Microsoft AI-300 (Operationalizing Machine Learning and Generative AI Solutions) Practice Test 45

Question 440

An AI engineer configures test inputs and model outputs for Foundry evaluation workflows. The team is required to map captured model output to the evaluator field representing the answer being assessed. Which option is the best fit?

  1. Response field mapping
  2. Ground-truth field mapping
  3. Query field mapping
  4. Context field mapping

Explanation

Correct Answer

A:

Response mapping ensures quality judgments score the generated content. A wrong column may yield meaningless metrics despite a successful evaluation run.

Incorrect Answers

B:

Ground truth supports comparisons to a known target; it is different from the model's generated response and must not be fabricated from that response itself; this approach leaves the team unable to map captured model output to the evaluator field representing the answer being assessed.

C:

Correct query mapping gives evaluators the actual task request; passing the reference answer in its place would measure an unintended comparison; the needed capability is to map captured model output to the evaluator field representing the answer being assessed.

D:

Context mapping enables groundedness checks against actual supplied evidence; without retrieved context, the evaluator cannot judge attribution to that evidence; the requirement, by contrast, is to map captured model output to the evaluator field representing the answer being assessed.

Question 441

An AI engineer configures test inputs and model outputs for Foundry evaluation workflows. An operator is tasked to pass the retrieved supporting passages as context for evaluation of evidence-based answers. Which mechanism provides that capability?

  1. Response field mapping
  2. Ground-truth field mapping
  3. Evaluation schema check
  4. Context field mapping
  5. Conversation turn grouping

Explanation

Correct Answer

D:

Context mapping enables groundedness checks against actual supplied evidence. Without retrieved context, the evaluator cannot judge attribution to that evidence.

Incorrect Answers

A:

Response mapping ensures quality judgments score the generated content; a wrong column may yield meaningless metrics despite a successful evaluation run; the requirement, by contrast, is to pass the retrieved supporting passages as context for evaluation of evidence-based answers.

B:

Ground truth supports comparisons to a known target; it is different from the model's generated response and must not be fabricated from that response itself; the team would still need to pass the retrieved supporting passages as context for evaluation of evidence-based answers.

C:

Schema validation prevents malformed test inputs from contaminating metrics; it does not establish that the reference labels themselves are accurate; the specified outcome instead requires the team to pass the retrieved supporting passages as context for evaluation of evidence-based answers.

E:

Turn grouping captures the state and dependencies of a conversation; flattening without context can misattribute errors to the wrong turn; the requested task remains to pass the retrieved supporting passages as context for evaluation of evidence-based answers.

Question 442

An AI engineer configures test inputs and model outputs for Foundry evaluation workflows. The immediate design goal is to provide the validated expected answer where a reference-based evaluator requires one. Which approach addresses that goal?

  1. Conversation turn grouping
  2. Ground-truth field mapping
  3. Context field mapping
  4. Response field mapping

Explanation

Correct Answer

B:

Ground truth supports comparisons to a known target. It is different from the model's generated response and must not be fabricated from that response itself.

Incorrect Answers

A:

Turn grouping captures the state and dependencies of a conversation; flattening without context can misattribute errors to the wrong turn; the implementation under review needs to provide the validated expected answer where a reference-based evaluator requires one.

C:

Context mapping enables groundedness checks against actual supplied evidence; without retrieved context, the evaluator cannot judge attribution to that evidence; this alternative does not achieve the requirement to provide the validated expected answer where a reference-based evaluator requires one.

D:

Response mapping ensures quality judgments score the generated content; a wrong column may yield meaningless metrics despite a successful evaluation run; the actual design criterion is to provide the validated expected answer where a reference-based evaluator requires one.

Question 443

An AI engineer configures test inputs and model outputs for Foundry evaluation workflows. The team is required to preserve ordering and identifiers when scoring a multi-turn agent interaction rather than treating each utterance as unrelated. Which option is the best fit?

  1. Evaluation schema check
  2. Ground-truth field mapping
  3. Context field mapping
  4. Conversation turn grouping

Explanation

Correct Answer

D:

Turn grouping captures the state and dependencies of a conversation. Flattening without context can misattribute errors to the wrong turn.

Incorrect Answers

A:

Schema validation prevents malformed test inputs from contaminating metrics; it does not establish that the reference labels themselves are accurate; the actual design criterion is to preserve ordering and identifiers when scoring a multi-turn agent interaction rather than treating each utterance as unrelated.

B:

Ground truth supports comparisons to a known target; it is different from the model's generated response and must not be fabricated from that response itself; the outstanding need is to preserve ordering and identifiers when scoring a multi-turn agent interaction rather than treating each utterance as unrelated.

C:

Context mapping enables groundedness checks against actual supplied evidence; without retrieved context, the evaluator cannot judge attribution to that evidence; this approach leaves the team unable to preserve ordering and identifiers when scoring a multi-turn agent interaction rather than treating each utterance as unrelated.

Question 444

A Foundry evaluator expects a response column, but the test dataset stores model completions under a different column name. Some rows also omit required fields, and paying for model-as-judge scoring on malformed records is wasteful. Query and ground-truth mappings are already correct. Which TWO configuration steps resolve these issues? Choose TWO.

  1. Query field mapping
  2. Conversation turn grouping
  3. Evaluation schema check
  4. Response field mapping
  5. Ground-truth field mapping

Explanation

Correct Answers

C:

Schema validation prevents malformed test inputs from contaminating metrics; it does not establish that the reference labels themselves are accurate. Here, malformed evaluation rows are caught before paid scoring calls are made.

D:

Response mapping ensures quality judgments score the generated content; a wrong column may yield meaningless metrics despite a successful evaluation run. Here, the actual model completion is mapped to the expected response field.

Incorrect Answers

A:

Correct query mapping gives evaluators the actual task request; passing the reference answer in its place would measure an unintended comparison; this mechanism leaves unresolved response-column mapping and rejection of malformed evaluation rows.

B:

Turn grouping captures the state and dependencies of a conversation; flattening without context can misattribute errors to the wrong turn; it does not by itself address response-column mapping and rejection of malformed evaluation rows.

E:

Ground truth supports comparisons to a known target; it is different from the model's generated response and must not be fabricated from that response itself; the system would still need a remedy for response-column mapping and rejection of malformed evaluation rows.

Question 445

A retrieval-based assistant must answer from supplied enterprise documents rather than invent unsupported claims. The immediate design goal is to assess whether the generated answer's substantive statements are supported by the supplied retrieved context. Which approach addresses that goal?

  1. Citation-source alignment
  2. Claim-level evidence check
  3. Groundedness score
  4. Context sufficiency evaluation

Explanation

Correct Answer

C:

Groundedness measures evidence support. A fluent statement may still score poorly if the retrieved sources do not justify it.

Incorrect Answers

A:

Citation checking tests attribution accuracy rather than mere presence of source links; a plausible citation identifier can still point to irrelevant text; the specified outcome instead requires the team to assess whether the generated answer's substantive statements are supported by the supplied retrieved context.

B:

Claim-level review can identify hallucinations obscured by mostly correct paragraphs; one broad answer score may not reveal which statement lacks support; the implementation under review needs to assess whether the generated answer's substantive statements are supported by the supplied retrieved context.

D:

Context sufficiency assesses the retrieved evidence independently of answer wording; it distinguishes retrieval limitations from a generator that fails to use otherwise adequate supporting information; this alternative does not achieve the requirement to assess whether the generated answer's substantive statements are supported by the supplied retrieved context.

Question 446

A retrieval-based assistant must answer from supplied enterprise documents rather than invent unsupported claims. The team is required to inspect individual factual assertions in an answer to locate the unsupported portion. Which option is the best fit?

  1. Grounding failure root cause
  2. Citation-source alignment
  3. Groundedness score
  4. Claim-level evidence check

Explanation

Correct Answer

D:

Claim-level review can identify hallucinations obscured by mostly correct paragraphs. One broad answer score may not reveal which statement lacks support.

Incorrect Answers

A:

Root-cause analysis determines whether search failed to supply facts or generation ignored them; the corrective action differs between retrieval and prompting; this alternative does not achieve the requirement to inspect individual factual assertions in an answer to locate the unsupported portion.

B:

Citation checking tests attribution accuracy rather than mere presence of source links; a plausible citation identifier can still point to irrelevant text; the actual design criterion is to inspect individual factual assertions in an answer to locate the unsupported portion.

C:

Groundedness measures evidence support; a fluent statement may still score poorly if the retrieved sources do not justify it; the outstanding need is to inspect individual factual assertions in an answer to locate the unsupported portion.

Question 447

A retrieval-based assistant must answer from supplied enterprise documents rather than invent unsupported claims. An operator is tasked to verify that a cited document actually supports the sentence attached to the citation. Which mechanism provides that capability?

  1. Grounding failure root cause
  2. Claim-level evidence check
  3. Citation-source alignment
  4. Context sufficiency evaluation

Explanation

Correct Answer

C:

Citation checking tests attribution accuracy rather than mere presence of source links. A plausible citation identifier can still point to irrelevant text.

Incorrect Answers

A:

Root-cause analysis determines whether search failed to supply facts or generation ignored them; the corrective action differs between retrieval and prompting; this approach leaves the team unable to verify that a cited document actually supports the sentence attached to the citation.

B:

Claim-level review can identify hallucinations obscured by mostly correct paragraphs; one broad answer score may not reveal which statement lacks support; the needed capability is to verify that a cited document actually supports the sentence attached to the citation.

D:

Context sufficiency assesses the retrieved evidence independently of answer wording; it distinguishes retrieval limitations from a generator that fails to use otherwise adequate supporting information; the team would still need to verify that a cited document actually supports the sentence attached to the citation.

Question 448

The retrieval system returns no authoritative policy passage for a customer's question. The assistant is prohibited from inventing an answer when there is no supporting evidence. Which response behavior should be enforced?

  1. Groundedness score
  2. Claim-level evidence check
  3. Insufficient-context abstention
  4. Context sufficiency evaluation
  5. Grounding failure root cause

Explanation

Correct Answer

C:

Abstention prevents unsupported certainty when sources do not cover the request. It is preferable to generating a guessed policy as fact.

Incorrect Answers

A:

Groundedness measures evidence support; a fluent statement may still score poorly if the retrieved sources do not justify it; the team would still need to require the assistant to state that evidence is unavailable when retrieval returns no authoritative answer.

B:

Claim-level review can identify hallucinations obscured by mostly correct paragraphs; one broad answer score may not reveal which statement lacks support; the specified outcome instead requires the team to require the assistant to state that evidence is unavailable when retrieval returns no authoritative answer.

D:

Context sufficiency assesses the retrieved evidence independently of answer wording; it distinguishes retrieval limitations from a generator that fails to use otherwise adequate supporting information; the requested task remains to require the assistant to state that evidence is unavailable when retrieval returns no authoritative answer.

E:

Root-cause analysis determines whether search failed to supply facts or generation ignored them; the corrective action differs between retrieval and prompting; this alternative does not achieve the requirement to require the assistant to state that evidence is unavailable when retrieval returns no authoritative answer.

Question 449

A retrieval-based assistant must answer from supplied enterprise documents rather than invent unsupported claims. The team is required to evaluate whether the passages retrieved for a user question contain the evidence needed to answer before measuring the generator's claim-level faithfulness. Which option is the best fit?

  1. Insufficient-context abstention
  2. Context sufficiency evaluation
  3. Groundedness score
  4. Citation-source alignment

Explanation

Correct Answer

B:

Context sufficiency assesses the retrieved evidence independently of answer wording. It distinguishes retrieval limitations from a generator that fails to use otherwise adequate supporting information.

Incorrect Answers

A:

Abstention prevents unsupported certainty when sources do not cover the request; it is preferable to generating a guessed policy as fact; the requested task remains to evaluate whether the passages retrieved for a user question contain the evidence needed to answer before measuring the generator's claim-level faithfulness.

C:

Groundedness measures evidence support; a fluent statement may still score poorly if the retrieved sources do not justify it; the actual design criterion is to evaluate whether the passages retrieved for a user question contain the evidence needed to answer before measuring the generator's claim-level faithfulness.

D:

Citation checking tests attribution accuracy rather than mere presence of source links; a plausible citation identifier can still point to irrelevant text; the outstanding need is to evaluate whether the passages retrieved for a user question contain the evidence needed to answer before measuring the generator's claim-level faithfulness.

Question 450

An enterprise RAG assistant's overall groundedness score suddenly drops. The team needs to identify exactly which assertions lack evidence and determine whether the evidence was never retrieved or was present but ignored during response generation. Which TWO diagnostic methods reveal those distinct root causes? Choose TWO.

  1. Insufficient-context abstention
  2. Claim-level evidence check
  3. Context sufficiency evaluation
  4. Grounding failure root cause
  5. Citation-source alignment

Explanation

Correct Answers

B:

Claim-level review can identify hallucinations obscured by mostly correct paragraphs; one broad answer score may not reveal which statement lacks support. Here, the review identifies unsupported answer claims at the assertion level.

D:

Root-cause analysis determines whether search failed to supply facts or generation ignored them; the corrective action differs between retrieval and prompting. Here, the failure is categorized as missing retrieval evidence versus incorrect generation over available evidence.

Incorrect Answers

A:

Abstention prevents unsupported certainty when sources do not cover the request; it is preferable to generating a guessed policy as fact; it does not by itself address unsupported answer assertions and the retrieval-versus-generation failure split.

C:

Context sufficiency assesses the retrieved evidence independently of answer wording; it distinguishes retrieval limitations from a generator that fails to use otherwise adequate supporting information; the immediate requirement instead concerns unsupported answer assertions and the retrieval-versus-generation failure split.

E:

Citation checking tests attribution accuracy rather than mere presence of source links; a plausible citation identifier can still point to irrelevant text; this mechanism leaves unresolved unsupported answer assertions and the retrieval-versus-generation failure split.

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