Question 340 A candidate model's card warns against personal-data profiling, but the proposed use case is automatic employee risk scoring. Which design change directly satisfies the stated constraints? Treat the benchmark as insufficient, run an Indonesian task evaluation and prefer a model with proven target-language behavior. Select an embedding model whose input limit fits the chunk budget or split sections at meaningful boundaries before encoding. Exclude the model from that disallowed use and perform a policy/legal review before considering any compliant alternative. Reject the incompatible model and choose an adequately performing candidate with a license permitting the intended commercial use. ExplanationCorrect Answer C:The action targets the stated cause: Published use restrictions and responsible-use limits are constraints, not optional annotations overcome by prompting. Check the repair by: audit the model-selection register for allowed use and verify approval references before enabling the scoring workflow. Other Answers A:Published use restrictions and responsible-use limits are constraints, not optional annotations overcome by prompting. B:Published use restrictions and responsible-use limits are constraints, not optional annotations overcome by prompting. D:Published use restrictions and responsible-use limits are constraints, not optional annotations overcome by prompting.
Question 341 A candidate model's card warns against personal-data profiling, but the proposed use case is automatic employee risk scoring. Which causal explanation best accounts for the failure? Embedding input truncation drops semantic evidence, creating vectors that omit later constraints and citations. Model-card benchmark coverage does not automatically generalize to untested languages or task distributions. Published use restrictions and responsible-use limits are constraints, not optional annotations overcome by prompting. A model card's license and permitted uses constrain deployment independently of benchmark quality. ExplanationCorrect Answer C:Exclude the model from that disallowed use and perform a policy/legal review before considering any compliant alternative. Other Answers A:That mechanism describes another problem: Technical manuals contain 5,000-token sections, but the selected embedding endpoint accepts only much shorter inputs and silently truncates endings. It does not explain this case, where published use restrictions and responsible-use limits are constraints, not optional annotations overcome by prompting. B:Published use restrictions and responsible-use limits are constraints, not optional annotations overcome by prompting. D:Published use restrictions and responsible-use limits are constraints, not optional annotations overcome by prompting.
Question 342 A candidate model's card warns against personal-data profiling, but the proposed use case is automatic employee risk scoring. Which concrete acceptance test best demonstrates the proposed change works? Compare source token counts with endpoint limits and query for details near the end of long sections. Check the exact published license terms and record the selection decision alongside measured task performance. Compare response correctness on a labeled Indonesian set and document coverage gaps from the model card. Audit the model-selection register for allowed use and verify approval references before enabling the scoring workflow. ExplanationCorrect Answer D:Published use restrictions and responsible-use limits are constraints, not optional annotations overcome by prompting. Other Answers A:Audit the model-selection register for allowed use and verify approval references before enabling the scoring workflow. B:Audit the model-selection register for allowed use and verify approval references before enabling the scoring workflow. C:Audit the model-selection register for allowed use and verify approval references before enabling the scoring workflow.
Question 343 A candidate model's card warns against personal-data profiling, but the proposed use case is automatic employee risk scoring. Which operational practice should be made a durable requirement? Require license and permitted-use review in every model-catalog promotion checklist. Add prohibited-use screening and documented human governance to the model registry intake process. Validate token lengths before embedding and maintain model-specific chunking profiles. Maintain target-language evidence requirements and uncertainty flags during model selection. ExplanationCorrect Answer B:Published use restrictions and responsible-use limits are constraints, not optional annotations overcome by prompting. Other Answers A:That published use restrictions and responsible-use limits are constraints, not optional annotations overcome by prompting. C:Published use restrictions and responsible-use limits are constraints, not optional annotations overcome by prompting. D:This control addresses a different exposure: A model listing reports excellent English QA scores but no evaluation on the company's Indonesian customer-support language. It would not prevent this failure, because published use restrictions and responsible-use limits are constraints, not optional annotations overcome by prompting.
Question 344 A new prompt version improves BLEU-like overlap with expected summaries but increases unsupported legal claims. What should the implementation team change first to resolve the described defect? Create a held-out labeled evaluation set that did not drive prompt tuning and compare versions without leakage. Compare factual groundedness and error severity alongside lexical overlap; reject the model change until unsupported claims are addressed. Log tool result identifiers and prompt context spans in the trace, then use recorded traces to review the failing branches. Measure quality, P95 response time and cost on the same stratified workload, then select the model meeting the published SLO. ExplanationCorrect Answer B:The action targets the stated cause: Surface-form similarity cannot establish factual entailment or safety in free-form generated explanations. Check the repair by: use adjudicated summary cases to score citation support and hallucinations separately from n-gram overlap. Other Answers A:Surface-form similarity cannot establish factual entailment or safety in free-form generated explanations. C:Surface-form similarity cannot establish factual entailment or safety in free-form generated explanations. D:Surface-form similarity cannot establish factual entailment or safety in free-form generated explanations.
Question 345 A new prompt version improves BLEU-like overlap with expected summaries but increases unsupported legal claims. Why does the current approach fail under these conditions? Averages conceal slow-tail behavior; model comparison must use comparable traffic and agreed operational thresholds. Evaluation leakage inflates metrics by testing memorized or overfitted examples rather than generalization. Trace context provides causality between retrieval, tool execution and final generation; isolated logs cannot reconstruct the path. Surface-form similarity cannot establish factual entailment or safety in free-form generated explanations. ExplanationCorrect Answer D:Compare factual groundedness and error severity alongside lexical overlap; reject the model change until unsupported claims are addressed. Other Answers A:Surface-form similarity cannot establish factual entailment or safety in free-form generated explanations. B:Surface-form similarity cannot establish factual entailment or safety in free-form generated explanations. C:That mechanism describes another problem: A multi-tool agent is returning unpredictable answers because intermediate tool results are not linked to the generated text in its MLflow experiment. It does not explain this case, where surface-form similarity cannot establish factual entailment or safety in free-form generated explanations.
Question 346 A new prompt version improves BLEU-like overlap with expected summaries but increases unsupported legal claims. Which observation would establish that the requirement is now met? Select a failed response in MLflow and navigate to the exact source/tool span responsible for its unsupported content. Measure quality on unseen intent clusters and report confidence ranges or failure counts alongside point estimates. Run controlled A/B load tests with matched prompts and compare P50, P95, token use and error rate. Use adjudicated summary cases to score citation support and hallucinations separately from n-gram overlap. ExplanationCorrect Answer D:Surface-form similarity cannot establish factual entailment or safety in free-form generated explanations. Other Answers A:Use adjudicated summary cases to score citation support and hallucinations separately from n-gram overlap. B:Use adjudicated summary cases to score citation support and hallucinations separately from n-gram overlap. C:Use adjudicated summary cases to score citation support and hallucinations separately from n-gram overlap.
Question 347 A new prompt version improves BLEU-like overlap with expected summaries but increases unsupported legal claims. What lasting control best prevents recurrence of the identified issue? Require reproducible trace IDs and stable experiment tagging for each agent revision. Treat high-impact factual errors as hard acceptance criteria in experimental model comparisons. Keep evaluation datasets versioned and separated from optimization inputs. Require reproducible experiments and release gates over both task quality and tail latency. ExplanationCorrect Answer B:Surface-form similarity cannot establish factual entailment or safety in free-form generated explanations. Other Answers A:Surface-form similarity cannot establish factual entailment or safety in free-form generated explanations. C:This control addresses a different exposure: A prompt change is evaluated on examples copied from the training set and appears perfect, but fails new user intents. It would not prevent this failure, because surface-form similarity cannot establish factual entailment or safety in free-form generated explanations. D:That surface-form similarity cannot establish factual entailment or safety in free-form generated explanations.
Question 348 Two models have similar answer quality, but one exhibits twice the P95 latency on the organization's real workload. What is the most targeted corrective step for this observed failure? Measure quality, P95 response time and cost on the same stratified workload, then select the model meeting the published SLO. Compare factual groundedness and error severity alongside lexical overlap; reject the model change until unsupported claims are addressed. Create a held-out labeled evaluation set that did not drive prompt tuning and compare versions without leakage. Benchmark smaller candidate models for classification accuracy and P95 latency, selecting the least costly model meeting both targets. ExplanationCorrect Answer A:The action targets the stated cause: Averages conceal slow-tail behavior; model comparison must use comparable traffic and agreed operational thresholds. Check the repair by: run controlled A/B load tests with matched prompts and compare P50, P95, token use and error rate. Other Answers B:Averages conceal slow-tail behavior; model comparison must use comparable traffic and agreed operational thresholds. C:Averages conceal slow-tail behavior; model comparison must use comparable traffic and agreed operational thresholds. D:Averages conceal slow-tail behavior; model comparison must use comparable traffic and agreed operational thresholds.
Question 349 Two models have similar answer quality, but one exhibits twice the P95 latency on the organization's real workload. What is the decisive mechanism behind the observed result? Model size does not determine whether a simple constrained task needs expensive general reasoning capacity. Averages conceal slow-tail behavior; model comparison must use comparable traffic and agreed operational thresholds. Evaluation leakage inflates metrics by testing memorized or overfitted examples rather than generalization. Surface-form similarity cannot establish factual entailment or safety in free-form generated explanations. ExplanationCorrect Answer B:Measure quality, P95 response time and cost on the same stratified workload, then select the model meeting the published SLO. Other Answers A:Averages conceal slow-tail behavior; model comparison must use comparable traffic and agreed operational thresholds. C:That mechanism describes another problem: A prompt change is evaluated on examples copied from the training set and appears perfect, but fails new user intents. It does not explain this case, where averages conceal slow-tail behavior; model comparison must use comparable traffic and agreed operational thresholds. D:Averages conceal slow-tail behavior; model comparison must use comparable traffic and agreed operational thresholds.
Question 350 Two models have similar answer quality, but one exhibits twice the P95 latency on the organization's real workload. Which verification is strongest for this particular failure mode? Use adjudicated summary cases to score citation support and hallucinations separately from n-gram overlap. Run labeled classification and load tests with cost per thousand requests, macro-F1 and P95 latency. Measure quality on unseen intent clusters and report confidence ranges or failure counts alongside point estimates. Run controlled A/B load tests with matched prompts and compare P50, P95, token use and error rate. ExplanationCorrect Answer D:Averages conceal slow-tail behavior; model comparison must use comparable traffic and agreed operational thresholds. Other Answers A:Run controlled A/B load tests with matched prompts and compare P50, P95, token use and error rate. B:Run controlled A/B load tests with matched prompts and compare P50, P95, token use and error rate. C:Run controlled A/B load tests with matched prompts and compare P50, P95, token use and error rate.
