Question 110
Global prompt maintenance accidentally alters a tenant's policy-specific behavior across unrelated services. Which acceptance result provides the most specific evidence?
- The measured evidence confirms that attack prompts embedded in reviews do not change tool permissions or task objective.
- The scenario test demonstrates that paraphrased customer requests produce the same refund-policy decision.
- The end-to-end check confirms that each output trace identifies the approved prompt revision and linked evaluation run.
- The observed outcome is that tenant-switch tests show only authorized tenant context influences a response.
Explanation
Correct Answer
D:Modularity reduces cross-tenant leakage and makes policy changes easier to audit; the safeguard must work when a reusable prompt contains extensive tenant-specific details hard-coded into global instructions.
Incorrect Answers
A:The embedded instruction has no authority to override the system's allowed summarization task; this makes sense when a user inserts 'ignore all rules' into a quoted customer review, rather than the risk that global prompt maintenance accidentally alters a tenant's policy-specific behavior across unrelated services.
B:Instruction hierarchy and fixed templates keep governing business rules consistent across requests; this technique fits where a billing assistant gives different refund criteria depending on the user's writing style, rather than the condition that global prompt maintenance accidentally alters a tenant's policy-specific behavior across unrelated services.
C:Prompt text is application logic whose untracked changes can cause regression and unsafe behavior; the trigger would be that a team edits production prompts directly and cannot explain changed answers, while the immediate question involves global prompt maintenance accidentally alters a tenant's policy-specific behavior across unrelated services.
Question 111
A summarizer blurs quoted evidence and its own inference, obscuring what the source actually supports. Which acceptance check evaluates this requirement rather than a proxy?
- The negative-case evaluation confirms that attack prompts embedded in reviews do not change tool permissions or task objective.
- The audit sample demonstrates that tenant-switch tests show only authorized tenant context influences a response.
- The measured evidence confirms that refund-limit probes consistently produce escalation instead of unauthorized commitments.
- The scenario test demonstrates that unsupported-claim probes trigger an uncertainty statement rather than fabricated citations.
Explanation
Correct Answer
D:Clear evidence instructions help distinguish supported facts from interpretation in the final response; the test covers the original risk that a model must cite provided materials yet sometimes blends source statements with inference.
Incorrect Answers
A:The embedded instruction has no authority to override the system's allowed summarization task; the other workload is one where a user inserts 'ignore all rules' into a quoted customer review, not the risk where a summarizer blurs quoted evidence and its own inference, obscuring what the source actually supports.
B:Modularity reduces cross-tenant leakage and makes policy changes easier to audit; this applies when a reusable prompt contains extensive tenant-specific details hard-coded into global instructions, not this situation where a summarizer blurs quoted evidence and its own inference, obscuring what the source actually supports.
C:Behavioral boundaries should make the assistant defer approval rather than invent authority; the alternative helps if a support bot makes undocumented promises about customer reimbursements, whereas this scenario involves a summarizer blurs quoted evidence and its own inference, obscuring what the source actually supports.
Question 112
A classifier misunderstands rare borderline categories with a bare one-line instruction. What intervention best matches the stated boundary?
- For the affected service, use a structured checklist of required sections and evaluate completion.
- In the project scope, provide a small, diverse set of labeled borderline examples with clear decision criteria.
- Within the rollout design, specify explicit verifiable intermediate fields and deterministic checks in the output contract.
- For the next architecture revision, decompose prompt stages into evidence selection, decision and final response.
Explanation
Correct Answer
B:Few-shot examples demonstrate nuanced boundaries more concretely than repeated abstract labels; a representative trial should prove that holdout borderline cases improve without overfitting to the few-shot exemplars.
Incorrect Answers
A:An explicit task structure increases coverage of mandatory dimensions without relying on length alone; it serves conditions where a generated summary omits minor but mandatory report categories, but this decision concerns a classifier misunderstands rare borderline categories with a bare one-line instruction.
C:Observable checks give auditors evidence without relying on opaque or speculative reasoning transcripts; this makes sense when a complex eligibility answer needs reproducible intermediate checks but not exposed private reasoning, but the requested boundary concerns a classifier misunderstands rare borderline categories with a bare one-line instruction.
D:Stagewise prompting exposes where a failure occurs and reduces competing objectives in one request; this technique fits where a complex task fails when requested in one step despite valid source materials, not the present case where a classifier misunderstands rare borderline categories with a bare one-line instruction.
Question 113
A straightforward document tagger has a stable schema and no unusual edge cases. Which implementation decision is the strongest fit?
- For the primary workstream, use a structured checklist of required sections and evaluate completion.
- For this workflow, provide a small, diverse set of labeled borderline examples with clear decision criteria.
- For the affected service, begin with a minimal zero-shot instruction and evaluate before adding complexity.
- In the project scope, delimit input data and instructions clearly using separated sections or XML-style tags.
Explanation
Correct Answer
C:Unnecessary examples increase context cost when direct instructions already meet the acceptance threshold; the implementation should produce evidence that the minimal prompt matches baseline quality at lower token usage.
Incorrect Answers
A:An explicit task structure increases coverage of mandatory dimensions without relying on length alone; it serves a setting where a generated summary omits minor but mandatory report categories, rather than the condition that a straightforward document tagger has a stable schema and no unusual edge cases.
B:Few-shot examples demonstrate nuanced boundaries more concretely than repeated abstract labels; that is defensible if a classifier misunderstands rare borderline categories with a bare one-line instruction, while the immediate question involves a straightforward document tagger has a stable schema and no unusual edge cases.
D:Boundary markers clarify which text is evidence and which text governs output behavior; this applies when a model confuses source documents with formatting instructions when both appear in one block, whereas the current issue is that a straightforward document tagger has a stable schema and no unusual edge cases.
Question 114
A complex eligibility answer needs reproducible intermediate checks but not exposed private reasoning. What should be prioritized during architecture review?
- Within the delivery plan, decompose prompt stages into evidence selection, decision and final response.
- At the system boundary, provide a small, diverse set of labeled borderline examples with clear decision criteria.
- For the primary workstream, specify explicit verifiable intermediate fields and deterministic checks in the output contract.
- For this workflow, delimit input data and instructions clearly using separated sections or XML-style tags.
- Under the reviewed approach, begin with a minimal zero-shot instruction and evaluate before adding complexity.
Explanation
Correct Answer
C:Observable checks give auditors evidence without relying on opaque or speculative reasoning transcripts; a suitable release result is that auditors can recompute each eligibility condition from approved fields.
Incorrect Answers
A:Stagewise prompting exposes where a failure occurs and reduces competing objectives in one request; that belongs to situations where a complex task fails when requested in one step despite valid source materials, not this situation where a complex eligibility answer needs reproducible intermediate checks but not exposed private reasoning.
B:Few-shot examples demonstrate nuanced boundaries more concretely than repeated abstract labels; it is relevant if a classifier misunderstands rare borderline categories with a bare one-line instruction, whereas this scenario involves a complex eligibility answer needs reproducible intermediate checks but not exposed private reasoning.
D:Boundary markers clarify which text is evidence and which text governs output behavior; the other case is when a model confuses source documents with formatting instructions when both appear in one block, instead of when a complex eligibility answer needs reproducible intermediate checks but not exposed private reasoning.
E:Unnecessary examples increase context cost when direct instructions already meet the acceptance threshold; this makes sense when a straightforward document tagger has a stable schema and no unusual edge cases, not the case in which a complex eligibility answer needs reproducible intermediate checks but not exposed private reasoning.
Question 115
A generated summary omits minor but mandatory report categories. Which response follows from the actual operational constraint?
- For this production path, begin with a minimal zero-shot instruction and evaluate before adding complexity.
- For the intended users, decompose prompt stages into evidence selection, decision and final response.
- Within the delivery plan, test prompt robustness with paraphrases and strengthen the task-specific decision rubric.
- At the system boundary, use a structured checklist of required sections and evaluate completion.
Explanation
Correct Answer
D:An explicit task structure increases coverage of mandatory dimensions without relying on length alone; the practical acceptance condition is that the report passes a section-level completeness test across varied source documents.
Incorrect Answers
A:Unnecessary examples increase context cost when direct instructions already meet the acceptance threshold; this control is needed when a straightforward document tagger has a stable schema and no unusual edge cases, but the actual risk is that a generated summary omits minor but mandatory report categories.
B:Stagewise prompting exposes where a failure occurs and reduces competing objectives in one request; that would be useful if a complex task fails when requested in one step despite valid source materials, but the requested boundary concerns a generated summary omits minor but mandatory report categories.
C:Robust instructions should track intent and policy constraints, not accidental surface phrasing; it serves a setting where an assistant's answers vary with rephrased task wording in the same operational class, not the present case where a generated summary omits minor but mandatory report categories.
Question 116
A model confuses source documents with formatting instructions when both appear in one block. Which change targets the constraint rather than a proxy metric?
- In the proposed design, specify explicit verifiable intermediate fields and deterministic checks in the output contract.
- Under the reviewed approach, delimit input data and instructions clearly using separated sections or XML-style tags.
- During architecture review, begin with a minimal zero-shot instruction and evaluate before adding complexity.
- For the intended users, test prompt robustness with paraphrases and strengthen the task-specific decision rubric.
Explanation
Correct Answer
B:Boundary markers clarify which text is evidence and which text governs output behavior; the implementation should produce evidence that injected instructions inside document sections remain treated as data during tests.
Incorrect Answers
A:Observable checks give auditors evidence without relying on opaque or speculative reasoning transcripts; it targets cases where a complex eligibility answer needs reproducible intermediate checks but not exposed private reasoning, while the immediate question involves a model confuses source documents with formatting instructions when both appear in one block.
C:Unnecessary examples increase context cost when direct instructions already meet the acceptance threshold; that belongs to situations where a straightforward document tagger has a stable schema and no unusual edge cases, whereas the current issue is that a model confuses source documents with formatting instructions when both appear in one block.
D:Robust instructions should track intent and policy constraints, not accidental surface phrasing; it is relevant if an assistant's answers vary with rephrased task wording in the same operational class, whereas the release criterion addresses a model confuses source documents with formatting instructions when both appear in one block.
Question 117
A complex task fails when requested in one step despite valid source materials. Which action is justified by the scenario?
- As a candidate control, decompose prompt stages into evidence selection, decision and final response.
- In the initial solution, use a structured checklist of required sections and evaluate completion.
- In the proposed design, provide a small, diverse set of labeled borderline examples with clear decision criteria.
- In the release design, begin with a minimal zero-shot instruction and evaluate before adding complexity.
Explanation
Correct Answer
A:Stagewise prompting exposes where a failure occurs and reduces competing objectives in one request; acceptance evidence must confirm that intermediate evaluation pinpoints evidence errors before wording evaluation.
Incorrect Answers
B:An explicit task structure increases coverage of mandatory dimensions without relying on length alone; that fits cases where a generated summary omits minor but mandatory report categories, but here a complex task fails when requested in one step despite valid source materials.
C:Few-shot examples demonstrate nuanced boundaries more concretely than repeated abstract labels; this control is needed when a classifier misunderstands rare borderline categories with a bare one-line instruction, instead of when a complex task fails when requested in one step despite valid source materials.
D:Unnecessary examples increase context cost when direct instructions already meet the acceptance threshold; that would be useful if a straightforward document tagger has a stable schema and no unusual edge cases, not the case in which a complex task fails when requested in one step despite valid source materials.
Question 118
An assistant's answers vary with rephrased task wording in the same operational class. What change directly resolves this production-design issue?
- In the proposed implementation, specify explicit verifiable intermediate fields and deterministic checks in the output contract.
- At this decision point, use a structured checklist of required sections and evaluate completion.
- As a candidate control, begin with a minimal zero-shot instruction and evaluate before adding complexity.
- In the initial solution, test prompt robustness with paraphrases and strengthen the task-specific decision rubric.
Explanation
Correct Answer
D:Robust instructions should track intent and policy constraints, not accidental surface phrasing; reviewers would verify that a paraphrase suite yields stable decisions across equivalent instructions.
Incorrect Answers
A:Observable checks give auditors evidence without relying on opaque or speculative reasoning transcripts; it is justified if a complex eligibility answer needs reproducible intermediate checks but not exposed private reasoning, but the requested boundary concerns an assistant's answers vary with rephrased task wording in the same operational class.
B:An explicit task structure increases coverage of mandatory dimensions without relying on length alone; its scope includes cases where a generated summary omits minor but mandatory report categories, not the present case where an assistant's answers vary with rephrased task wording in the same operational class.
C:Unnecessary examples increase context cost when direct instructions already meet the acceptance threshold; it targets cases where a straightforward document tagger has a stable schema and no unusual edge cases, while this problem involves an assistant's answers vary with rephrased task wording in the same operational class.
Question 119
An edge-category classifier succeeds on ordinary cases but flips decisions at a narrowly defined threshold. Which finding should the release gate require?
- The reviewer observes that a paraphrase suite yields stable decisions across equivalent instructions.
- The release report verifies that the minimal prompt matches baseline quality at lower token usage.
- The controlled replay verifies that holdout borderline cases improve without overfitting to the few-shot exemplars.
- A reproducible run shows that intermediate evaluation pinpoints evidence errors before wording evaluation.
Explanation
Correct Answer
C:Few-shot examples demonstrate nuanced boundaries more concretely than repeated abstract labels; the evidence must be robust when a classifier misunderstands rare borderline categories with a bare one-line instruction.
Incorrect Answers
A:Robust instructions should track intent and policy constraints, not accidental surface phrasing; this technique fits where an assistant's answers vary with rephrased task wording in the same operational class, but the present constraint is that an edge-category classifier succeeds on ordinary cases but flips decisions at a narrowly defined threshold.
B:Unnecessary examples increase context cost when direct instructions already meet the acceptance threshold; the trigger would be that a straightforward document tagger has a stable schema and no unusual edge cases, whereas the current issue is that an edge-category classifier succeeds on ordinary cases but flips decisions at a narrowly defined threshold.
D:Stagewise prompting exposes where a failure occurs and reduces competing objectives in one request; that fits cases where a complex task fails when requested in one step despite valid source materials, while the required safeguard concerns an edge-category classifier succeeds on ordinary cases but flips decisions at a narrowly defined threshold.
Question 120
A lightweight tagging benchmark is inflated by demonstrations copied from the test set. Which verified outcome most directly addresses the reported risk?
- A production-like test shows that a paraphrase suite yields stable decisions across equivalent instructions.
- The release check confirms that auditors can recompute each eligibility condition from approved fields.
- The reviewer observes that holdout borderline cases improve without overfitting to the few-shot exemplars.
- The scenario test demonstrates that the minimal prompt matches baseline quality at lower token usage.
Explanation
Correct Answer
D:Unnecessary examples increase context cost when direct instructions already meet the acceptance threshold; this checks the situation in which a straightforward document tagger has a stable schema and no unusual edge cases.
Incorrect Answers
A:Robust instructions should track intent and policy constraints, not accidental surface phrasing; this applies when an assistant's answers vary with rephrased task wording in the same operational class, but here a lightweight tagging benchmark is inflated by demonstrations copied from the test set.
B:Observable checks give auditors evidence without relying on opaque or speculative reasoning transcripts; the alternative helps if a complex eligibility answer needs reproducible intermediate checks but not exposed private reasoning, instead of when a lightweight tagging benchmark is inflated by demonstrations copied from the test set.
C:Few-shot examples demonstrate nuanced boundaries more concretely than repeated abstract labels; it is justified if a classifier misunderstands rare borderline categories with a bare one-line instruction, not the case in which a lightweight tagging benchmark is inflated by demonstrations copied from the test set.