Question 240
The same procedure appears twice with different amendment dates and no indication which version supersedes the other. What should the team do next?
- Audit the currently enabled Project connectors
- Refresh an uploaded file after a controlled update
- Check underlying permission when a connector returns no document
- Avoid duplicating documents with conflicting titles
Explanation
Correct Answer
D:Duplicate overlapping references can create ambiguity. Maintain a single controlled document or explicit precedence rules.
Incorrect Answers
A:Remove unnecessary integrations and periodically check owner and purpose; that distinction matters because the specific gap is resolving conflicts among repeated knowledge files, not minimizing unused connector exposure.
B:Uploaded material needs lifecycle maintenance; replace the old copy and re-test prompts referencing changed sections; this helps with keeping static knowledge synchronized with revisions while leaving resolving conflicts among repeated knowledge files unsolved.
C:Connectors operate within authorized access boundaries; the missing result may reflect permissions rather than a model failure, yet the scenario asks for resolving conflicts among repeated knowledge files rather than distinguishing retrieval authorization from generation quality.
Question 241
A vendor agreement explicitly forbids transferring customer records to unapproved AI processors, but a staff member proposes direct upload for speed. Which action would address the stated problem most directly?
- Aggregate forbidden customer records before sending them to the disallowed processor
- Honor a contractual restriction on third-party processing
- Classify data before connecting a new source
- Strip unique identifiers not needed for the question
Explanation
Correct Answer
B:Contractual controls define what processing is permitted. Obtain appropriate approval or use a compliant alternative; prompt wording cannot waive the restriction.
Incorrect Answers
A:Aggregation can reduce personal-data exposure when designed to prevent disclosure of small groups; that distinction matters because the specific gap is respecting external processing restrictions, not reducing person-level detail in analytical inputs.
C:Determine sensitivity and permitted processing before enabling access; connector availability is not blanket data authorization; this helps with assessing source sensitivity before access while leaving respecting external processing restrictions unsolved.
D:Only collect or process fields necessary for the approved purpose; unused sensitive columns add risk without analytical benefit, yet the scenario asks for respecting external processing restrictions rather than minimizing unnecessary identity attributes.
Question 242
Every weekly update needs the same correction to include the reporting period; staff currently apply it manually in each new chat. What should the responsible employee choose?
- Preserve the present update’s facts while continuing to retype the missing period each week
- Re-run a small acceptance suite after prompt edits
- Update the core template for a recurring issue
- Make change reasons explicit in the defect log
Explanation
Correct Answer
C:Fix the persistent specification rather than depend on repeated follow-ups. Re-test to confirm the field appears reliably.
Incorrect Answers
A:Scope editing to voice and wording while explicitly retaining approved dates, amounts, and obligations; that distinction matters because the specific gap is eliminating a repeated instruction omission, not protecting checked facts during editorial change.
B:Regression cases check both targeted fixes and unaffected requirements; prompt success must be measured across representative tasks, while the requirement here is eliminating a repeated instruction omission instead of verifying that repairs did not break other behavior.
D:Link each change to a reproducible issue and verification result; this supports evidence-based iteration and rollback, yet the scenario asks for eliminating a repeated instruction omission rather than tracing improvements to observed failures.
Question 243
Claude produces a new customer-service script, but frontline agents have not tested whether its steps match available tools and escalation routes. Which approach best satisfies this requirement?
- Score the draft script at a desk
- Separate extraction of facts from interpretation
- Insert stakeholder validation between draft and rollout
- Design a data-cleaning step before calculating metrics
Explanation
Correct Answer
C:A pilot review with users reveals operational gaps before the script becomes standard. Generating the text is not the same as establishing that it works.
Incorrect Answers
A:An agreed rubric aligns fields, scoring definitions, and exclusions across all submissions; this helps with standardizing evaluation dimensions ahead of review while leaving testing a proposed process with actual users unsolved.
B:Extracting verifiable events first creates an evidentiary base; interpretation can then be labeled and checked separately instead of contaminating the timeline; that is relevant to distinguishing events from causal judgments, not the stated need for testing a proposed process with actual users.
D:Cleaning and defining records first stops duplicate observations and malformed dates from corrupting the metric, while the requirement here is testing a proposed process with actual users instead of fixing record quality before calculation.
Question 244
A team prompts Claude to 'prove that our campaign succeeded' and receives only favorable metrics while returns and complaints increased. What is the most defensible approach?
- Investigate whether the campaign meeting notes contain fabricated promises
- Identify confirmation bias in evidence selection
- Check omission of accessibility needs
- Validate that an attribution is not composite
Explanation
Correct Answer
B:Leading instructions distort evidence selection. Ask for a neutral assessment with supportive and adverse indicators and consistent thresholds.
Incorrect Answers
A:Decision extraction must distinguish proposals, discussion, approvals, and follow-ups, yet the scenario asks for preventing a leading prompt from skewing evaluation rather than distinguishing proposals from recorded decisions.
C:A useful plan must account for affected user groups rather than silently adopting a narrow default; this helps with including accessibility requirements while leaving preventing a leading prompt from skewing evaluation unsolved.
D:A quotation must trace to one actual speaker and passage; composite paraphrases should not be presented as verbatim attributable speech; that distinction matters because the specific gap is preventing a leading prompt from skewing evaluation, not preventing misattributed composite quotes.
Question 245
A document operations team copies dates and reference numbers from standardized forms and already has strong automated reconciliation. How should the team proceed?
- Batch standardized copying into larger AI requests
- Measure error-adjusted cost rather than price per request only
- Use a small validated model for simple extraction
- Reserve expert review for high-risk conclusions
Explanation
Correct Answer
C:Routine extraction may not require the highest reasoning tier. Validate against representative forms and track exceptions.
Incorrect Answers
A:Grouping similar work reduces process overhead while keeping task boundaries and traceability, while the requirement here is reducing model expense on well-defined extraction instead of improving throughput without losing record boundaries.
B:Total workflow cost includes human rework and rejected outputs; per-request price alone can misrepresent economic value; that distinction matters because the specific gap is reducing model expense on well-defined extraction, not comparing end-to-end work cost.
D:Quality controls should be risk-weighted rather than eliminated for cost savings, yet the scenario asks for reducing model expense on well-defined extraction rather than protecting high-impact review gates.
Question 246
A workflow drafts customer-facing claims with Claude and must prevent publication of unsupported promises or inaccurate service dates. What is the most defensible approach?
- Send the customer-claim drafts for production API design
- Design a human approval checkpoint
- Draft acceptance tests for a planned solution
- Prototype a screen concept as an Artifact
Explanation
Correct Answer
B:An authorized reviewer at the publication boundary controls external commitments. The checkpoint belongs in the process, not merely in the prompt.
Incorrect Answers
A:Authentication, authorization, and failure recovery need engineering design and organizational controls, yet the scenario asks for keeping approval before external release rather than routing complex systems engineering to specialists.
C:Acceptance tests turn ambitions into observable input-output cases; they guide implementation review without assuming success, while the requirement here is keeping approval before external release instead of defining testable behavior for a solution.
D:A lightweight prototype makes layout and decision flow tangible for stakeholder feedback; that distinction matters because the specific gap is keeping approval before external release, not exploring interface usability before development.
Question 247
The assistant is asked to authorize a discount beyond an employee's delegation, and the team wants a consistent response. Which action would address the stated problem most directly?
- Place the recurring rule only in one transient conversation for the shared Project
- Describe role, task and output expectations
- Specify acceptable tone without conflicting prohibitions
- Define source precedence explicitly
- State escalation conditions for out-of-scope requests
Explanation
Correct Answer
E:Instructions can define approval thresholds and when to refer a case to a manager. They cannot create actual business approval authority.
Incorrect Answers
A:Chat-level guidance does not automatically become persistent instructions for every new conversation in a Project; it leaves the specified need for identifying the point where a human must authorize unresolved.
B:Specific role, audience, format, and sourcing rules make behavior testable; this helps with creating actionable reusable behavioral guidance while leaving identifying the point where a human must authorize unsolved.
C:Coherent instructions avoid contradictory tone rules; resolve conflicts based on the actual audience and purpose, while the requirement here is identifying the point where a human must authorize instead of removing inconsistent style directives.
D:Priority rules should name the governing source and behavior on conflict, a useful capability for resolving which instruction or source controls but not a solution to identifying the point where a human must authorize.
Question 248
A staff member uploads restricted files and asks the assistant to 'forget everything immediately' as the only proposed protection. Which decision is the strongest fit for this task?
- Treat model instructions as distinct from technical privacy controls
- Check cross-border vendor terms but treat the assistant’s forget request as sufficient deletion
- Check indirect reidentification in supposedly anonymous data
- Apply the relevant retention and records policy
Explanation
Correct Answer
A:A request in the prompt does not establish contractual, account, or system-level handling guarantees. Apply actual approved controls before processing.
Incorrect Answers
B:Authorization must include applicable transfer and processing constraints, while the requirement here is using enforceable privacy mechanisms instead of validating geographic processing obligations.
C:Combinations of quasi-identifiers can reveal a person; evaluate linkage risk before publication or AI processing; this helps with preventing identity inference from sparse records while leaving using enforceable privacy mechanisms unsolved.
D:Retention, deletion, and access obligations may apply to both input and output, a useful capability for managing sensitive generated records across their lifecycle but not a solution to using enforceable privacy mechanisms.
Question 249
A team rewrites prompts repeatedly to retrieve a payroll report that the connected account lacks permission to open. How should the team proceed?
- Keep generating alternative payroll-report prompts
- Escalate when a prompt fix cannot grant data access
- Add tests for an omitted exception rule
- Stop using a failing shortcut when error costs rise
Explanation
Correct Answer
B:Authorization problems require an approved access change or different workflow. Language instructions cannot override source permissions.
Incorrect Answers
A:Name the dimensions to vary and compare versions against the same source facts; that is relevant to producing meaningfully different drafting alternatives, not the stated need for resolving a system access boundary.
C:Encode the exception and test both promoted and ordinary purchases; narrow tests expose whether the rule works across conditions, while the requirement here is resolving a system access boundary instead of correcting a missing special-case rule.
D:Measured error costs justify redesign or escalation of hard cases; price per call is not the only optimization criterion; this helps with correcting a cost-saving method that harms quality while leaving resolving a system access boundary unsolved.
Question 250
An onboarding draft improves stylistically but continues to omit the data-privacy warning that policy requires. What should the responsible employee choose?
- Ask for another broad self-critique
- Revise the prompt when its original goal was ambiguous
- Compare the revision against an explicit acceptance checklist
- Point to the missing evidence and request a correction
Explanation
Correct Answer
C:A checklist catches mandatory omissions independently of how polished the text becomes. It converts feedback from subjective preference into observable completion.
Incorrect Answers
A:A structured critique can surface weaknesses for follow-up; it is a review aid rather than independent proof that the draft is accurate, a useful capability for identifying risks for subsequent human checking but not a solution to verifying that mandatory content survived revision.
B:Repairing the original definition removes a systematic ambiguity from the task, yet the scenario asks for verifying that mandatory content survived revision rather than correcting an ambiguous initial metric.
D:Targeted feedback directs repair to identifiable claims and asks for verifiable support; that is relevant to repairing specifically unsupported claims, not the stated need for verifying that mandatory content survived revision.