Question 550
An Azure AI Search system needs better ordering of already retrieved passages. The immediate design goal is to expand initial search recall when reranking appears unable to find the correct answer despite good ranking logic. Which approach addresses that goal?
- Semantic ranker activation
- Candidate set widening
- Semantic configuration content
- Reranking latency budget
Explanation
Correct Answer
B:A reranker can reorder only candidates it receives. Poor first-stage recall cannot be fixed by downstream scoring alone.
Incorrect Answers
A:Semantic ranking evaluates meaning and relevance among candidate documents; it operates after initial retrieval, so candidate recall remains important; the outstanding need is to expand initial search recall when reranking appears unable to find the correct answer despite good ranking logic.
C:Field configuration determines which text the reranker can assess; ranking a record with no useful textual content limits semantic judgment; the needed capability is to expand initial search recall when reranking appears unable to find the correct answer despite good ranking logic.
D:Reranking can improve relevance but adds processing work; its benefits should be balanced against end-to-end latency limits; the requirement, by contrast, is to expand initial search recall when reranking appears unable to find the correct answer despite good ranking logic.
Question 551
An Azure AI Search system needs better ordering of already retrieved passages. The team is required to measure the additional response time introduced by semantic ranking against the application's service objective. Which option is the best fit?
- Reranking A/B test
- Candidate set widening
- Reranking latency budget
- Reranker score analysis
Explanation
Correct Answer
C:Reranking can improve relevance but adds processing work. Its benefits should be balanced against end-to-end latency limits.
Incorrect Answers
A:Controlled comparison measures whether a second-stage ranker improves actual results; an isolated impressive example is not sufficient evidence; the requirement, by contrast, is to measure the additional response time introduced by semantic ranking against the application's service objective.
B:A reranker can reorder only candidates it receives; poor first-stage recall cannot be fixed by downstream scoring alone; the team would still need to measure the additional response time introduced by semantic ranking against the application's service objective.
D:Semantic scores represent a later ranking stage and need their own interpretation; they are not interchangeable with raw vector similarity values; the implementation under review needs to measure the additional response time introduced by semantic ranking against the application's service objective.
Question 552
An Azure AI Search system needs better ordering of already retrieved passages. An operator is tasked to compare semantic reranking on and off using the same query-to-document judgments. Which mechanism provides that capability?
- Reranking A/B test
- Semantic configuration content
- Reranker score analysis
- Reranking latency budget
Explanation
Correct Answer
A:Controlled comparison measures whether a second-stage ranker improves actual results. An isolated impressive example is not sufficient evidence.
Incorrect Answers
B:Field configuration determines which text the reranker can assess; ranking a record with no useful textual content limits semantic judgment; the requested task remains to compare semantic reranking on and off using the same query-to-document judgments.
C:Semantic scores represent a later ranking stage and need their own interpretation; they are not interchangeable with raw vector similarity values; this alternative does not achieve the requirement to compare semantic reranking on and off using the same query-to-document judgments.
D:Reranking can improve relevance but adds processing work; its benefits should be balanced against end-to-end latency limits; the actual design criterion is to compare semantic reranking on and off using the same query-to-document judgments.
Question 553
A RAG assistant serves internal documents with different access permissions. The immediate design goal is to filter retrievable documents to the permissions of the requesting user before including them as model context. Which approach addresses that goal?
- Cross-tenant isolation test
- Citation access validation
- Metadata filter expression
- Indexed access metadata
- Identity-aware security trimming
Explanation
Correct Answer
E:Security trimming enforces document-level access at retrieval. A prompt instruction not to reveal secrets is not an authorization boundary.
Incorrect Answers
A:Isolation tests exercise filters and identity mappings under adversarial conditions; strong semantic relevance does not authorize cross-tenant data access; the actual design criterion is to filter retrievable documents to the permissions of the requesting user before including them as model context.
B:Citation checks prevent the answer from leaking inaccessible source titles or snippets; formatting a valid-looking reference does not establish permission; the outstanding need is to filter retrievable documents to the permissions of the requesting user before including them as model context.
C:Structured filters reduce irrelevant or unauthorized results where indexed metadata supports them; a vector similarity score cannot replace an access rule; this approach leaves the team unable to filter retrievable documents to the permissions of the requesting user before including them as model context.
D:Access metadata enables filters tied to user entitlements; indexing content without its access scope risks cross-user disclosure; the needed capability is to filter retrievable documents to the permissions of the requesting user before including them as model context.
Question 554
A RAG assistant serves internal documents with different access permissions. The team is required to store the necessary document or group access information to support query-time permissions filtering. Which option is the best fit?
- Cross-tenant isolation test
- Citation access validation
- Source deletion synchronization
- Indexed access metadata
Explanation
Correct Answer
D:Access metadata enables filters tied to user entitlements. Indexing content without its access scope risks cross-user disclosure.
Incorrect Answers
A:Isolation tests exercise filters and identity mappings under adversarial conditions; strong semantic relevance does not authorize cross-tenant data access; the needed capability is to store the necessary document or group access information to support query-time permissions filtering.
B:Citation checks prevent the answer from leaking inaccessible source titles or snippets; formatting a valid-looking reference does not establish permission; the requirement, by contrast, is to store the necessary document or group access information to support query-time permissions filtering.
C:Index lifecycle synchronization prevents stale content remaining retrievable; the source system's deletion may not instantly remove a previously indexed copy; the team would still need to store the necessary document or group access information to support query-time permissions filtering.
Question 555
A RAG assistant serves internal documents with different access permissions. An operator is tasked to restrict search candidates to the correct tenant, document class, or approved time period before answer synthesis. Which mechanism provides that capability?
- Cross-tenant isolation test
- Source deletion synchronization
- Indexed access metadata
- Metadata filter expression
Explanation
Correct Answer
D:Structured filters reduce irrelevant or unauthorized results where indexed metadata supports them. A vector similarity score cannot replace an access rule.
Incorrect Answers
A:Isolation tests exercise filters and identity mappings under adversarial conditions; strong semantic relevance does not authorize cross-tenant data access; the specified outcome instead requires the team to restrict search candidates to the correct tenant, document class, or approved time period before answer synthesis.
B:Index lifecycle synchronization prevents stale content remaining retrievable; the source system's deletion may not instantly remove a previously indexed copy; the implementation under review needs to restrict search candidates to the correct tenant, document class, or approved time period before answer synthesis.
C:Access metadata enables filters tied to user entitlements; indexing content without its access scope risks cross-user disclosure; the requested task remains to restrict search candidates to the correct tenant, document class, or approved time period before answer synthesis.
Question 556
A RAG assistant serves internal documents with different access permissions. The immediate design goal is to remove or update indexed passages when their original restricted documents are deleted or permissions change. Which approach addresses that goal?
- Metadata filter expression
- Source deletion synchronization
- Identity-aware security trimming
- Citation access validation
Explanation
Correct Answer
B:Index lifecycle synchronization prevents stale content remaining retrievable. The source system's deletion may not instantly remove a previously indexed copy.
Incorrect Answers
A:Structured filters reduce irrelevant or unauthorized results where indexed metadata supports them; a vector similarity score cannot replace an access rule; this alternative does not achieve the requirement to remove or update indexed passages when their original restricted documents are deleted or permissions change.
C:Security trimming enforces document-level access at retrieval; a prompt instruction not to reveal secrets is not an authorization boundary; the outstanding need is to remove or update indexed passages when their original restricted documents are deleted or permissions change.
D:Citation checks prevent the answer from leaking inaccessible source titles or snippets; formatting a valid-looking reference does not establish permission; this approach leaves the team unable to remove or update indexed passages when their original restricted documents are deleted or permissions change.
Question 557
A RAG assistant serves internal documents with different access permissions. The team is required to ensure each cited document identifier refers to a passage the current user is authorized to see. Which option is the best fit?
- Identity-aware security trimming
- Indexed access metadata
- Citation access validation
- Source deletion synchronization
Explanation
Correct Answer
C:Citation checks prevent the answer from leaking inaccessible source titles or snippets. Formatting a valid-looking reference does not establish permission.
Incorrect Answers
A:Security trimming enforces document-level access at retrieval; a prompt instruction not to reveal secrets is not an authorization boundary; this approach leaves the team unable to ensure each cited document identifier refers to a passage the current user is authorized to see.
B:Access metadata enables filters tied to user entitlements; indexing content without its access scope risks cross-user disclosure; the needed capability is to ensure each cited document identifier refers to a passage the current user is authorized to see.
D:Index lifecycle synchronization prevents stale content remaining retrievable; the source system's deletion may not instantly remove a previously indexed copy; the team would still need to ensure each cited document identifier refers to a passage the current user is authorized to see.
Question 558
A RAG assistant serves internal documents with different access permissions. An operator is tasked to verify that users from one customer tenant cannot retrieve another tenant's indexed content. Which mechanism provides that capability?
- Citation access validation
- Identity-aware security trimming
- Metadata filter expression
- Cross-tenant isolation test
Explanation
Correct Answer
D:Isolation tests exercise filters and identity mappings under adversarial conditions. Strong semantic relevance does not authorize cross-tenant data access.
Incorrect Answers
A:Citation checks prevent the answer from leaking inaccessible source titles or snippets; formatting a valid-looking reference does not establish permission; the team would still need to verify that users from one customer tenant cannot retrieve another tenant's indexed content.
B:Security trimming enforces document-level access at retrieval; a prompt instruction not to reveal secrets is not an authorization boundary; the specified outcome instead requires the team to verify that users from one customer tenant cannot retrieve another tenant's indexed content.
C:Structured filters reduce irrelevant or unauthorized results where indexed metadata supports them; a vector similarity score cannot replace an access rule; the implementation under review needs to verify that users from one customer tenant cannot retrieve another tenant's indexed content.
Question 559
A search team must separate retrieval failure from generation failure in a low-quality assistant. The immediate design goal is to measure whether the known relevant passage appears among the top k retrieved candidates. Which approach addresses that goal?
- Retrieval precision at k
- Retrieval recall at k
- End-to-end answer success rate
- Grounded answer scoring
Explanation
Correct Answer
B:Recall at k tests evidence discovery. It does not judge whether the language model uses that evidence faithfully.
Incorrect Answers
A:Precision at k captures ranking noise within the evidence budget; it can decline as more marginal results are included; the requested task remains to measure whether the known relevant passage appears among the top k retrieved candidates.
C:End-to-end success combines evidence discovery and answer construction; a good individual search metric may not translate into an acceptable user outcome; the actual design criterion is to measure whether the known relevant passage appears among the top k retrieved candidates.
D:Groundedness evaluates answer synthesis against evidence; strong retrieval recall does not guarantee the model will avoid unsupported assertions; the outstanding need is to measure whether the known relevant passage appears among the top k retrieved candidates.
Question 560
A search-quality review finds that only four passages in the top ten retrieval results are marked relevant. Which metric reports the fraction of retrieved top-k candidates that are relevant, rather than the proportion of all relevant passages recovered?
- End-to-end answer success rate
- Retrieval precision at k
- Retrieval recall at k
- Grounded answer scoring
- Retrieval miss categorization
Explanation
Correct Answer
B:Precision at k captures ranking noise within the evidence budget. It can decline as more marginal results are included.
Incorrect Answers
A:End-to-end success combines evidence discovery and answer construction; a good individual search metric may not translate into an acceptable user outcome; the outstanding need is to measure the fraction of the top k candidates that are labeled relevant to the query.
C:Recall at k tests evidence discovery; it does not judge whether the language model uses that evidence faithfully; the needed capability is to measure the fraction of the top k candidates that are labeled relevant to the query.
D:Groundedness evaluates answer synthesis against evidence; strong retrieval recall does not guarantee the model will avoid unsupported assertions; the requirement, by contrast, is to measure the fraction of the top k candidates that are labeled relevant to the query.
E:Categorization links symptoms to the stage responsible for absent evidence; prompt rewriting alone cannot recover unindexed documents; the team would still need to measure the fraction of the top k candidates that are labeled relevant to the query.