Question 160 A product warranty assistant indexes public brochures and confidential supplier negotiations together without separating who may read each. What prevention measure is most targeted to this underlying cause? Track source owners, approval status, effective windows and retirement dates in the document catalog. Require document classification, ACL propagation and access-filter regression tests for every source refresh. Maintain a required-source coverage checklist tied to representative user intents before deployment. Treat access filtering as a pre-ranking invariant and test it with negative authorization fixtures. ExplanationCorrect Answer B:Corpus inclusion and access control determine what evidence a model can see; an answer prompt cannot reliably conceal already exposed confidential text. Other Answers A:That corpus inclusion and access control determine what evidence a model can see; an answer prompt cannot reliably conceal already exposed confidential text. C:Corpus inclusion and access control determine what evidence a model can see; an answer prompt cannot reliably conceal already exposed confidential text. D:This control addresses a different exposure: A re-ranker receives confidential chunks before caller entitlements are checked and places one unauthorized report first in the candidate list. It would not prevent this failure, because corpus inclusion and access control determine what evidence a model can see; an answer prompt cannot reliably conceal already exposed confidential text.
Question 161 A retrieval evaluation reports high precision because the top result is relevant, but eight of ten required policy clauses never appear. Which implementation decision best restores the intended behavior? Compare precision@k, recall@k, latency and token usage while tuning the cutoff; avoid treating higher recall as the sole goal. Use paraphrased and negative queries with relevance judgments, measuring hit ranks and false retrievals as well as easy exact matches. Curate the authoritative FAQ and safety notices as required retrieval sources, assigning explicit owner and review dates. Measure recall at the chosen top-k against a labeled set of all required clauses, not precision alone. ExplanationCorrect Answer D:The action targets the stated cause: Precision measures relevance among returned hits, whereas recall measures the fraction of relevant documents that the system actually found. Check the repair by: compute recall@k for known multi-clause questions and inspect missing relevant source IDs. Other Answers A:Precision measures relevance among returned hits, whereas recall measures the fraction of relevant documents that the system actually found. B:Precision measures relevance among returned hits, whereas recall measures the fraction of relevant documents that the system actually found. C:Precision measures relevance among returned hits, whereas recall measures the fraction of relevant documents that the system actually found.
Question 162 A retrieval evaluation reports high precision because the top result is relevant, but eight of ten required policy clauses never appear. Which underlying technical principle explains why this occurs? Retrieval quality is bounded by corpus completeness; relevant-looking promotional documents cannot replace missing authoritative guidance. In-sample lexical overlap overestimates semantic retrieval generalization and hides ambiguity or out-of-domain failures. Precision measures relevance among returned hits, whereas recall measures the fraction of relevant documents that the system actually found. Increasing k often captures additional relevant items at the cost of more irrelevant context and downstream generation expense. ExplanationCorrect Answer C:Measure recall at the chosen top-k against a labeled set of all required clauses, not precision alone. Other Answers A:Precision measures relevance among returned hits, whereas recall measures the fraction of relevant documents that the system actually found. B:Precision measures relevance among returned hits, whereas recall measures the fraction of relevant documents that the system actually found. D:That mechanism describes another problem: An index tuning exercise improves recall@20 but adds many irrelevant chunks, doubling the generator context and token bill. It does not explain this case, where precision measures relevance among returned hits, whereas recall measures the fraction of relevant documents that the system actually found.
Question 163 A retrieval evaluation reports high precision because the top result is relevant, but eight of ten required policy clauses never appear. After correction, what targeted test should demonstrate success? Compute recall@k for known multi-clause questions and inspect missing relevant source IDs. Test exception-related patient questions and confirm the validated FAQ plus safety notice are retrieved and cited. Benchmark held-out questions at several k values and record both answer grounding and total token spend. Run held-out paraphrases and unanswerable queries, then record top-k relevance judgments and abstention outcomes. ExplanationCorrect Answer A:Precision measures relevance among returned hits, whereas recall measures the fraction of relevant documents that the system actually found. Other Answers B:Compute recall@k for known multi-clause questions and inspect missing relevant source IDs. C:Compute recall@k for known multi-clause questions and inspect missing relevant source IDs. D:Compute recall@k for known multi-clause questions and inspect missing relevant source IDs.
Question 164 A retrieval evaluation reports high precision because the top result is relevant, but eight of ten required policy clauses never appear. Which release safeguard most directly addresses this failure class? Define a multi-metric retrieval acceptance budget and periodically retune top-k against representative traffic. Maintain a required-source coverage checklist tied to representative user intents before deployment. Track both recall and precision on a domain-stratified labeled retrieval set after index changes. Maintain independent query splits and adjudicated relevance labels rather than relying on source-text copies. ExplanationCorrect Answer C:Precision measures relevance among returned hits, whereas recall measures the fraction of relevant documents that the system actually found. Other Answers A:Precision measures relevance among returned hits, whereas recall measures the fraction of relevant documents that the system actually found. B:This control addresses a different exposure: A clinic information assistant finds detailed marketing copy but misses the approved treatment FAQ containing medical exceptions. It would not prevent this failure, because precision measures relevance among returned hits, whereas recall measures the fraction of relevant documents that the system actually found. D:That precision measures relevance among returned hits, whereas recall measures the fraction of relevant documents that the system actually found.
Question 165 An index tuning exercise improves recall@20 but adds many irrelevant chunks, doubling the generator context and token bill. Which design change directly satisfies the stated constraints? Budget instructions and output tokens first, then choose smaller relevant chunks to fit a measured total input limit. Compare precision@k, recall@k, latency and token usage while tuning the cutoff; avoid treating higher recall as the sole goal. Use paraphrased and negative queries with relevance judgments, measuring hit ranks and false retrievals as well as easy exact matches. Measure recall at the chosen top-k against a labeled set of all required clauses, not precision alone. ExplanationCorrect Answer B:The action targets the stated cause: Increasing k often captures additional relevant items at the cost of more irrelevant context and downstream generation expense. Check the repair by: benchmark held-out questions at several k values and record both answer grounding and total token spend. Other Answers A:Increasing k often captures additional relevant items at the cost of more irrelevant context and downstream generation expense. C:Increasing k often captures additional relevant items at the cost of more irrelevant context and downstream generation expense. D:Increasing k often captures additional relevant items at the cost of more irrelevant context and downstream generation expense.
Question 166 An index tuning exercise improves recall@20 but adds many irrelevant chunks, doubling the generator context and token bill. Which causal explanation best accounts for the failure? Retrieval chunk size, top-k and prompt overhead jointly consume context; a large per-chunk setting can still overflow the model. Increasing k often captures additional relevant items at the cost of more irrelevant context and downstream generation expense. In-sample lexical overlap overestimates semantic retrieval generalization and hides ambiguity or out-of-domain failures. Precision measures relevance among returned hits, whereas recall measures the fraction of relevant documents that the system actually found. ExplanationCorrect Answer B:Compare precision@k, recall@k, latency and token usage while tuning the cutoff; avoid treating higher recall as the sole goal. Other Answers A:Increasing k often captures additional relevant items at the cost of more irrelevant context and downstream generation expense. C:That mechanism describes another problem: A team declares vector-search quality perfect after testing only questions copied word-for-word from stored paragraph headings. It does not explain this case, where increasing k often captures additional relevant items at the cost of more irrelevant context and downstream generation expense. D:Increasing k often captures additional relevant items at the cost of more irrelevant context and downstream generation expense.
Question 167 An index tuning exercise improves recall@20 but adds many irrelevant chunks, doubling the generator context and token bill. Which concrete acceptance test best demonstrates the proposed change works? Benchmark held-out questions at several k values and record both answer grounding and total token spend. Record actual prompt token totals and verify the final cited evidence is present without truncation across worst-case queries. Run held-out paraphrases and unanswerable queries, then record top-k relevance judgments and abstention outcomes. Compute recall@k for known multi-clause questions and inspect missing relevant source IDs. ExplanationCorrect Answer A:Increasing k often captures additional relevant items at the cost of more irrelevant context and downstream generation expense. Other Answers B:Benchmark held-out questions at several k values and record both answer grounding and total token spend. C:Benchmark held-out questions at several k values and record both answer grounding and total token spend. D:Benchmark held-out questions at several k values and record both answer grounding and total token spend.
Question 168 An index tuning exercise improves recall@20 but adds many irrelevant chunks, doubling the generator context and token bill. Which operational practice should be made a durable requirement? Maintain independent query splits and adjudicated relevance labels rather than relying on source-text copies. Track both recall and precision on a domain-stratified labeled retrieval set after index changes. Enforce token-aware packing, alert on truncation and review maximum per-source chunk settings. Define a multi-metric retrieval acceptance budget and periodically retune top-k against representative traffic. ExplanationCorrect Answer D:Increasing k often captures additional relevant items at the cost of more irrelevant context and downstream generation expense. Other Answers A:This control addresses a different exposure: A team declares vector-search quality perfect after testing only questions copied word-for-word from stored paragraph headings. It would not prevent this failure, because increasing k often captures additional relevant items at the cost of more irrelevant context and downstream generation expense. B:That increasing k often captures additional relevant items at the cost of more irrelevant context and downstream generation expense. C:Increasing k often captures additional relevant items at the cost of more irrelevant context and downstream generation expense.
Question 169 A team declares vector-search quality perfect after testing only questions copied word-for-word from stored paragraph headings. What should the implementation team change first to resolve the described defect? Use paraphrased and negative queries with relevance judgments, measuring hit ranks and false retrievals as well as easy exact matches. Include the complete hierarchy of region, chapter and section headings with each chunk's text and metadata. Compare precision@k, recall@k, latency and token usage while tuning the cutoff; avoid treating higher recall as the sole goal. Measure recall at the chosen top-k against a labeled set of all required clauses, not precision alone. ExplanationCorrect Answer A:The action targets the stated cause: In-sample lexical overlap overestimates semantic retrieval generalization and hides ambiguity or out-of-domain failures. Check the repair by: run held-out paraphrases and unanswerable queries, then record top-k relevance judgments and abstention outcomes. Other Answers B:In-sample lexical overlap overestimates semantic retrieval generalization and hides ambiguity or out-of-domain failures. C:In-sample lexical overlap overestimates semantic retrieval generalization and hides ambiguity or out-of-domain failures. D:In-sample lexical overlap overestimates semantic retrieval generalization and hides ambiguity or out-of-domain failures.
Question 170 A team declares vector-search quality perfect after testing only questions copied word-for-word from stored paragraph headings. Why does the current approach fail under these conditions? In-sample lexical overlap overestimates semantic retrieval generalization and hides ambiguity or out-of-domain failures. Precision measures relevance among returned hits, whereas recall measures the fraction of relevant documents that the system actually found. Increasing k often captures additional relevant items at the cost of more irrelevant context and downstream generation expense. Headings create disambiguating context; removing ancestors yields near-identical embeddings for distinct jurisdiction-specific procedures. ExplanationCorrect Answer A:Use paraphrased and negative queries with relevance judgments, measuring hit ranks and false retrievals as well as easy exact matches. Other Answers B:That mechanism describes another problem: A retrieval evaluation reports high precision because the top result is relevant, but eight of ten required policy clauses never appear. It does not explain this case, where in-sample lexical overlap overestimates semantic retrieval generalization and hides ambiguity or out-of-domain failures. C:In-sample lexical overlap overestimates semantic retrieval generalization and hides ambiguity or out-of-domain failures. D:In-sample lexical overlap overestimates semantic retrieval generalization and hides ambiguity or out-of-domain failures.
