the project contains no document about the change.", "question": "What is the most likely cause and the right fix?", "options": [{"letter": "A", "text": "Claude has no dependable source for it, so she should supply the announcement or use Research on it.", "correct": true, "explanation": "Nothing in the conversation or the project describes the change, so the gap is in the available information; supplying the source, or using Research so claims come with citations, is what closes it."}, {"letter": "B", "text": "The conversation has grown too long, so she should begin again and put the question in a fresh chat.", "correct": false, "explanation": "Conversation length would produce drift and contradiction, not confident detail about an unavailable topic."}, {"letter": "C", "text": "The model tier is too small, so she should put the same question to a model with deeper reasoning.", "correct": false, "explanation": "Reasoning depth does not create knowledge of an announcement that was never provided."}, {"letter": "D", "text": "Her request was too vague, so she should ask the same question again with a stricter output format set.", "correct": false, "explanation": "The question was specific enough; format instructions cannot supply facts that were never available."}], "correct": "A", "select": 1}, {"id": "a7-047", "domain": 7, "task_id": "7.2", "objective": "Adjust approach based on feedback and results", "situation": "A team has drafted a revised instruction that seems to fix a recurring problem in a shared project.", "question": "Which sequence best confirms and adopts it?", "options": [{"letter": "A", "text": "Try the revision on two recent real cases; compare the results with the old output on those cases; update the shared project instructions; tell the team what changed and why.", "correct": true, "explanation": "The revision is tested on real cases and compared against what it replaces before it becomes the shared default, and colleagues are told once the change is actually in place."}, {"letter": "B", "text": "Update the shared project instructions; tell the team what changed and why; try the revision on two recent real cases; compare the results with the old output on those cases.", "correct": false, "explanation": "Rolling the change out to everyone before testing it makes the whole team the experiment."}, {"letter": "C", "text": "Compare the results with the old output on those cases; try the revision on two recent real cases; update the shared project instructions; tell the team what changed and why.", "correct": false, "explanation": "Comparing results before producing them is not possible, so the ordering cannot work."}, {"letter": "D", "text": "Tell the team what changed and why; update the shared project instructions; compare the results with the old output on those cases; try the revision on two recent real cases.", "correct": false, "explanation": "Announcing a change before it exists, and testing it last, leaves colleagues acting on something unverified."}], "correct": "A", "select": 1}, {"id": "a7-048", "domain": 7, "task_id": "7.2", "objective": "Adjust approach based on feedback and results", "situation": "A small operations team wants to know whether a revised set of instructions genuinely improved results.", "question": "What is the most reliable check?", "options": [{"letter": "A", "text": "Ask Claude to compare the two sets of instructions and state which one ought to perform better.", "correct": false, "explanation": "Claude can comment on the wording but cannot know how either version performs on the team's real work."}, {"letter": "B", "text": "Ask the team at the end of the month whether things have felt better since the change was made.", "correct": false, "explanation": "A recollection a month later is influenced by everything else that happened in the meantime."}, {"letter": "C", "text": "Try the new instructions on one fresh request and adopt them if that single result looks better.", "correct": false, "explanation": "A single new request confounds the change with whatever was distinctive about that request."}, {"letter": "D", "text": "Run the old and new instructions on the same recent cases and compare against agreed criteria.", "correct": true, "explanation": "Holding the cases and the criteria constant is what makes the comparison mean something; only the instructions differ, so any difference in quality is attributable."}], "correct": "D", "select": 1}, {"id": "a7-049", "domain": 7, "task_id": "7.1", "objective": "Identify, diagnose, and resolve issues with underperforming prompts or poor outputs", "situation": "An operations specialist keeps a Claude project for vendor onboarding, with the onboarding policy uploaded as project knowledge. Claude now states the approval threshold confidently, but it is last year's figure, and rewording the question changes nothing.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Move to a more capable model, which should reason more carefully about the approval rules.", "correct": false, "explanation": "A stronger model reading an out-of-date document will still report the out-of-date figure."}, {"letter": "B", "text": "Start a new conversation inside the project and ask the same question in a different way.", "correct": false, "explanation": "A fresh conversation still draws on the same project knowledge, so the same figure comes back."}, {"letter": "C", "text": "Add an instruction telling Claude to be accurate and to double-check any figure it reports.", "correct": false, "explanation": "Instructions to be accurate cannot correct a number that the supplied source states incorrectly."}, {"letter": "D", "text": "Check whether the policy document in the project knowledge is still the current version.", "correct": true, "explanation": "A confidently wrong figure that survives rewording points at the source rather than the prompt, and a superseded document in project knowledge produces exactly this pattern."}], "correct": "D", "select": 1}, {"id": "a7-050", "domain": 7, "task_id": "7.1", "objective": "Identify, diagnose, and resolve issues with underperforming prompts or poor outputs", "situation": "A team debugging a poorly performing prompt shows it, with no additional context, to a colleague unfamiliar with the project and asks them to follow it exactly. The colleague is confused about what output is expected.", "question": "What does this diagnostic exercise indicate, per Anthropic's prompting guidance?", "options": [{"letter": "A", "text": "The prompt lacks clarity and specificity, because a reader with minimal context could not follow it and Claude likely cannot either.", "correct": true, "explanation": "Anthropic's golden rule of clear prompting is to show the prompt to a colleague with minimal context on the task and ask them to follow it; if they would be confused, Claude will be too, so the fix is a clearer and more specific prompt."}, {"letter": "B", "text": "The colleague simply needs more subject-matter training on this project before they can judge whether the prompt itself is clear enough.", "correct": false, "explanation": "The diagnostic deliberately uses a reader with minimal context on the task, so training that reader up would remove the very gap the exercise is designed to expose."}, {"letter": "C", "text": "The team should move the work to a more capable model tier, since a confusing prompt is a model limitation rather than a prompting problem.", "correct": false, "explanation": "A confused human reader points at an under-specified prompt, and the documented response is to add the missing clarity and specificity. Moving to a stronger model tier does not supply context that the prompt never contained."}, {"letter": "D", "text": "The exercise is inconclusive, because whether a person can follow a prompt says nothing about how Claude will interpret the same text.", "correct": false, "explanation": "Anthropic's prompting guidance uses human comprehension as the signal precisely because it predicts model confusion, so the exercise is informative rather than inconclusive."}], "correct": "A", "select": 1}, {"id": "a7-051", "domain": 7, "task_id": "7.2", "objective": "Adjust approach based on feedback and results", "situation": "Several engineers tell a site coordinator that Claude's shift-handover digests 'skip the open risks', but nobody says which risks or which shifts. Which two steps should the coordinator take first?", "question": "Select two.", "options": [{"letter": "A", "text": "Switch the digests into a table format, since open risks may then be easier to spot in a table.", "correct": false, "explanation": "A format change may help presentation, but it cannot add a risk the digest never captured."}, {"letter": "B", "text": "Rewrite the shared project instructions straight away to stress that open risks matter most of all.", "correct": false, "explanation": "Changing shared instructions before the cause is known risks fixing the wrong thing for everyone at once."}, {"letter": "C", "text": "Ask everyone to score every future digest out of ten so that a trend can be tracked over time.", "correct": false, "explanation": "A score captures sentiment but still never says which risks went missing, or why they went missing."}, {"letter": "D", "text": "Compare those digests against the shift logs to see exactly what was left out of them.", "correct": true, "explanation": "on separate grounds: only the source shows whether the risk was absent from the log, buried, or genuinely dropped."}, {"letter": "E", "text": "Ask for two digests where an open risk was missed, with the missing risk actually named.", "correct": true, "explanation": "Vague feedback cannot be acted on; specific failing examples turn an impression into something diagnosable."}], "correct": ["D", "E"], "select": 2}, {"id": "a7-052", "domain": 7, "task_id": "7.3", "objective": "Optimize workflows for efficiency and effectiveness", "situation": "A finance analyst follows the same twelve-step month-end review whenever the figures land, and wants Claude to carry it out the same way whichever conversation she happens to be working in.", "question": "What fits this need best?", "options": [{"letter": "A", "text": "Capture the review as a reusable skill, so the same procedure is available wherever she works.", "correct": true, "explanation": "A repeatable multi-step procedure that should behave the same way everywhere is what a skill captures; skills provide specialised procedures that work across Claude rather than in one workspace."}, {"letter": "B", "text": "Upload a document describing the twelve steps and point Claude to it at the start of each review.", "correct": false, "explanation": "A document she must point to each time relies on her remembering, and describes the steps rather than performing them."}, {"letter": "C", "text": "Keep one long month-end conversation and continue it each month so the steps stay in the history.", "correct": false, "explanation": "One long-running conversation becomes unwieldy and ties the procedure to a single chat history."}, {"letter": "D", "text": "Add the twelve steps to the instructions of the project where she keeps her month-end files.", "correct": false, "explanation": "Project instructions apply only inside that project, so the procedure would not follow her elsewhere."}], "correct": "A", "select": 1}, {"id": "a7-053", "domain": 7, "task_id": "7.2", "objective": "Adjust approach based on feedback and results", "situation": "An adjustment to a shared project's instructions has clearly improved results across several weeks.", "question": "What most reliably keeps that improvement in place for the whole team?", "options": [{"letter": "A", "text": "Telling the colleague who reported the original problem that it is now fixed", "correct": false, "explanation": "Closing the loop with one colleague is good practice but does nothing to preserve the change itself."}, {"letter": "B", "text": "Reminding people in each chat to phrase their requests the improved way", "correct": false, "explanation": "Relying on people to remember a phrasing in every chat re-solves the problem by hand each time."}, {"letter": "C", "text": "Keeping a copy of the improved output to show anyone who asks about it", "correct": false, "explanation": "A sample of good output demonstrates the result without capturing what produced it."}, {"letter": "D", "text": "Recording the change and the reason for it where the whole team can see", "correct": true, "explanation": "A fix that is written down, with its reason, survives staff changes and stops someone unpicking it later because they cannot see why it is there. That is what makes the adjustment durable."}], "correct": "D", "select": 1}, {"id": "a7-054", "domain": 7, "task_id": "7.1", "objective": "Identify, diagnose, and resolve issues with underperforming prompts or poor outputs", "situation": "A programme officer notices that Claude's answers in a long-running chat now ignore constraints she stated near the start.", "question": "Which action best addresses the cause?", "options": [{"letter": "A", "text": "Start a fresh chat and restate the constraints that must hold throughout", "correct": true, "explanation": "The conversation itself has become the problem: it is long enough that early material no longer carries. A fresh chat carrying the constraints forward restores them as live instructions."}, {"letter": "B", "text": "Repeat the constraints once more at the end of the current conversation", "correct": false, "explanation": "Adding the constraints again extends the same overloaded conversation, so the next few answers may improve before the drift returns."}, {"letter": "C", "text": "Ask Claude to confirm that it still remembers everything said earlier on", "correct": false, "explanation": "A reassurance that it remembers is not evidence, and it does nothing about the length that caused the drift."}, {"letter": "D", "text": "Ask for shorter answers so that the conversation grows more slowly now", "correct": false, "explanation": "Shorter answers slow the growth of the conversation but leave the material already lost from view exactly where it is."}], "correct": "A", "select": 1}, {"id": "a7-055", "domain": 7, "task_id": "7.2", "objective": "Adjust approach based on feedback and results", "situation": "After poor feedback, a team changes the wording, the examples, the required format, the model and the source documents all at once. The next outputs are better.", "question": "What is the greatest risk?", "options": [{"letter": "A", "text": "The improvement may take several days to become consistent across everyone working in the project.", "correct": false, "explanation": "A delay in consistency is a minor operational point beside not knowing what caused the change."}, {"letter": "B", "text": "Colleagues may find the new format harder to read than the one they had grown accustomed to.", "correct": false, "explanation": "Readability of the format is a preference that can be adjusted once the cause is understood."}, {"letter": "C", "text": "The revised instructions may be long enough that new joiners take a while to understand them.", "correct": false, "explanation": "Longer instructions are a maintenance cost, not the reason the improvement cannot be explained."}, {"letter": "D", "text": "They cannot tell which change helped, so they cannot keep the useful one or undo a harmful one.", "correct": true, "explanation": "Five simultaneous changes and one outcome give no information about cause, so the team cannot simplify, cannot reverse a change that hurt, and cannot repeat the success elsewhere."}], "correct": "D", "select": 1}, {"id": "a7-056", "domain": 7, "task_id": "7.1", "objective": "Identify, diagnose, and resolve issues with underperforming prompts or poor outputs", "situation": "A prompt asks Claude to solve multi-step word problems, but final answers are frequently arithmetically wrong even though the described approach sounds reasonable.", "question": "Which change is most likely to reduce these errors?", "options": [{"letter": "A", "text": "Shorten the word problem's wording considerably so there is simply much less text overall for Claude to process carefully before it answers.", "correct": false, "explanation": "Shortening the wording doesn't address the missing step-by-step reasoning that catches arithmetic mistakes, and could remove necessary problem details."}, {"letter": "B", "text": "Repeat the exact same word problem twice within the prompt so Claude effectively gets two separate chances to parse it correctly.", "correct": false, "explanation": "Duplicating the problem statement does not add a reasoning process and is not a documented technique for reducing calculation errors."}, {"letter": "C", "text": "Have Claude work through the problem step by step before the final answer, separating its reasoning from the answer using structured tags.", "correct": true, "explanation": "Encouraging step-by-step reasoning (chain-of-thought), with structured tags to separate reasoning from the final answer, is Anthropic's documented technique for catching errors reliably, especially in coding and math tasks."}, {"letter": "D", "text": "Ask Claude to skip any intermediate explanation entirely and output only the final numeric answer so there is less room for error.", "correct": false, "explanation": "Removing the reasoning step is the opposite of the documented fix; skipping intermediate steps tends to increase arithmetic errors on multi-step problems."}], "correct": "C", "select": 1}, {"id": "a7-057", "domain": 7, "task_id": "7.2", "objective": "Adjust approach based on feedback and results", "situation": "Objections to a Claude-assisted arrangement come mostly from colleagues who have not tried it themselves.", "question": "What is the most appropriate response?", "options": [{"letter": "A", "text": "Have some of them run it on their own real work and then gather their views", "correct": true, "explanation": "Feedback based on real use is the only kind that can be acted on, and it converts an argument about expectations into evidence about results."}, {"letter": "B", "text": "Discount the objections, since only current users can comment on the results", "correct": false, "explanation": "Dismissing non-users ignores concerns that may be well founded and hardens the opposition."}, {"letter": "C", "text": "Drop the arrangement, because adoption will fail without their support first", "correct": false, "explanation": "Abandoning the arrangement over untested objections discards a change that may be working well."}, {"letter": "D", "text": "Send them the strongest outputs the team produced during its trial fortnight", "correct": false, "explanation": "Showing a curated set of best outputs invites the reasonable suspicion that it is not representative."}], "correct": "A", "select": 1}, {"id": "a7-058", "domain": 7, "task_id": "7.2", "objective": "Adjust approach based on feedback and results", "situation": "Half a team says a Claude-assisted drafting pilot saves hours; the other half says the drafts need so much editing they are not worth it.", "question": "What should the pilot lead do first?", "options": [{"letter": "A", "text": "Extend the pilot by a month so that more people have the chance to form a settled view of it.", "correct": false, "explanation": "More time produces more opinion without a shared standard to interpret any of it against."}, {"letter": "B", "text": "Agree what an acceptable draft looks like, then gather specific examples from both groups.", "correct": true, "explanation": "The two groups may be applying different standards to different work, so a shared definition of an acceptable draft, plus real examples, turns opinion into something that can be acted on."}, {"letter": "C", "text": "Rewrite the shared instructions using the suggestions from the colleagues who were most critical.", "correct": false, "explanation": "Rewriting to satisfy the critics assumes their cases are representative, which nobody has checked."}, {"letter": "D", "text": "Adopt the workflow for the group that likes it and withdraw it from the group that does not.", "correct": false, "explanation": "Splitting the group accepts the disagreement without ever discovering what caused it."}], "correct": "B", "select": 1}, {"id": "a7-059", "domain": 7, "task_id": "7.3", "objective": "Optimize workflows for efficiency and effectiveness", "situation": "An operations lead pastes the same thirty-page service handbook into a new conversation every time a query about it comes in.", "question": "What is the most effective first change?", "options": [{"letter": "A", "text": "Paste only the two or three sections that appear most likely to be relevant to each query.", "correct": false, "explanation": "Choosing sections by hand each time re-introduces the effort and risks omitting the relevant part."}, {"letter": "B", "text": "Put the handbook into a project's knowledge so every chat in that project can draw on it.", "correct": true, "explanation": "Background material that every conversation needs belongs in project knowledge, where it is available to each chat without being supplied again."}, {"letter": "C", "text": "Have Claude summarise the handbook once, then paste that summary at the start of each chat.", "correct": false, "explanation": "A summary loses the detail that specific queries turn on, so answers become less reliable."}, {"letter": "D", "text": "Keep one very long conversation open and ask every handbook query inside that conversation.", "correct": false, "explanation": "One ever-growing conversation eventually loses the thread and consumes more of the allowance."}], "correct": "B", "select": 1}, {"id": "a7-060", "domain": 7, "task_id": "7.3", "objective": "Optimize workflows for efficiency and effectiveness", "situation": "A coordinator has been running a Claude-assisted intake process successfully on his own and now wants twelve colleagues to use it.", "question": "What should he do first?", "options": [{"letter": "A", "text": "Write down the steps that actually worked and run them on two recent real cases to confirm.", "correct": true, "explanation": "What works in one person's hands is often partly undocumented habit; writing the steps down and re-running them on real cases is what turns it into something twelve people can repeat."}, {"letter": "B", "text": "Ask Claude to write a guide from his description of the process and circulate that instead.", "correct": false, "explanation": "A guide written from his description records what he thinks he does, not what actually worked."}, {"letter": "C", "text": "Send colleagues the wording he uses, so each can adapt it to the way they prefer to work.", "correct": false, "explanation": "Twelve personal adaptations produce twelve different results and no shared practice to improve."}, {"letter": "D", "text": "Arrange training for the whole group so everyone starts using the process on the same day.", "correct": false, "explanation": "Training people in a procedure that has not been written down or re-tested spreads the gaps as well."}], "correct": "A", "select": 1}, {"id": "a7-061", "domain": 7, "task_id": "7.1", "objective": "Identify, diagnose, and resolve issues with underperforming prompts or poor outputs", "situation": "A prompt intended to elicit a plain, matter-of-fact answer instead produces a math-heavy explanation formatted with LaTeX-style notation, which breaks downstream rendering in the team's plain-text chat interface.", "question": "What is the most direct fix?", "options": [{"letter": "A", "text": "Ask Claude to respond twice and then manually pick whichever version happens to avoid LaTeX notation on that particular occasion.", "correct": false, "explanation": "Relying on chance across repeated generations is not a reliable or documented technique for controlling output formatting."}, {"letter": "B", "text": "Add an explicit instruction to format the response in plain text only, avoiding LaTeX or markup, using standard characters for math.", "correct": true, "explanation": "Anthropic notes that current models default to LaTeX for mathematical expressions, and documents an explicit instruction — requesting plain text only and specifying standard characters for math — as the fix for teams that prefer plain-text output."}, {"letter": "C", "text": "Rephrase the question to avoid using any numbers, since numeric content is what actually triggers LaTeX formatting in responses.", "correct": false, "explanation": "Avoiding numbers entirely isn't practical for many tasks and doesn't address the underlying formatting default; the fix is to specify the desired format, not avoid the topic."}, {"letter": "D", "text": "Switch to a noticeably smaller model on the assumption that smaller models never default to LaTeX-style formatting for math at all whatsoever.", "correct": false, "explanation": "LaTeX defaulting is not documented as being tied to model size, so switching models is not a reliable or targeted fix for this formatting issue."}], "correct": "B", "select": 1}, {"id": "a7-062", "domain": 7, "task_id": "7.1", "objective": "Identify, diagnose, and resolve issues with underperforming prompts or poor outputs", "situation": "A support team's Claude Project answers policy questions from a handbook uploaded eleven months ago. The handbook was revised twice since, but nobody updated the project. Answers still read as confident and well written.", "question": "What is the greatest risk here?", "options": [{"letter": "A", "text": "The team acts on withdrawn policy because the answers read as authoritative and nothing flags the source as stale", "correct": true, "explanation": "A stale knowledge source is the hardest failure to spot: the output is fluent and confident, so nothing prompts anyone to check, and no amount of rewording the prompt will fix an answer grounded in a withdrawn document."}, {"letter": "B", "text": "The project will refuse to answer any policy question at all until somebody replaces the handbook with its current revision", "correct": false, "explanation": "The project answers from whatever it holds; it does not detect a revision elsewhere and withhold responses."}, {"letter": "C", "text": "Claude will notice the handbook is outdated and will say so before answering, which slows the team down on every request", "correct": false, "explanation": "Claude has no way to know a newer revision exists outside the project, so it cannot warn the team about it."}, {"letter": "D", "text": "The uploaded handbook will be removed automatically once it passes a year, leaving the project with no source to draw on", "correct": false, "explanation": "Uploaded project knowledge does not expire on a timer; it stays until someone changes it."}], "correct": "A", "select": 1}, {"id": "a7-063", "domain": 7, "task_id": "7.1", "objective": "Identify, diagnose, and resolve issues with underperforming prompts or poor outputs", "situation": "A prompt for generating short marketing taglines includes ten examples, all of which happen to be puns. The team is puzzled that every generated tagline is now also a pun, even for products where a pun-based tagline is a poor fit.", "question": "What is the most likely diagnosis?", "options": [{"letter": "A", "text": "The examples are causing Claude to ignore the instructions entirely and rely solely on memorized training data instead.", "correct": false, "explanation": "There's no basis for claiming the examples cause instructions to be ignored in favor of training data; the documented issue with narrow examples is pattern overfitting to the examples themselves."}, {"letter": "B", "text": "Ten examples is too many; Anthropic caps effective examples at two, and anything beyond that causes overfitting.", "correct": false, "explanation": "Anthropic recommends 3-5 examples for best results as a general guideline, not a hard cap of two, and the core problem here is lack of diversity, not raw example count."}, {"letter": "C", "text": "The examples are not diverse enough, so Claude picked up the unintended pattern that taglines must always be puns.", "correct": true, "explanation": "Anthropic's guidance on using examples effectively specifies they should be diverse, varying enough that Claude doesn't pick up unintended patterns; a set of examples that are uniformly puns teaches the model that puns are required, not just one possible style."}, {"letter": "D", "text": "The issue is unrelated to the examples; it stems from an implicit temperature default that favors humorous phrasing.", "correct": false, "explanation": "This misattributes a documented examples-diversity issue to an unrelated sampling parameter with no such documented default behavior."}], "correct": "C", "select": 1}, {"id": "a7-064", "domain": 7, "task_id": "7.2", "objective": "Adjust approach based on feedback and results", "situation": "A payroll supervisor has given Claude the same correction in a shared project every week for a month and wants to stop repeating it. Which two steps make the fix stick?", "question": "Select two.", "options": [{"letter": "A", "text": "Tell the rest of the team about it, so nobody keeps solving the same problem in their own chats.", "correct": true, "explanation": "for a different reason: the fix is only really closed when the practice is shared, otherwise colleagues rediscover it."}, {"letter": "B", "text": "Keep a personal note of the correction and paste it in whenever the problem happens to come back.", "correct": false, "explanation": "A personal note keeps the fix manual and private, so it is re-applied by hand and helps nobody else."}, {"letter": "C", "text": "Get faster at spotting it, and paste the correction at the start of each new chat in the project.", "correct": false, "explanation": "Pasting the correction faster is still re-solving the same problem in every single conversation."}, {"letter": "D", "text": "Write the correction into the project's instructions so that it applies by default from now on.", "correct": true, "explanation": "Project instructions are the durable home for a fix, so it stops depending on somebody remembering it."}, {"letter": "E", "text": "Ask the team to flag it in feedback each time, so the pattern can be tracked more precisely.", "correct": false, "explanation": "Tracking the pattern documents the problem more precisely but does not actually change any output."}], "correct": ["A", "D"], "select": 2}, {"id": "a7-065", "domain": 7, "task_id": "7.3", "objective": "Optimize workflows for efficiency and effectiveness", "situation": "A researcher is doing straightforward drafting work while several connectors and tools she does not need for it remain switched on.", "question": "What is a reasonable step to make routine work leaner?", "options": [{"letter": "A", "text": "Ask Claude in each message to avoid using any connector while she works on the drafting.", "correct": false, "explanation": "An instruction in every message is a per-message reminder rather than a change to the setup."}, {"letter": "B", "text": "Do the drafting inside a separate project that has no documents in its knowledge at all.", "correct": false, "explanation": "An empty project addresses knowledge documents, not the tools and connections that stay enabled."}, {"letter": "C", "text": "Turn off the connectors and tools this task does not need, and switch them back on later.", "correct": true, "explanation": "Turning off tools and connectors that a task does not need is a recognised way to keep routine work efficient, and they can be re-enabled when the next task needs them."}, {"letter": "D", "text": "Remove the connectors from her account permanently, so nothing can call on them by mistake.", "correct": false, "explanation": "Removing them permanently costs her the reconnection work every time the other tasks come round."}], "correct": "C", "select": 1}, {"id": "a7-066", "domain": 7, "task_id": "7.1", "objective": "Identify, diagnose, and resolve issues with underperforming prompts or poor outputs", "situation": "A programme manager has worked inside a single conversation for three days. Claude now blends details from an earlier workstream into its answers and contradicts decisions taken yesterday.", "question": "What is the most effective response?", "options": [{"letter": "A", "text": "Ask Claude to disregard everything discussed before yesterday and carry on in the same conversation.", "correct": false, "explanation": "An instruction to ignore earlier content leaves that content in the conversation, still available to be drawn on."}, {"letter": "B", "text": "Move to a model with deeper reasoning, so the length of the conversation is handled more reliably.", "correct": false, "explanation": "A different model inherits the same crowded conversation and the same contradictions inside it."}, {"letter": "C", "text": "Start a fresh chat with a short recap and the current documents, keeping key material in a project.", "correct": true, "explanation": "The conversation itself is the problem, so the fix is a clean start carrying only what still matters, with the durable material kept somewhere it does not have to be re-established each time."}, {"letter": "D", "text": "Repeat the most recent decisions in every message so the earlier workstream stops shaping the answers.", "correct": false, "explanation": "Restating decisions in every message is laborious and still competes with three days of superseded discussion."}], "correct": "C", "select": 1}, {"id": "a7-067", "domain": 7, "task_id": "7.3", "objective": "Optimize workflows for efficiency and effectiveness", "situation": "A specialist has given Claude the same correction three weeks running in the same project.", "question": "Which sequence best makes that fix stick?", "options": [{"letter": "A", "text": "Confirm it is a standing requirement rather than a one-off; add it to the project instructions; check the next output honours it unprompted; tell colleagues working in the project.", "correct": true, "explanation": "Establish that the correction is genuinely standing, make it durable in the instructions, verify it now works without the reminder, and only then tell the people who share the project."}, {"letter": "B", "text": "Tell colleagues working in the project; check the next output honours it unprompted; confirm it is a standing requirement rather than a one-off; add it to the project instructions.", "correct": false, "explanation": "Announcing the change before it has been made or verified leaves colleagues relying on something that is not there."}, {"letter": "C", "text": "Check the next output honours it unprompted; tell colleagues working in the project; add it to the project instructions; confirm it is a standing requirement rather than a one-off.", "correct": false, "explanation": "Checking that the output honours a rule that has not yet been written cannot tell anyone anything."}, {"letter": "D", "text": "Add it to the project instructions; confirm it is a standing requirement rather than a one-off; tell colleagues working in the project; check the next output honours it unprompted.", "correct": false, "explanation": "Writing it into shared instructions before checking it is a standing rule can impose a one-off preference on everyone."}], "correct": "A", "select": 1}, {"id": "a7-068", "domain": 7, "task_id": "7.3", "objective": "Optimize workflows for efficiency and effectiveness", "situation": "A team's Claude project produces good analysis, but they have to ask for British spelling and a summary table in every single conversation.", "question": "Where do those two requirements belong?", "options": [{"letter": "A", "text": "In the first message of each conversation, since requirements can change from one task to another.", "correct": false, "explanation": "Repeating them in every conversation is the effort they are trying to remove, and it will be forgotten."}, {"letter": "B", "text": "In a document uploaded to the project's knowledge, alongside the reference material already there.", "correct": false, "explanation": "Knowledge documents supply background material; they are not read as directions about how to respond."}, {"letter": "C", "text": "In the project's instructions, where they apply to every conversation started in that project.", "correct": true, "explanation": "These are standing rules for how work in this project should be presented, which is exactly what project instructions are for."}, {"letter": "D", "text": "In a reusable skill, so that the two requirements can be invoked from any conversation at all.", "correct": false, "explanation": "A skill is heavier machinery than two presentation rules need, and these apply only to this project's work."}], "correct": "C", "select": 1}, {"id": "a7-069", "domain": 7, "task_id": "7.2", "objective": "Adjust approach based on feedback and results", "situation": "After a month in use, the only feedback on a Claude-assisted rota digest is \"it's fine\". The lead wants to know whether it is genuinely working.", "question": "What should she do first?", "options": [{"letter": "A", "text": "Ask the team to score the digest out of ten at the end of each week ahead", "correct": false, "explanation": "A weekly score records a mood without saying what to change, and the answer is a month away."}, {"letter": "B", "text": "Ask the users what they still change by hand before they use the digest", "correct": true, "explanation": "What people quietly fix is the most honest signal available. \"It's fine\" often means \"I edit it every week and stopped mentioning it\", and the edits show exactly where the gap is."}, {"letter": "C", "text": "Ask Claude to produce a longer digest and see whether anyone objects to it", "correct": false, "explanation": "Deliberately degrading the digest to provoke a reaction wastes a month and annoys the users."}, {"letter": "D", "text": "Take the absence of complaints as confirmation and extend it to two teams", "correct": false, "explanation": "Silence is not evidence of success, and extending on that basis spreads an unverified arrangement."}], "correct": "B", "select": 1}, {"id": "a7-070", "domain": 7, "task_id": "7.1", "objective": "Identify, diagnose, and resolve issues with underperforming prompts or poor outputs", "situation": "A prompt for a tone-sensitive customer email generator gives Claude no persona or role, just the raw task instructions. Emails come back with an inconsistent, generic tone that doesn't match the brand voice.", "question": "What is a lightweight change likely to help?", "options": [{"letter": "A", "text": "Increase the max_tokens limit so Claude has more space to naturally settle into a consistent tone over a longer response.", "correct": false, "explanation": "A larger token budget affects available response length, not tone consistency, and is not the documented lever for brand voice."}, {"letter": "B", "text": "Add a system prompt sentence establishing Claude's role, such as being a customer service specialist for the brand.", "correct": true, "explanation": "Anthropic notes that setting a role in the system prompt focuses Claude's behavior and tone for a given use case, and that even a single sentence establishing a role makes a noticeable difference in output consistency."}, {"letter": "C", "text": "Remove the existing task instructions entirely and rely only on examples to communicate the desired tone.", "correct": false, "explanation": "Removing the task instructions would leave Claude without clear direction on what to actually write, which is a separate problem from tone and likely to make output worse."}, {"letter": "D", "text": "Ask Claude to write each email in a different language and then translate it back, hoping tone carries over more consistently.", "correct": false, "explanation": "Round-tripping through translation is not a documented technique for tone control and introduces unnecessary risk of meaning drift."}], "correct": "B", "select": 1}]