Question 160
A prototype misroutes requests containing two issue categories, and the project lead considers rewriting all instructions without diagnosing examples. What should the responsible employee choose?
- Route mixed-category cases for escalation
- Keep a versioned decision log during iterations
- Use real sample records in a controlled pilot
- Iterate from specific observed failures
Explanation
Correct Answer
D:Failure examples identify whether rules, categories, or context are inadequate. Narrow, tested corrections are easier to attribute.
Incorrect Answers
A:A threshold for uncertainty or rule conflict protects the workflow from overconfident routing, a useful capability for routing ambiguous cases to accountable people but not a solution to correcting reproducible prototype errors.
B:Versioned decisions preserve approved changes and their rationale; it avoids implementing contradictory drafts; that distinction matters because the specific gap is correcting reproducible prototype errors, not tracking changes and approved decisions.
C:A controlled, permitted pilot with representative inputs measures actual performance, yet the scenario asks for correcting reproducible prototype errors rather than assessing operation on representative permitted data.
Question 161
A template writer wants to place real client passwords and account identifiers into persistent instructions to simplify later support conversations. What is the best response to this situation?
- Avoid pasting confidential facts into general instructions
- Stress-test the persistent template with conflicting requests while real passwords remain embedded
- Keep a change note for significant rule updates
- Use project instructions rather than profile preferences for team rules
Explanation
Correct Answer
A:Instructions should contain behavior and references, not secrets or unnecessary private data. Credentials belong in approved secure systems.
Incorrect Answers
B:Adversarial cases test whether the configured source hierarchy is respected, yet the scenario asks for keeping sensitive secrets out of configuration text rather than checking resilience against conflicting lower-trust content.
C:Recording configuration changes supports audit and diagnosis; versioned rules let reviewers attribute output shifts to deliberate edits; this helps with maintaining traceability of instructional changes while leaving keeping sensitive secrets out of configuration text unsolved.
D:Workspace-specific instructions are the appropriate shared locus for team workflow rules, while the requirement here is keeping sensitive secrets out of configuration text instead of scoping a rule to the common project workflow.
Question 162
A generated customer letter is ready for distribution, but company policy requires approval by the account owner for pricing or delivery commitments. What should the team do next?
- Report a suspected data exposure through incident channels
- Allow a fluent draft to bypass required specialist sign-off in the organization’s AI process
- Follow the defined human sign-off policy
- Consult the approved AI-use policy before starting
- Document AI assistance when disclosure is required
Explanation
Correct Answer
C:Organizational sign-off remains mandatory even if the prose appears correct. A model cannot confer approval rights.
Incorrect Answers
A:Potential breaches require timely reporting and containment according to policy; this helps with escalating a potential confidentiality incident while leaving preserving mandated business approvals unsolved.
B:Generated judgments may be incorrect or biased and do not transfer legal or professional accountability, but this shortcut offers no dependable way of preserving mandated business approvals.
D:Corporate rules determine approved tools and permitted data classes; verify authorization before using an unfamiliar service, while the requirement here is preserving mandated business approvals instead of checking which AI environment is sanctioned.
E:Follow the organization's transparency requirements about use of generated material, yet the scenario asks for preserving mandated business approvals rather than complying with required AI-use transparency.
Question 163
A connected policy service sometimes times out, and the assistant currently guesses current rules rather than alert staff. Which approach best satisfies this requirement?
- Review total processing cost across all stages
- Switch model tiers before checking task instructions or input quality during fault diagnosis
- Prioritize recurring errors over cosmetic edits
- Track accepted output rather than generated volume
- Introduce a fallback when source retrieval fails
Explanation
Correct Answer
E:Fallback behavior should withhold unsupported claims and route to a manual source when necessary. Reliability includes graceful failure.
Incorrect Answers
A:Total process economics include model consumption, operator time, corrections, escalations, and rejected outcomes; that distinction matters because the specific gap is handling unavailable evidence safely, not including downstream effort in optimization.
B:Changing models does not supply absent source data, fix ambiguous requirements, or remove contradictory instructions; it leaves the specified need for handling unavailable evidence safely unresolved.
C:Correctness defects with material impact take priority; cosmetic polish cannot compensate for systemic analytic errors, while the requirement here is handling unavailable evidence safely instead of focusing improvement on high-impact defects.
D:The meaningful productivity measure is the number of complete, approved outputs reaching the recipient; this helps with measuring usable finished production while leaving handling unavailable evidence safely unsolved.
Question 164
A classification prompt works on two obvious cases but fails when customers express both praise and a complaint in one message. Which option best resolves the stated issue?
- Use varied examples to expose a fragile instruction
- Approve the classification prompt for reuse after only its two easiest test examples
- Add a measurable output limit after reviewing length
- Explicitly preserve accepted portions during revision
Explanation
Correct Answer
A:Challenging prompts with ambiguous and edge examples tests the decision rule rather than just easy cases. Failures identify what needs clarification.
Incorrect Answers
B:Saving the effective specification turns an accidental success into a repeatable process, a useful capability for reusing validated instructions across cycles but not a solution to measuring robustness on boundary cases.
C:A concrete length and structure target translates reader feedback into a repeatable production constraint, while the requirement here is measuring robustness on boundary cases instead of turning subjective feedback into a size rule.
D:Locking approved sections prevents an aesthetic edit from introducing new factual errors; that is relevant to protecting already-approved source facts, not the stated need for measuring robustness on boundary cases.
Question 165
A global meeting summary gives a deadline that shifts one day for overseas teams because the original event timezone was omitted. How should the team proceed?
- Treat Claude’s own confidence estimate as the primary validation evidence during evidence review
- Compare two independent source records
- Check the publication date of external findings
- Review date and timezone conversions
- Audit whether extracted data matches source rows
Explanation
Correct Answer
D:Convert timestamps from an explicit source zone and account for calendar rollover. Ambiguous local times cannot be safely normalized by intuition.
Incorrect Answers
A:Self-ratings are model-produced opinions rather than independent checks of source claims, but this shortcut offers no dependable way of verifying timezone-dependent deadlines.
B:Independent operational records give cross-checks when one generated narrative may reflect a misread document, while the requirement here is verifying timezone-dependent deadlines instead of corroborating events across authoritative records.
C:Evidence must be assessed for age and relevance to the versions being discussed, yet the scenario asks for verifying timezone-dependent deadlines rather than validating evidence freshness.
E:Row-and-column cross-checking identifies extraction shifts and schema errors; this helps with checking extraction against original records while leaving verifying timezone-dependent deadlines unsolved.
Question 166
An extended project conversation is losing earlier constraints; the team wants to continue without silently abandoning approved requirements. What should the responsible employee choose?
- Begin a new blank conversation for the ongoing project
- Place recurring source documents in Project knowledge
- Do not rely on imagined cross-chat recall
- Summarize confirmed decisions before restarting a long chat
Explanation
Correct Answer
D:A concise handoff of decisions, sources, and unresolved items preserves continuity when moving to a fresh context. Verify that the summary is accurate.
Incorrect Answers
A:Topic separation reduces irrelevant history and contradictory instructions; this helps with isolating an independent discussion while leaving transferring stable decisions into new context unsolved.
B:Centralizing curated reference files reduces version drift and repetitive uploads; that distinction matters because the specific gap is transferring stable decisions into new context, not reusing authoritative shared reference material.
C:Claude needs accessible context for task-specific private facts; missing material should be supplied through approved context rather than assumed, while the requirement here is transferring stable decisions into new context instead of ensuring the necessary context is actually available.
Question 167
The design works for routine orders, yet refunds, partial fulfillment, and duplicated identifiers were excluded from tests. How should the team proceed?
- Document ordinary approval owners
- Inspect the operational exception path
- Pilot within a small reversible scope
- Test accessibility of a proposed interface
Explanation
Correct Answer
B:Exception-path testing challenges assumptions that normal cases conceal. Production readiness depends on more than the happy path.
Incorrect Answers
A:A design should name who checks, authorizes, and remediates outputs; otherwise accountability is lost during handoffs; that distinction matters because the specific gap is proving handling of atypical records, not assigning accountable review in the new process.
C:A bounded pilot limits potential disruption and supplies evidence for staged rollout; this helps with limiting initial operational exposure while leaving proving handling of atypical records unsolved.
D:Accessibility testing belongs in solution validation, not after launch; visual appearance does not establish usability for all users, while the requirement here is proving handling of atypical records instead of checking inclusive interaction requirements.
Question 168
A new organizational writing standard removes a previously required disclaimer, but Project outputs continue to include the obsolete paragraph. What should the team do next?
- Collect more complaints about the outdated disclaimer
- Schedule periodic review of knowledge currency
- Assign responsibility for approving changes
- Replace withdrawn instructions and retest
Explanation
Correct Answer
D:Update the controlling instruction, then run representative checks for both corrected behavior and regressions.
Incorrect Answers
A:Reproducible issue reports make defects diagnosable; gather inputs, expected behavior, observed behavior, and frequency; this helps with turning vague user complaints into test cases while leaving verifying behavior after a policy-driven change unsolved.
B:A review cadence catches stale sources before they influence repeated outputs; that is relevant to periodically confirming reference freshness, not the stated need for verifying behavior after a policy-driven change.
C:Change control prevents conflicting edits and establishes accountability, a useful capability for governing who can alter shared rules but not a solution to verifying behavior after a policy-driven change.
Question 169
Reviewers routinely correct an AI-generated eligibility summary, but no log captures what changed or why. Which action would address the stated problem most directly?
- Retain human eligibility decisions but omit an auditable log of reviewer corrections
- Apply version control to approved templates
- Keep a record of material human overrides
- Limit use to approved accounts and connectors
Explanation
Correct Answer
C:Logging material reviewer corrections provides an audit trail and reveals repeated failure patterns. The organization can use this evidence to prioritize safeguards where its policy requires records.
Incorrect Answers
A:Governance policy can prohibit or condition autonomous decisions; a technical capability does not create a policy exception, a useful capability for keeping consequential actions within authority but not a solution to recording consequential corrections.
B:A versioned approved template communicates which rules are current, prevents silent conflicts, and lets editors trace changes after a policy revision; this helps with maintaining current approved prompt standards while leaving recording consequential corrections unsolved.
D:Organization-approved connections preserve intended access and governance boundaries; that is relevant to using permitted integration channels, not the stated need for recording consequential corrections.
Question 170
An approved classification workflow handles standard requests accurately, while a small set of high-impact anomalies always needs manual judgment. What should the responsible employee choose?
- Expand a regression suite of already-successful routine cases
- Measure quality after each optimization change
- Separate human review for exceptions from routine processing
- Keep a known-good configuration for rollback
Explanation
Correct Answer
C:Exception routing limits reviewer workload without relinquishing control over uncertain cases. Criteria and monitoring keep the boundary safe.
Incorrect Answers
A:A maintained suite covering typical and edge cases detects quality drift, a useful capability for checking robustness across representative inputs but not a solution to concentrating human attention on atypical high-risk items.
B:Cost, accuracy, and completion must be measured together on the same test set; that is relevant to validating trade-offs after a cost reduction, not the stated need for concentrating human attention on atypical high-risk items.
D:Versioned instructions and rollback criteria allow recovery while the defect is investigated, while the requirement here is concentrating human attention on atypical high-risk items instead of reversing a harmful prompt release.