McAfee-Secured Website

Exam Code: AI-300

Exam Name: Operationalizing Machine Learning and Generative AI Solutions

Certification Provider: Microsoft

Corresponding Certification: Microsoft Certified: Machine Learning Operations Engineer Associate

100% Microsoft AI-300 Guaranteed Success With Testking Training Materials

Get Real AI-300 Exam Questions & Expert Verified Answers as Experienced on the Real Exam!

213 Questions & Answers with Testing Engine

"Operationalizing Machine Learning and Generative AI Solutions Exam", also known as AI-300 exam, is a Microsoft certification exam.

Pass your tests with the always up-to-date AI-300 Exam Engine. Your AI-300 training materials keep you at the head of the pack!

Satisfaction Guaranteed

Money Back Guarantee

Testking provides hassle-free money back guarantee with our products. That is because we have 100% trust in the abilities of our professional and experience product team, and our record is a proof of that.

99.6% PASS RATE
Was: $137.49
Now: $124.99

Microsoft AI-300 Practice Test 74

Question 730

A production hybrid endpoint starts timing out after a team maximizes text recall, k, and expensive semantic ranking simultaneously. What should be tuned first?

  1. Reduce expensive candidate and reranking settings and benchmark p95 latency before scaling.
  2. Interpret RRF scores on the hybrid scale and calibrate relevance cutoffs separately.
  3. Retain meaningful exact terms for BM25 while generating a compatible vector query.
  4. Adjust vector weighting and lexical candidate contribution using labeled exact-reference queries.

Explanation

Correct Answer

A:

Hybrid retrieval cost grows with parallel subqueries and reranking; indiscriminately maximizing every setting raises overload risk.

Incorrect Answers

B:

Interpret RRF scores on the hybrid scale and calibrate relevance cutoffs separately. It addresses another case: An engineer rejects all hybrid hits because their fused scores are around 0.03, below the vector-only threshold of 0.8. For that case, the distinguishing justification or acceptance check is: Fused rank scores have different numeric behavior from standalone vector similarity scores; thresholds cannot be borrowed directly. The present case instead states: A production hybrid endpoint starts timing out after a team maximizes text recall, k, and expensive semantic ranking simultaneously. It calls for Reduce expensive candidate and reranking settings and benchmark p95 latency before scaling. Showing success for the other case would not establish: Hybrid retrieval cost grows with parallel subqueries and reranking; indiscriminately maximizing every setting raises overload risk.

C:

The alternative, Retain meaningful exact terms for BM25 while generating a compatible vector query, belongs to a different technical condition: An index has excellent keyword matches but hybrid relevance worsens when all rare terms are stripped during query preprocessing. There the source-specific mechanism or proof is: Hybrid retrieval is valuable precisely because lexical signals and semantic representations contribute different information. The condition asked about here is: A production hybrid endpoint starts timing out after a team maximizes text recall, k, and expensive semantic ranking simultaneously. A suitable correction is Reduce expensive candidate and reranking settings and benchmark p95 latency before scaling; the targeted justification or result is: Hybrid retrieval cost grows with parallel subqueries and reranking; indiscriminately maximizing every setting raises overload risk.

D:

Adjust vector weighting and lexical candidate contribution using labeled exact-reference queries. That operation would be relevant if: The vector side dominates hybrid results even for precise regulation references. Its corresponding reason or verification is: Hybrid relevance can be tuned to preserve precise lexical matches instead of treating all semantic neighbors as equally important. However the problem in this question is: A production hybrid endpoint starts timing out after a team maximizes text recall, k, and expensive semantic ranking simultaneously. The requirement instead calls for Reduce expensive candidate and reranking settings and benchmark p95 latency before scaling; the question-specific check is: Hybrid retrieval cost grows with parallel subqueries and reranking; indiscriminately maximizing every setting raises overload risk.

Question 731

An RAG team claims recall improved but has no labels showing which passages actually answer its benchmark questions. What should they prepare?

  1. Create an independent query-to-relevant-document judgment set and calculate recall at k.
  2. Capture corpus and index snapshot identifiers with each evaluation run.
  3. Include unanswerable and contradictory-source questions with expected abstention behavior.
  4. Measure p95 latency, throttling, token cost, and task success under production-like concurrency.

Explanation

Correct Answer

A:

Retrieval recall requires known relevant sources; raw model confidence alone cannot establish coverage.

Incorrect Answers

B:

Capture corpus and index snapshot identifiers with each evaluation run. This is useful for the distinct issue: An enterprise knowledge index is updated weekly, but its evaluation results compare answers without tracking document snapshots. The relevant technical rationale or evidence for that issue is: Document changes can alter expected evidence; reproducible A/B comparisons require immutable retrieval-source versions. Here the reported issue is: An RAG team claims recall improved but has no labels showing which passages actually answer its benchmark questions. The relevant response is: Create an independent query-to-relevant-document judgment set and calculate recall at k. The first result cannot verify this different correction: Retrieval recall requires known relevant sources; raw model confidence alone cannot establish coverage.

C:

Include unanswerable and contradictory-source questions with expected abstention behavior. It addresses another case: A regulatory bot must refuse to invent answers when the source corpus has no applicable rule. For that case, the distinguishing justification or acceptance check is: A benchmark of only answerable questions cannot detect confident fabrication on missing or conflicting evidence. The present case instead states: An RAG team claims recall improved but has no labels showing which passages actually answer its benchmark questions. It calls for Create an independent query-to-relevant-document judgment set and calculate recall at k. Showing success for the other case would not establish: Retrieval recall requires known relevant sources; raw model confidence alone cannot establish coverage.

D:

The alternative, Measure p95 latency, throttling, token cost, and task success under production-like concurrency, belongs to a different technical condition: An offline RAG benchmark passes, but live requests trigger rate limits and slower answers. There the source-specific mechanism or proof is: Offline retrieval accuracy alone cannot establish operational reliability at actual service demand. The condition asked about here is: An RAG team claims recall improved but has no labels showing which passages actually answer its benchmark questions. A suitable correction is Create an independent query-to-relevant-document judgment set and calculate recall at k; the targeted justification or result is: Retrieval recall requires known relevant sources; raw model confidence alone cannot establish coverage.

Question 732

A generated answer quotes a policy correctly but cites an unrelated page. Which evaluation should expose this?

  1. Randomize assignment within comparable user cohorts and record fixed variant exposure.
  2. Stratify relevance and groundedness results by real user workflow and document category.
  3. Capture corpus and index snapshot identifiers with each evaluation run.
  4. Measure citation-to-claim support and source attribution correctness separately from answer fluency.

Explanation

Correct Answer

D:

A fluent correct-looking response is not properly grounded when the cited document does not support its statement.

Incorrect Answers

A:

Randomize assignment within comparable user cohorts and record fixed variant exposure. A separate source case supports that approach when: Two retrieval pipelines are tested on production users, but premium accounts all receive variant B. Conversion differs by account value. The supporting rationale or observable result is: Unbalanced account cohorts confound retrieval quality with different user populations and spending behavior. This question describes a different issue: A generated answer quotes a policy correctly but cites an unrelated page. Address it through Measure citation-to-claim support and source attribution correctness separately from answer fluency, with the required justification or evidence: A fluent correct-looking response is not properly grounded when the cited document does not support its statement.

B:

Stratify relevance and groundedness results by real user workflow and document category. This is useful for the distinct issue: A RAG test set contains only headquarters policies, but field workers use different procedures. The relevant technical rationale or evidence for that issue is: Aggregate scores can conceal important retrieval failures for less common but operationally critical user cohorts. Here the reported issue is: A generated answer quotes a policy correctly but cites an unrelated page. The relevant response is: Measure citation-to-claim support and source attribution correctness separately from answer fluency. The first result cannot verify this different correction: A fluent correct-looking response is not properly grounded when the cited document does not support its statement.

C:

Capture corpus and index snapshot identifiers with each evaluation run. It addresses another case: An enterprise knowledge index is updated weekly, but its evaluation results compare answers without tracking document snapshots. For that case, the distinguishing justification or acceptance check is: Document changes can alter expected evidence; reproducible A/B comparisons require immutable retrieval-source versions. The present case instead states: A generated answer quotes a policy correctly but cites an unrelated page. It calls for Measure citation-to-claim support and source attribution correctness separately from answer fluency. Showing success for the other case would not establish: A fluent correct-looking response is not properly grounded when the cited document does not support its statement.

Question 733

Two retrieval pipelines are tested on production users, but premium accounts all receive variant B. Conversion differs by account value. What experiment change is needed?

  1. Apply predefined safety guardrails and withhold promotion until risk thresholds are satisfied.
  2. Measure citation-to-claim support and source attribution correctness separately from answer fluency.
  3. Randomize assignment within comparable user cohorts and record fixed variant exposure.
  4. Evaluate generation groundedness separately from retrieval-hit metrics.

Explanation

Correct Answer

C:

Unbalanced account cohorts confound retrieval quality with different user populations and spending behavior.

Incorrect Answers

A:

Apply predefined safety guardrails and withhold promotion until risk thresholds are satisfied. That operation would be relevant if: Variant B yields better answer relevance but causes a sharp increase in unsafe responses. Its corresponding reason or verification is: Quality gains do not justify deployment when safety regression violates the experiment's acceptance criteria. However the problem in this question is: Two retrieval pipelines are tested on production users, but premium accounts all receive variant B. Conversion differs by account value. The requirement instead calls for Randomize assignment within comparable user cohorts and record fixed variant exposure; the question-specific check is: Unbalanced account cohorts confound retrieval quality with different user populations and spending behavior.

B:

Measure citation-to-claim support and source attribution correctness separately from answer fluency. A separate source case supports that approach when: A generated answer quotes a policy correctly but cites an unrelated page. The supporting rationale or observable result is: A fluent correct-looking response is not properly grounded when the cited document does not support its statement. This question describes a different issue: Two retrieval pipelines are tested on production users, but premium accounts all receive variant B. Conversion differs by account value. Address it through Randomize assignment within comparable user cohorts and record fixed variant exposure, with the required justification or evidence: Unbalanced account cohorts confound retrieval quality with different user populations and spending behavior.

D:

Evaluate generation groundedness separately from retrieval-hit metrics. This is useful for the distinct issue: A RAG answer is unsupported even though the correct document was retrieved and included in context. The relevant technical rationale or evidence for that issue is: Successful retrieval does not guarantee the model uses that evidence accurately when composing an answer. Here the reported issue is: Two retrieval pipelines are tested on production users, but premium accounts all receive variant B. Conversion differs by account value. The relevant response is: Randomize assignment within comparable user cohorts and record fixed variant exposure. The first result cannot verify this different correction: Unbalanced account cohorts confound retrieval quality with different user populations and spending behavior.

Question 734

A RAG answer is unsupported even though the correct document was retrieved and included in context. Which stage failed?

  1. Include unanswerable and contradictory-source questions with expected abstention behavior.
  2. Evaluate generation groundedness separately from retrieval-hit metrics.
  3. Measure p95 latency, throttling, token cost, and task success under production-like concurrency.
  4. Improve retrieval recall on the labeled gold set before attributing all failures to generation.

Explanation

Correct Answer

B:

Successful retrieval does not guarantee the model uses that evidence accurately when composing an answer.

Incorrect Answers

A:

The alternative, Include unanswerable and contradictory-source questions with expected abstention behavior, belongs to a different technical condition: A regulatory bot must refuse to invent answers when the source corpus has no applicable rule. There the source-specific mechanism or proof is: A benchmark of only answerable questions cannot detect confident fabrication on missing or conflicting evidence. The condition asked about here is: A RAG answer is unsupported even though the correct document was retrieved and included in context. A suitable correction is Evaluate generation groundedness separately from retrieval-hit metrics; the targeted justification or result is: Successful retrieval does not guarantee the model uses that evidence accurately when composing an answer.

C:

Measure p95 latency, throttling, token cost, and task success under production-like concurrency. That operation would be relevant if: An offline RAG benchmark passes, but live requests trigger rate limits and slower answers. Its corresponding reason or verification is: Offline retrieval accuracy alone cannot establish operational reliability at actual service demand. However the problem in this question is: A RAG answer is unsupported even though the correct document was retrieved and included in context. The requirement instead calls for Evaluate generation groundedness separately from retrieval-hit metrics; the question-specific check is: Successful retrieval does not guarantee the model uses that evidence accurately when composing an answer.

D:

Improve retrieval recall on the labeled gold set before attributing all failures to generation. A separate source case supports that approach when: An evaluation identifies missing correct documents in retrieval, while answers often appear fabricated. The supporting rationale or observable result is: A generator cannot reliably cite evidence absent from the context supplied by the retrieval system. This question describes a different issue: A RAG answer is unsupported even though the correct document was retrieved and included in context. Address it through Evaluate generation groundedness separately from retrieval-hit metrics, with the required justification or evidence: Successful retrieval does not guarantee the model uses that evidence accurately when composing an answer.

Question 735

An evaluation identifies missing correct documents in retrieval, while answers often appear fabricated. What should be fixed first?

  1. Improve retrieval recall on the labeled gold set before attributing all failures to generation.
  2. Capture corpus and index snapshot identifiers with each evaluation run.
  3. Apply predefined safety guardrails and withhold promotion until risk thresholds are satisfied.
  4. Include unanswerable and contradictory-source questions with expected abstention behavior.

Explanation

Correct Answer

A:

A generator cannot reliably cite evidence absent from the context supplied by the retrieval system.

Incorrect Answers

B:

Capture corpus and index snapshot identifiers with each evaluation run. It addresses another case: An enterprise knowledge index is updated weekly, but its evaluation results compare answers without tracking document snapshots. For that case, the distinguishing justification or acceptance check is: Document changes can alter expected evidence; reproducible A/B comparisons require immutable retrieval-source versions. The present case instead states: An evaluation identifies missing correct documents in retrieval, while answers often appear fabricated. It calls for Improve retrieval recall on the labeled gold set before attributing all failures to generation. Showing success for the other case would not establish: A generator cannot reliably cite evidence absent from the context supplied by the retrieval system.

C:

The alternative, Apply predefined safety guardrails and withhold promotion until risk thresholds are satisfied, belongs to a different technical condition: Variant B yields better answer relevance but causes a sharp increase in unsafe responses. There the source-specific mechanism or proof is: Quality gains do not justify deployment when safety regression violates the experiment's acceptance criteria. The condition asked about here is: An evaluation identifies missing correct documents in retrieval, while answers often appear fabricated. A suitable correction is Improve retrieval recall on the labeled gold set before attributing all failures to generation; the targeted justification or result is: A generator cannot reliably cite evidence absent from the context supplied by the retrieval system.

D:

Include unanswerable and contradictory-source questions with expected abstention behavior. That operation would be relevant if: A regulatory bot must refuse to invent answers when the source corpus has no applicable rule. Its corresponding reason or verification is: A benchmark of only answerable questions cannot detect confident fabrication on missing or conflicting evidence. However the problem in this question is: An evaluation identifies missing correct documents in retrieval, while answers often appear fabricated. The requirement instead calls for Improve retrieval recall on the labeled gold set before attributing all failures to generation; the question-specific check is: A generator cannot reliably cite evidence absent from the context supplied by the retrieval system.

Question 736

A regulatory bot must refuse to invent answers when the source corpus has no applicable rule. Which cases must be in the test suite?

  1. Capture corpus and index snapshot identifiers with each evaluation run.
  2. Measure p95 latency, throttling, token cost, and task success under production-like concurrency.
  3. Measure citation-to-claim support and source attribution correctness separately from answer fluency.
  4. Include unanswerable and contradictory-source questions with expected abstention behavior.

Explanation

Correct Answer

D:

A benchmark of only answerable questions cannot detect confident fabrication on missing or conflicting evidence.

Incorrect Answers

A:

Capture corpus and index snapshot identifiers with each evaluation run. This is useful for the distinct issue: An enterprise knowledge index is updated weekly, but its evaluation results compare answers without tracking document snapshots. The relevant technical rationale or evidence for that issue is: Document changes can alter expected evidence; reproducible A/B comparisons require immutable retrieval-source versions. Here the reported issue is: A regulatory bot must refuse to invent answers when the source corpus has no applicable rule. The relevant response is: Include unanswerable and contradictory-source questions with expected abstention behavior. The first result cannot verify this different correction: A benchmark of only answerable questions cannot detect confident fabrication on missing or conflicting evidence.

B:

Measure p95 latency, throttling, token cost, and task success under production-like concurrency. It addresses another case: An offline RAG benchmark passes, but live requests trigger rate limits and slower answers. For that case, the distinguishing justification or acceptance check is: Offline retrieval accuracy alone cannot establish operational reliability at actual service demand. The present case instead states: A regulatory bot must refuse to invent answers when the source corpus has no applicable rule. It calls for Include unanswerable and contradictory-source questions with expected abstention behavior. Showing success for the other case would not establish: A benchmark of only answerable questions cannot detect confident fabrication on missing or conflicting evidence.

C:

The alternative, Measure citation-to-claim support and source attribution correctness separately from answer fluency, belongs to a different technical condition: A generated answer quotes a policy correctly but cites an unrelated page. There the source-specific mechanism or proof is: A fluent correct-looking response is not properly grounded when the cited document does not support its statement. The condition asked about here is: A regulatory bot must refuse to invent answers when the source corpus has no applicable rule. A suitable correction is Include unanswerable and contradictory-source questions with expected abstention behavior; the targeted justification or result is: A benchmark of only answerable questions cannot detect confident fabrication on missing or conflicting evidence.

Question 737

Variant B yields better answer relevance but causes a sharp increase in unsafe responses. What rollout decision fits?

  1. Capture corpus and index snapshot identifiers with each evaluation run.
  2. Randomize assignment within comparable user cohorts and record fixed variant exposure.
  3. Apply predefined safety guardrails and withhold promotion until risk thresholds are satisfied.
  4. Run a versioned regression evaluation on sampled production traces with a rollback threshold.

Explanation

Correct Answer

C:

Quality gains do not justify deployment when safety regression violates the experiment's acceptance criteria.

Incorrect Answers

A:

Capture corpus and index snapshot identifiers with each evaluation run. A separate source case supports that approach when: An enterprise knowledge index is updated weekly, but its evaluation results compare answers without tracking document snapshots. The supporting rationale or observable result is: Document changes can alter expected evidence; reproducible A/B comparisons require immutable retrieval-source versions. This question describes a different issue: Variant B yields better answer relevance but causes a sharp increase in unsafe responses. Address it through Apply predefined safety guardrails and withhold promotion until risk thresholds are satisfied, with the required justification or evidence: Quality gains do not justify deployment when safety regression violates the experiment's acceptance criteria.

B:

Randomize assignment within comparable user cohorts and record fixed variant exposure. This is useful for the distinct issue: Two retrieval pipelines are tested on production users, but premium accounts all receive variant B. Conversion differs by account value. The relevant technical rationale or evidence for that issue is: Unbalanced account cohorts confound retrieval quality with different user populations and spending behavior. Here the reported issue is: Variant B yields better answer relevance but causes a sharp increase in unsafe responses. The relevant response is: Apply predefined safety guardrails and withhold promotion until risk thresholds are satisfied. The first result cannot verify this different correction: Quality gains do not justify deployment when safety regression violates the experiment's acceptance criteria.

D:

Run a versioned regression evaluation on sampled production traces with a rollback threshold. It addresses another case: A rollout changes the retriever while preserving the prompt; the answer-quality metric drops after launch. For that case, the distinguishing justification or acceptance check is: Continuous evaluation can detect retrieval-induced answer degradation after the offline acceptance test. The present case instead states: Variant B yields better answer relevance but causes a sharp increase in unsafe responses. It calls for Apply predefined safety guardrails and withhold promotion until risk thresholds are satisfied. Showing success for the other case would not establish: Quality gains do not justify deployment when safety regression violates the experiment's acceptance criteria.

Question 738

A RAG test set contains only headquarters policies, but field workers use different procedures. Which evaluation design reduces this blind spot?

  1. Randomize assignment within comparable user cohorts and record fixed variant exposure.
  2. Stratify relevance and groundedness results by real user workflow and document category.
  3. Improve retrieval recall on the labeled gold set before attributing all failures to generation.
  4. Evaluate generation groundedness separately from retrieval-hit metrics.

Explanation

Correct Answer

B:

Aggregate scores can conceal important retrieval failures for less common but operationally critical user cohorts.

Incorrect Answers

A:

Randomize assignment within comparable user cohorts and record fixed variant exposure. That operation would be relevant if: Two retrieval pipelines are tested on production users, but premium accounts all receive variant B. Conversion differs by account value. Its corresponding reason or verification is: Unbalanced account cohorts confound retrieval quality with different user populations and spending behavior. However the problem in this question is: A RAG test set contains only headquarters policies, but field workers use different procedures. The requirement instead calls for Stratify relevance and groundedness results by real user workflow and document category; the question-specific check is: Aggregate scores can conceal important retrieval failures for less common but operationally critical user cohorts.

C:

Improve retrieval recall on the labeled gold set before attributing all failures to generation. A separate source case supports that approach when: An evaluation identifies missing correct documents in retrieval, while answers often appear fabricated. The supporting rationale or observable result is: A generator cannot reliably cite evidence absent from the context supplied by the retrieval system. This question describes a different issue: A RAG test set contains only headquarters policies, but field workers use different procedures. Address it through Stratify relevance and groundedness results by real user workflow and document category, with the required justification or evidence: Aggregate scores can conceal important retrieval failures for less common but operationally critical user cohorts.

D:

Evaluate generation groundedness separately from retrieval-hit metrics. This is useful for the distinct issue: A RAG answer is unsupported even though the correct document was retrieved and included in context. The relevant technical rationale or evidence for that issue is: Successful retrieval does not guarantee the model uses that evidence accurately when composing an answer. Here the reported issue is: A RAG test set contains only headquarters policies, but field workers use different procedures. The relevant response is: Stratify relevance and groundedness results by real user workflow and document category. The first result cannot verify this different correction: Aggregate scores can conceal important retrieval failures for less common but operationally critical user cohorts.

Question 739

A team changes chunk sizes and embedding models at the same time, then cannot attribute recall improvement. What experimental control is missing?

  1. Change one retrieval factor at a time or use a factorial design with recorded configurations.
  2. Score evidence relevance and version agreement for each retrieved citation.
  3. Include unanswerable and contradictory-source questions with expected abstention behavior.
  4. Run a versioned regression evaluation on sampled production traces with a rollback threshold.

Explanation

Correct Answer

A:

Controlled comparisons identify which specific tuning choice improves relevance instead of conflating unrelated modifications.

Incorrect Answers

B:

The alternative, Score evidence relevance and version agreement for each retrieved citation, belongs to a different technical condition: A support agent returns accurate facts but often cites passages that discuss a different product release. There the source-specific mechanism or proof is: Answers can be broadly correct yet inappropriate for the requested release when sources refer to other versions. The condition asked about here is: A team changes chunk sizes and embedding models at the same time, then cannot attribute recall improvement. A suitable correction is Change one retrieval factor at a time or use a factorial design with recorded configurations; the targeted justification or result is: Controlled comparisons identify which specific tuning choice improves relevance instead of conflating unrelated modifications.

C:

Include unanswerable and contradictory-source questions with expected abstention behavior. That operation would be relevant if: A regulatory bot must refuse to invent answers when the source corpus has no applicable rule. Its corresponding reason or verification is: A benchmark of only answerable questions cannot detect confident fabrication on missing or conflicting evidence. However the problem in this question is: A team changes chunk sizes and embedding models at the same time, then cannot attribute recall improvement. The requirement instead calls for Change one retrieval factor at a time or use a factorial design with recorded configurations; the question-specific check is: Controlled comparisons identify which specific tuning choice improves relevance instead of conflating unrelated modifications.

D:

Run a versioned regression evaluation on sampled production traces with a rollback threshold. A separate source case supports that approach when: A rollout changes the retriever while preserving the prompt; the answer-quality metric drops after launch. The supporting rationale or observable result is: Continuous evaluation can detect retrieval-induced answer degradation after the offline acceptance test. This question describes a different issue: A team changes chunk sizes and embedding models at the same time, then cannot attribute recall improvement. Address it through Change one retrieval factor at a time or use a factorial design with recorded configurations, with the required justification or evidence: Controlled comparisons identify which specific tuning choice improves relevance instead of conflating unrelated modifications.

Question 740

An enterprise knowledge index is updated weekly, but its evaluation results compare answers without tracking document snapshots. What must be versioned?

  1. Score evidence relevance and version agreement for each retrieved citation.
  2. Apply predefined safety guardrails and withhold promotion until risk thresholds are satisfied.
  3. Measure citation-to-claim support and source attribution correctness separately from answer fluency.
  4. Capture corpus and index snapshot identifiers with each evaluation run.

Explanation

Correct Answer

D:

Document changes can alter expected evidence; reproducible A/B comparisons require immutable retrieval-source versions.

Incorrect Answers

A:

The alternative, Score evidence relevance and version agreement for each retrieved citation, belongs to a different technical condition: A support agent returns accurate facts but often cites passages that discuss a different product release. There the source-specific mechanism or proof is: Answers can be broadly correct yet inappropriate for the requested release when sources refer to other versions. The condition asked about here is: An enterprise knowledge index is updated weekly, but its evaluation results compare answers without tracking document snapshots. A suitable correction is Capture corpus and index snapshot identifiers with each evaluation run; the targeted justification or result is: Document changes can alter expected evidence; reproducible A/B comparisons require immutable retrieval-source versions.

B:

Apply predefined safety guardrails and withhold promotion until risk thresholds are satisfied. That operation would be relevant if: Variant B yields better answer relevance but causes a sharp increase in unsafe responses. Its corresponding reason or verification is: Quality gains do not justify deployment when safety regression violates the experiment's acceptance criteria. However the problem in this question is: An enterprise knowledge index is updated weekly, but its evaluation results compare answers without tracking document snapshots. The requirement instead calls for Capture corpus and index snapshot identifiers with each evaluation run; the question-specific check is: Document changes can alter expected evidence; reproducible A/B comparisons require immutable retrieval-source versions.

C:

Measure citation-to-claim support and source attribution correctness separately from answer fluency. A separate source case supports that approach when: A generated answer quotes a policy correctly but cites an unrelated page. The supporting rationale or observable result is: A fluent correct-looking response is not properly grounded when the cited document does not support its statement. This question describes a different issue: An enterprise knowledge index is updated weekly, but its evaluation results compare answers without tracking document snapshots. Address it through Capture corpus and index snapshot identifiers with each evaluation run, with the required justification or evidence: Document changes can alter expected evidence; reproducible A/B comparisons require immutable retrieval-source versions.

Operationalizing Machine Learning and Generative AI Solutions (AI-300) Practice Tests