Question 220 A repair assistant names the correct replacement part but omits a mandatory electrical isolation step from its response. Which verification is strongest for this particular failure mode? Compare each stated monetary limit with the cited plan passage and score unsupported claims separately from style. Score factuality and harmful-language behavior separately on adversarial tone prompts and confirm safe alternatives pass. Check a labeled procedure checklist and confirm every critical prerequisite appears in the final instruction order. Query with a public identity and confirm no restricted source ID enters the prompt or returned text. ExplanationCorrect Answer C:Answer completeness differs from factual correctness: an otherwise true statement can still be unsafe if a required step is missing. Other Answers A:Check a labeled procedure checklist and confirm every critical prerequisite appears in the final instruction order. B:Check a labeled procedure checklist and confirm every critical prerequisite appears in the final instruction order. D:Check a labeled procedure checklist and confirm every critical prerequisite appears in the final instruction order.
Question 221 A repair assistant names the correct replacement part but omits a mandatory electrical isolation step from its response. What change to engineering governance would reduce this risk over time? Use groundedness evaluations and explicit missing-evidence paths on high-stakes factual outputs. Enforce pre-retrieval access checks and monitor suspected sensitive-data leakage in traces. Maintain independent safety evaluation thresholds rather than collapsing every outcome into one overall correctness score. Maintain critical-step rubrics with negative test cases for omitted safety conditions. ExplanationCorrect Answer D:Answer completeness differs from factual correctness: an otherwise true statement can still be unsafe if a required step is missing. Other Answers A:Answer completeness differs from factual correctness: an otherwise true statement can still be unsafe if a required step is missing. B:This control addresses a different exposure: An internal assistant quotes a confidential HR note in a public answer after a vector search included employee-only documents. It would not prevent this failure, because answer completeness differs from factual correctness: an otherwise true statement can still be unsafe if a required step is missing. C:That answer completeness differs from factual correctness: an otherwise true statement can still be unsafe if a required step is missing.
Question 222 An internal assistant quotes a confidential HR note in a public answer after a vector search included employee-only documents. Which engineering action most directly addresses this specific requirement? Treat the response as incomplete and add procedure-coverage checks that require the safety prerequisite before any repair instructions. Rebalance metadata filters and lexical/vector hybrid search for exact code matches while measuring answer support. Mark the result as unsupported, require claim-to-passage grounding and abstain or escalate when the source lacks the needed limit. Classify the incident as a privacy/access-control failure, block unauthorized retrieval and validate output against disclosure rules. ExplanationCorrect Answer D:The action targets the stated cause: A grounded answer may still be unsafe when the grounding source was outside the caller's entitlement. Check the repair by: query with a public identity and confirm no restricted source ID enters the prompt or returned text. Other Answers A:A grounded answer may still be unsafe when the grounding source was outside the caller's entitlement. B:A grounded answer may still be unsafe when the grounding source was outside the caller's entitlement. C:A grounded answer may still be unsafe when the grounding source was outside the caller's entitlement.
Question 223 An internal assistant quotes a confidential HR note in a public answer after a vector search included employee-only documents. What architectural constraint is most directly responsible? A syntactically valid citation token is not proof that a source exists or supports the answer. Correct factual content does not make abusive presentation acceptable; quality evaluation needs separate safety dimensions. A grounded answer may still be unsafe when the grounding source was outside the caller's entitlement. Fluent generation is not factual evidence; a claimed number must be supported by the retrieved authoritative document. ExplanationCorrect Answer C:Classify the incident as a privacy/access-control failure, block unauthorized retrieval and validate output against disclosure rules. Other Answers A:A grounded answer may still be unsafe when the grounding source was outside the caller's entitlement. B:That mechanism describes another problem: A customer-service assistant gives a helpful policy summary but uses harassing language toward the requester in its final paragraph. It does not explain this case, where a grounded answer may still be unsafe when the grounding source was outside the caller's entitlement. D:A grounded answer may still be unsafe when the grounding source was outside the caller's entitlement.
Question 224 An internal assistant quotes a confidential HR note in a public answer after a vector search included employee-only documents. What validation evidence would most specifically confirm the resolution? Query held-out part numbers and compare exact-ID recall and groundedness before and after hybrid filtering. Compare each stated monetary limit with the cited plan passage and score unsupported claims separately from style. Query with a public identity and confirm no restricted source ID enters the prompt or returned text. Check a labeled procedure checklist and confirm every critical prerequisite appears in the final instruction order. ExplanationCorrect Answer C:A grounded answer may still be unsafe when the grounding source was outside the caller's entitlement. Other Answers A:Query with a public identity and confirm no restricted source ID enters the prompt or returned text. B:Query with a public identity and confirm no restricted source ID enters the prompt or returned text. D:Query with a public identity and confirm no restricted source ID enters the prompt or returned text.
Question 225 An internal assistant quotes a confidential HR note in a public answer after a vector search included employee-only documents. What prevention measure is most targeted to this underlying cause? Use groundedness evaluations and explicit missing-evidence paths on high-stakes factual outputs. Enforce pre-retrieval access checks and monitor suspected sensitive-data leakage in traces. Maintain critical-step rubrics with negative test cases for omitted safety conditions. Monitor identifier coverage and apply corpus-drift checks when ingestion volume changes. ExplanationCorrect Answer B:A grounded answer may still be unsafe when the grounding source was outside the caller's entitlement. Other Answers A:This control addresses a different exposure: A benefits assistant invents a coverage limit absent from the retrieved plan section while stating it with high confidence. It would not prevent this failure, because a grounded answer may still be unsafe when the grounding source was outside the caller's entitlement. C:That a grounded answer may still be unsafe when the grounding source was outside the caller's entitlement. D:A grounded answer may still be unsafe when the grounding source was outside the caller's entitlement.
Question 226 A generator responds to a question about an unavailable product with a fabricated but well-formed citation identifier. Which implementation decision best restores the intended behavior? Fail the safety and tone criterion independently of policy factuality and route the case through controlled moderation or rewrite review. Treat the response as incomplete and add procedure-coverage checks that require the safety prerequisite before any repair instructions. Mark the result as unsupported, require claim-to-passage grounding and abstain or escalate when the source lacks the needed limit. Flag the citation as ungrounded, validate references against the actual retrieved set and return uncertainty when evidence is absent. ExplanationCorrect Answer D:The action targets the stated cause: A syntactically valid citation token is not proof that a source exists or supports the answer. Check the repair by: verify every generated reference against the retrieval manifest and inspect whether the cited passage entails the claim. Other Answers A:A syntactically valid citation token is not proof that a source exists or supports the answer. B:A syntactically valid citation token is not proof that a source exists or supports the answer. C:A syntactically valid citation token is not proof that a source exists or supports the answer.
Question 227 A generator responds to a question about an unavailable product with a fabricated but well-formed citation identifier. Which underlying technical principle explains why this occurs? Fluent generation is not factual evidence; a claimed number must be supported by the retrieved authoritative document. Answer completeness differs from factual correctness: an otherwise true statement can still be unsafe if a required step is missing. A syntactically valid citation token is not proof that a source exists or supports the answer. Coarse chunks can match topic vocabulary yet bury the actual answer span among unrelated context. ExplanationCorrect Answer C:Flag the citation as ungrounded, validate references against the actual retrieved set and return uncertainty when evidence is absent. Other Answers A:That mechanism describes another problem: A benefits assistant invents a coverage limit absent from the retrieved plan section while stating it with high confidence. It does not explain this case, where a syntactically valid citation token is not proof that a source exists or supports the answer. B:A syntactically valid citation token is not proof that a source exists or supports the answer. D:A syntactically valid citation token is not proof that a source exists or supports the answer.
Question 228 A generator responds to a question about an unavailable product with a fabricated but well-formed citation identifier. After correction, what targeted test should demonstrate success? Compare each stated monetary limit with the cited plan passage and score unsupported claims separately from style. Score factuality and harmful-language behavior separately on adversarial tone prompts and confirm safe alternatives pass. Verify every generated reference against the retrieval manifest and inspect whether the cited passage entails the claim. Check a labeled procedure checklist and confirm every critical prerequisite appears in the final instruction order. ExplanationCorrect Answer C:A syntactically valid citation token is not proof that a source exists or supports the answer. Other Answers A:Verify every generated reference against the retrieval manifest and inspect whether the cited passage entails the claim. B:Verify every generated reference against the retrieval manifest and inspect whether the cited passage entails the claim. D:Verify every generated reference against the retrieval manifest and inspect whether the cited passage entails the claim.
Question 229 A generator responds to a question about an unavailable product with a fabricated but well-formed citation identifier. Which release safeguard most directly addresses this failure class? Use groundedness evaluations and explicit missing-evidence paths on high-stakes factual outputs. Maintain critical-step rubrics with negative test cases for omitted safety conditions. Add citation-existence and entailment checks to the answer-quality evaluation suite. Maintain independent safety evaluation thresholds rather than collapsing every outcome into one overall correctness score. ExplanationCorrect Answer C:A syntactically valid citation token is not proof that a source exists or supports the answer. Other Answers A:That a syntactically valid citation token is not proof that a source exists or supports the answer. B:A syntactically valid citation token is not proof that a source exists or supports the answer. D:This control addresses a different exposure: A customer-service assistant gives a helpful policy summary but uses harassing language toward the requester in its final paragraph. It would not prevent this failure, because a syntactically valid citation token is not proof that a source exists or supports the answer.
Question 230 A customer-service assistant gives a helpful policy summary but uses harassing language toward the requester in its final paragraph. Which design change directly satisfies the stated constraints? Report sampled monitoring coverage and uncertainty separately from full-bank offline evaluation or qualified human review. Treat the response as incomplete and add procedure-coverage checks that require the safety prerequisite before any repair instructions. Mark the result as unsupported, require claim-to-passage grounding and abstain or escalate when the source lacks the needed limit. Fail the safety and tone criterion independently of policy factuality and route the case through controlled moderation or rewrite review. ExplanationCorrect Answer D:The action targets the stated cause: Correct factual content does not make abusive presentation acceptable; quality evaluation needs separate safety dimensions. Check the repair by: score factuality and harmful-language behavior separately on adversarial tone prompts and confirm safe alternatives pass. Other Answers A:Correct factual content does not make abusive presentation acceptable; quality evaluation needs separate safety dimensions. B:Correct factual content does not make abusive presentation acceptable; quality evaluation needs separate safety dimensions. C:Correct factual content does not make abusive presentation acceptable; quality evaluation needs separate safety dimensions.
