Question 370
A ticket-routing pilot should send only uncertain or high-impact cases to a reviewer. Its new prompt version has produced a harmful error, and a cost-cutting proposal now needs a measured quality comparison with the previous baseline. Which THREE measures address exception review, prompt rollback, and post-change quality? (Choose THREE.)
- Measure quality after each optimization change
- Measure only the average generation time after switching to the cheaper model
- Separate human review for exceptions from routine processing
- Keep a known-good configuration for rollback
- Repeat the same prompt unchanged several more times
- Ignore user feedback when the model follows the original format
Explanation
Correct Answers
A:Cost, accuracy, and completion must be measured together on the same test set; efficiency gains that erode essential quality are not improvements. Compare the proposed cheaper workflow with the prior baseline on routing accuracy and correction effort, not on cost alone.
C:Exception routing limits reviewer workload without relinquishing control over uncertain cases; criteria and monitoring keep the boundary safe. Escalate uncertain or high-impact tickets while keeping routine tickets on the standard supported path.
D:Versioned instructions and rollback criteria allow recovery while the defect is investigated; uncontrolled changes undermine reliability. Restore the last verified prompt after the harmful routing error and investigate the change before redeployment.
Incorrect Answers
B:Average generation time measures latency, not whether the cheaper workflow routes tickets correctly. The team must compare errors and exception handling against the former baseline; optimizing only speed can hide a harmful quality regression even while the interface appears faster.
E:Additional samples without diagnosis or better constraints seldom fix a systematic missing-input problem; it leaves the specified need for concentrating human attention on atypical high-risk items and reversing a harmful prompt release and validating trade-offs after a cost reduction unresolved.
F:An output may conform to formatting while failing the user’s task, audience, or acceptance criteria, which does not replace the required process of concentrating human attention on atypical high-risk items and reversing a harmful prompt release and validating trade-offs after a cost reduction.
Question 371
An event planner receives ten nearly identical webinar ideas and wants genuinely different formats that suit the same budget. Which action would address the stated problem most directly?
- Ask for current sources and a defined timeframe for research
- Demand deliberate variety in brainstorming
- Request assumptions and computational inputs for analysis
- Append additional raw notes but leave the required action unspecified during prompt refinement
- Provide audience, voice and length for a draft
Explanation
Correct Answer
B:Asking for variation across audience, channel, format, and effort broadens the idea space. Scoring can follow once the options are diverse.
Incorrect Answers
A:Research instructions should define recency and require checkable citations, a useful capability for gathering time-bounded external evidence but not a solution to exploring differentiated candidate ideas.
C:Asking for the underlying figures and calculation method makes the analysis reproducible and easy to recompute outside Claude; this helps with making numerical analysis auditable while leaving exploring differentiated candidate ideas unsolved.
D:More pasted material enlarges context without telling Claude which decision or output matters; the actual task calls for exploring differentiated candidate ideas instead.
E:Drafting constraints guide tone, concision, and authorized commitments; a factual research method alone would not determine an appropriate customer message; that distinction matters because the specific gap is exploring differentiated candidate ideas, not drafting a message within communication guardrails.
Question 372
A subsidy summary misapplies an eligibility threshold near its cutoff, cites provisions that reviewers cannot locate, and quotes two conflicting official notices as if both were controlling. Which THREE validation actions directly test the threshold, make evidence inspectable, and resolve authority conflicts? (Choose THREE.)
- Ask Claude to restate the cutoff in simpler language before checking the rule
- Compare a second generated summary without opening the underlying notices
- Preserve a reference to the exact policy version
- Request exact cited extracts for review
- Test numerical claims on edge cases
- Resolve conflicting authoritative sources explicitly
Explanation
Correct Answers
D:Brief source extracts with locations focus inspection on specific evidence; the reviewer still checks them against the actual standard. The subsidy reviewer must be able to locate the actual cited provisions and compare the extract with the published notice.
E:Boundary inputs reveal rule implementation defects hidden by typical examples; verification should include relevant thresholds and exception conditions. Test values immediately below, exactly at, and above the subsidy threshold to reveal an off-by-one eligibility mistake.
F:Identify the controlling revision and jurisdiction rather than arbitrarily selecting a convenient date; record unresolved conflict for human confirmation. Determine which notice is current and legally applicable; the model cannot treat mutually inconsistent rules as equally controlling.
Incorrect Answers
A:Fluent, authoritative wording is not evidence; models can state fabricated specifics with certainty, but this shortcut offers no dependable way of testing a numeric rule at its boundaries and making evidence passages inspectable and reconciling conflicting official statements.
B:A citation-looking string can be irrelevant, outdated, inaccessible, or fabricated; the underlying source must be inspected; it leaves the specified need for testing a numeric rule at its boundaries and making evidence passages inspectable and reconciling conflicting official statements unresolved.
C:Version identifiers and retrieval dates make evidence traceable; a generic link to a changing document may not reproduce the same rule later; that distinction matters because the specific gap is testing a numeric rule at its boundaries and making evidence passages inspectable and reconciling conflicting official statements, not preserving an auditable source version.
Question 373
A project continues to rely on a 2025 procedure while a controlled 2026 revision replaces several approval steps. Which option best resolves the stated issue?
- Refresh sources when a new edition supersedes an old one
- Keep a chat on one coherent problem
- Separate reusable rules from ephemeral facts
- Use a concise carry-forward summary of a resolved task
- Restart the conversation each time without supplying the needed reference files when selecting a Claude feature
Explanation
Correct Answer
A:Persistent knowledge must be curated for currency. Remembering outdated information well is not operational correctness.
Incorrect Answers
B:Continuing a focused thread preserves immediate assumptions and revisions, a useful capability for maintaining useful same-task continuity but not a solution to updating long-lived context to the current revision.
C:Persistent instructions belong in shared configuration, while current inputs remain task-specific; this helps with distinguishing durable instructions from changing data while leaving updating long-lived context to the current revision unsolved.
D:A structured handoff discards irrelevant exploratory chatter while preserving actionable evidence, yet the scenario asks for updating long-lived context to the current revision rather than compressing history into verified work products.
E:Restarting discards task continuity; reusable project knowledge or the correct tool is a better fit for recurring requirements, but this shortcut offers no dependable way of updating long-lived context to the current revision.
Question 374
A redesign proposal removes the final compliance check to claim maximal time savings on externally published advice. Which approach best satisfies this requirement?
- Maintain a manual operational fallback while deleting the compliance sign-off for published advice
- Redesign the end-to-end workflow when approval delays dominate
- Augment an existing drafting step first
- Retain review for consequential actions
Explanation
Correct Answer
D:Control points should be optimized, not deleted blindly. Consequential output still requires authorized validation.
Incorrect Answers
A:A fallback protects essential operations during outages or degraded outputs, a useful capability for maintaining service during tool failures but not a solution to preserving mandatory assurance in a new flow.
B:When the bottleneck moves, reconsider handoffs, exception handling, and reviewer allocation; this helps with addressing a downstream constraint through redesign while leaving preserving mandatory assurance in a new flow unsolved.
C:Inserting Claude at the drafting stage can reduce effort while preserving approvals and handoffs, yet the scenario asks for preserving mandatory assurance in a new flow rather than improving one step without changing roles.
Question 375
A team member changes roles, but a shared workflow still assumes they can retrieve the previous department's restricted files. Which method best fits these constraints?
- Categorize general prompt failures while ignoring the former team member’s now-invalid file permissions
- Verify continuing connector permissions
- Remove superseded material rather than accumulating layers
- Run regression tests after a new file upload
Explanation
Correct Answer
B:Access and connector permissions must reflect current entitlements. Workflow design should not presume historical access remains valid.
Incorrect Answers
A:Categorized feedback reveals systematic defects and directs targeted configuration changes, yet the scenario asks for keeping live-source permissions current rather than prioritizing fixes from recurring failure patterns.
C:Curated replacement reduces version conflict; an archive can remain elsewhere without being active reference context; that is relevant to preventing historical policy mixing, not the stated need for keeping live-source permissions current.
D:Standard test prompts compare expected behaviors before and after the update, while the requirement here is keeping live-source permissions current instead of detecting unintended effects of knowledge updates.
Question 376
A customer-support team starts using generated responses for external publication instead of its original low-risk internal brainstorming use case. What is the best response to this situation?
- Include training before a large rollout
- Limit vendor promises to verified platform behavior
- Review policy compliance after a workflow changes
- Allow a fluent draft to bypass required specialist sign-off in the organization’s AI process
- Request an exception through formal channels
Explanation
Correct Answer
C:Changed purpose and audience may change required approvals and controls. Reassess governance when risk scope expands.
Incorrect Answers
A:Governance requires user awareness of permitted tasks, data rules, escalation paths, and review obligations, yet the scenario asks for updating controls when the use case evolves rather than preparing workers to follow AI policy.
B:Product and data-handling assurances depend on actual terms and configuration; that is relevant to validating organizational data-handling claims, not the stated need for updating controls when the use case evolves.
D:Generated judgments may be incorrect or biased and do not transfer legal or professional accountability, but this shortcut offers no dependable way of updating controls when the use case evolves.
E:Document the business need and request review through the designated exception process, while the requirement here is updating controls when the use case evolves instead of seeking authorized deviation from standard tools.
Question 377
An innovation workshop gets familiar incremental suggestions; the chair wants options including process changes, partnerships, and low-tech approaches. Which decision is the strongest fit for this task?
- Demand a heavily cited literature overview
- Ask for unconventional alternatives before narrowing
- Provide brand examples for persuasive writing
- Specify a decision framework for trade-off analysis
Explanation
Correct Answer
B:Broad category prompts counter anchoring on the first obvious solution. Feasibility review belongs in a later evaluation stage.
Incorrect Answers
A:A research-and-synthesis task needs the original materials and traceable support for comparative claims, a useful capability for comparing verified documents rather than remembered claims but not a solution to widening the initial solution space.
C:Concrete references let the model imitate style while keeping new details original, while the requirement here is widening the initial solution space instead of matching a specific communication voice.
D:Analysis requires explicit criteria, relative importance, and assumptions so a reviewer can challenge the decision rather than accept intuition; that distinction matters because the specific gap is widening the initial solution space, not weighing competing decision factors.
Question 378
A staff member asks Claude for alternative icebreaker questions for a routine team meeting, with no external commitments or sensitive claims. What should the responsible employee choose?
- Require legal approval before sending binding advice
- Require human adjudication for hiring decisions
- Treat Claude’s own confidence estimate as the primary validation evidence during evidence review
- Permit lighter review for an internal low-risk draft
- Require clinician review for health-related recommendations
Explanation
Correct Answer
D:A basic relevance and appropriateness review is proportionate to low-impact ideation. Intensive legal sign-off would add little value here.
Incorrect Answers
A:Legal specialists must confirm the interpretation and approved wording; a well-phrased model output is not authorization to offer binding terms, while the requirement here is using proportionate oversight for low-stakes ideas instead of obtaining competent legal sign-off.
B:Employment decisions need authorized human evaluation, relevant evidence, fairness controls, and compliant processes, yet the scenario asks for using proportionate oversight for low-stakes ideas rather than preserving accountability in personnel decisions.
C:Self-ratings are model-produced opinions rather than independent checks of source claims, but this shortcut offers no dependable way of using proportionate oversight for low-stakes ideas.
E:A qualified healthcare professional must validate medical claims and safe scope before release; that is relevant to obtaining clinical oversight for consequential advice, not the stated need for using proportionate oversight for low-stakes ideas.
Question 379
A conversation handoff condenses several manager constraints but accidentally drops the requirement for bilingual output. Which approach best satisfies this requirement?
- Write a fresh decision summary
- Confirm a summarized instruction before discarding the original
- Avoid treating memory as a source of current law
- Use narrowly scoped source excerpts when context is crowded
Explanation
Correct Answer
B:Check summary fidelity against the original decisions before restarting. Compression can lose seemingly small but binding requirements.
Incorrect Answers
A:A handoff summary should preserve verified decisions and uncertainties, not simply copy the entire chat; this helps with transferring project state between workers while leaving validating a compressed context handoff unsolved.
C:Remembered context supports continuity, not authoritative currentness; changing external rules still need fresh source checks, a useful capability for separating continuity from source authority but not a solution to validating a compressed context handoff.
D:Focused relevant extracts reduce distractions while preserving the governing details; that distinction matters because the specific gap is validating a compressed context handoff, not concentrating context on pertinent evidence.
Question 380
A redesigned workflow introduces AI-generated intake suggestions, yet staff believe the suggestions replace their existing duty to verify customer eligibility. What should the responsible employee choose?
- Count approved intake throughput while staff still misunderstand who decides eligibility
- Train operators on changed responsibilities
- Use a structured exception queue
- Map the full process before automating individual tasks
Explanation
Correct Answer
B:Change management must state new boundaries and accountability. A technically functioning draft system can still fail through misuse.
Incorrect Answers
A:Completed accepted outputs are a more meaningful measure than raw generation, a useful capability for counting usable completed work but not a solution to aligning staff behavior with revised roles.
C:Exception routing separates safe routine work from cases requiring human judgment; that distinction matters because the specific gap is aligning staff behavior with revised roles, not escalating high-risk or ambiguous cases.
D:End-to-end mapping identifies shared constraints and opportunities; local acceleration can otherwise overload the next stage; this helps with seeing the entire process before task selection while leaving aligning staff behavior with revised roles unsolved.