Microsoft AB-731 Practice Test 35
Question 340
At Solstice Retail, a research lead wants similar past cases retrieved before writing advice. Evidence: Similarity matching is distinct from composing a final recommendation. The business goal is to distinguish embeddings for retrieval from LLMs for generation. The proposed solution must respect documented authority and organizational controls. Which approach most directly meets the stated business objective within the approved authority?
- Add quality checks and structured table parsing for critical documents
- Index embeddings for matching and use a generator only after retrieval
- Select a generative text model for reviewed descriptions and reserve prediction for forecasts
- Route basic classification to an evaluated efficient model and complex exceptions upward
Explanation
Correct Answer
B:
Similarity matching is distinct from composing a final recommendation. Index embeddings for matching and use a generator only after retrieval.
Incorrect Answers
A:
“Add quality checks and structured table parsing for critical documents” is a design or implementation response to the condition: Corrupt source extraction can make even a grounded answer wrong. This question instead requires an action that can distinguish embeddings for retrieval from LLMs for generation. Similarity matching is distinct from composing a final recommendation. Index embeddings for matching and use a generator only after retrieval.
C:
“Select a generative text model for reviewed descriptions and reserve prediction for forecasts” is a design or implementation response to the condition: The deliverable is new natural-language copy, not a numeric sales prediction. This question instead requires an action that can distinguish embeddings for retrieval from LLMs for generation. Similarity matching is distinct from composing a final recommendation. Index embeddings for matching and use a generator only after retrieval.
D:
“Route basic classification to an evaluated efficient model and complex exceptions upward” is a design or implementation response to the condition: Simple tags may be handled by a lower-cost capable model. This question instead requires an action that can distinguish embeddings for retrieval from LLMs for generation. Similarity matching is distinct from composing a final recommendation. Index embeddings for matching and use a generator only after retrieval.
Question 341
At Solstice Retail, an AI-assisted employee service provides no way to challenge incorrect answers. Evidence: Users may be unable to correct harmful mistakes. The business goal is to create accessible escalation and correction procedures. The proposed solution must respect documented authority and organizational controls. Which approach most directly meets the stated business objective within the approved authority?
- Define stop/rollback, notification and correction procedures
- Include contractual safeguards, tests and incident obligations
- Set outcome targets and collect matched pilot comparisons
- Provide feedback capture, human review and response tracking
Explanation
Correct Answer
D:
Users may be unable to correct harmful mistakes. Provide feedback capture, human review and response tracking.
Incorrect Answers
A:
“Define stop/rollback, notification and correction procedures” is a design or implementation response to the condition: There is no owner or process for AI-specific incidents. This question instead requires an action that can create accessible escalation and correction procedures. Users may be unable to correct harmful mistakes. Provide feedback capture, human review and response tracking.
B:
“Include contractual safeguards, tests and incident obligations” is a design or implementation response to the condition: Contract data and prompts may be handled under unclear commitments. This question instead requires an action that can create accessible escalation and correction procedures. Users may be unable to correct harmful mistakes. Provide feedback capture, human review and response tracking.
C:
“Set outcome targets and collect matched pilot comparisons” is a design or implementation response to the condition: No customer quality, time savings or cost data were collected. This question instead requires an action that can create accessible escalation and correction procedures. Users may be unable to correct harmful mistakes. Provide feedback capture, human review and response tracking.
Question 342
At Orion Distribution, a research platform retrieves many superficially similar but irrelevant paragraphs. Evidence: First-stage keyword search has poor top-result precision. The business goal is to use semantic reranking when it measurably improves relevance. The proposed solution must respect documented authority and organizational controls. Which approach most directly meets the stated business objective within the approved authority?
- Add groundedness criteria to the model evaluation and release gate
- Shortlist eligible vision-language options and evaluate on sample images
- Test semantic ranker on candidate search responses
- Ask for a source-linked summary across relevant allowed records
Explanation
Correct Answer
C:
First-stage keyword search has poor top-result precision. Test semantic ranker on candidate search responses. Semantic reranking reorders an initial candidate set by contextual relevance. It can improve top results only if useful candidates were retrieved first and the measured quality gain justifies latency.
Incorrect Answers
A:
“Add groundedness criteria to the model evaluation and release gate” is a design or implementation response to the condition: Citations exist but conclusions are not entailed by the passages. This question instead requires an action that can use semantic reranking when it measurably improves relevance. First-stage keyword search has poor top-result precision. Test semantic ranker on candidate search responses. Semantic reranking reorders an initial candidate set by contextual relevance. It can improve top results only if useful candidates were retrieved first and the measured quality gain justifies latency.
B:
“Shortlist eligible vision-language options and evaluate on sample images” is a design or implementation response to the condition: Text-only models cannot inspect image evidence. This question instead requires an action that can use semantic reranking when it measurably improves relevance. First-stage keyword search has poor top-result precision. Test semantic ranker on candidate search responses. Semantic reranking reorders an initial candidate set by contextual relevance. It can improve top results only if useful candidates were retrieved first and the measured quality gain justifies latency.
D:
“Ask for a source-linked summary across relevant allowed records” is a design or implementation response to the condition: People, files and collaboration events have permissioned relationships. This question instead requires an action that can use semantic reranking when it measurably improves relevance. First-stage keyword search has poor top-result precision. Test semantic ranker on candidate search responses. Semantic reranking reorders an initial candidate set by contextual relevance. It can improve top results only if useful candidates were retrieved first and the measured quality gain justifies latency.
Question 343
At Ashwood Public Services, a bank proposes testing credit-risk fairness only on AI-generated applicants. Evidence: Synthetic samples may reproduce omissions and hide rare real-world issues. The business goal is to distinguish synthetic test utility from representative production evidence. The sponsor needs an evidence-based diagnosis rather than a guessed fix. Which initial investigation best identifies the controlling constraint before any solution is chosen?
- Inspect source ACLs and query identity propagation
- Compare synthetic cohorts with approved anonymized real-data distributions
- Measure document length, supported limits and missing critical sections
- Inspect input, output and retrieval tokens per request type
Explanation
Correct Answer
B:
Synthetic samples may reproduce omissions and hide rare real-world issues. Compare synthetic cohorts with approved anonymized real-data distributions. Synthetic examples are good for safe boundary testing, but generated cases may reproduce the omissions of the prompt. Keep a separately governed representative evaluation set for real-world coverage.
Incorrect Answers
A:
“Inspect source ACLs and query identity propagation” is a diagnostic step for the reported condition: Retrieval happens before checking the caller's permissions. This question instead requires an action that can distinguish synthetic test utility from representative production evidence. Synthetic samples may reproduce omissions and hide rare real-world issues. Compare synthetic cohorts with approved anonymized real-data distributions. Synthetic examples are good for safe boundary testing, but generated cases may reproduce the omissions of the prompt. Keep a separately governed representative evaluation set for real-world coverage.
C:
“Measure document length, supported limits and missing critical sections” is a diagnostic step for the reported condition: Critical terms from the oldest section may be omitted or truncated. This question instead requires an action that can distinguish synthetic test utility from representative production evidence. Synthetic samples may reproduce omissions and hide rare real-world issues. Compare synthetic cohorts with approved anonymized real-data distributions. Synthetic examples are good for safe boundary testing, but generated cases may reproduce the omissions of the prompt. Keep a separately governed representative evaluation set for real-world coverage.
D:
“Inspect input, output and retrieval tokens per request type” is a diagnostic step for the reported condition: Longer retrieved passages and responses increase processed input and output tokens. This question instead requires an action that can distinguish synthetic test utility from representative production evidence. Synthetic samples may reproduce omissions and hide rare real-world issues. Compare synthetic cohorts with approved anonymized real-data distributions. Synthetic examples are good for safe boundary testing, but generated cases may reproduce the omissions of the prompt. Keep a separately governed representative evaluation set for real-world coverage.
Question 344
At Lakeshore Commerce, a retail chatbot expects sharp holiday peaks. Evidence: Quota limits and token throughput may prevent desired traffic. The business goal is to plan capacity and graceful degradation around measured quotas. The project owner requires a measurable acceptance criterion before signoff. Which evidence would most convincingly demonstrate that the requirement has been met?
- Latency targets are met or users receive safe fallback guidance
- The restricted user does not receive protected excerpts
- Observed unit cost matches forecast within tolerance
- Chosen model meets quality and cost limits for its intended task
Explanation
Correct Answer
A:
Quota limits and token throughput may prevent desired traffic. Latency targets are met or users receive safe fallback guidance.
Incorrect Answers
B:
“The restricted user does not receive protected excerpts” validates an outcome in a case where the evidence is: The user lacks access even though the files exist in Microsoft 365. This question instead requires an action that can plan capacity and graceful degradation around measured quotas. Quota limits and token throughput may prevent desired traffic. Latency targets are met or users receive safe fallback guidance.
C:
“Observed unit cost matches forecast within tolerance” validates an outcome in a case where the evidence is: Input and output tokens and deployment type all affect billing. This question instead requires an action that can plan capacity and graceful degradation around measured quotas. Quota limits and token throughput may prevent desired traffic. Latency targets are met or users receive safe fallback guidance.
D:
“Chosen model meets quality and cost limits for its intended task” validates an outcome in a case where the evidence is: Candidate models differ in modality, region, cost and accuracy. This question instead requires an action that can plan capacity and graceful degradation around measured quotas. Quota limits and token throughput may prevent desired traffic. Latency targets are met or users receive safe fallback guidance.
Question 345
At Northlake Services, a procurement manager needs supplier status answers from one maintained knowledge source. Evidence: A prebuilt agent may already support search without bespoke code. The business goal is to choose the simplest governed agent route. Only non-sensitive synthetic data and approved sandbox systems may be used. Which bounded trial with synthetic records best tests the proposed approach?
- Test mock restricted files under permitted and denied identities
- Compare a fictional work-grounded summary against source documents
- Compare prebuilt answers against a mock custom prototype
- Create a harmless weekly sandbox summary and inspect next-run status
Explanation
Correct Answer
C:
A prebuilt agent may already support search without bespoke code. Compare prebuilt answers against a mock custom prototype. A supported prebuilt agent may already supply retrieval and administration features. Verify its permissions, limitations and support terms before adding unnecessary custom integration code.
Incorrect Answers
A:
“Test mock restricted files under permitted and denied identities” tests a proposed change for the condition: Index query results lack caller-specific authorization filters. This question instead requires an action that can choose the simplest governed agent route. A prebuilt agent may already support search without bespoke code. Compare prebuilt answers against a mock custom prototype. A supported prebuilt agent may already supply retrieval and administration features. Verify its permissions, limitations and support terms before adding unnecessary custom integration code.
B:
“Compare a fictional work-grounded summary against source documents” tests a proposed change for the condition: The answer depends on authorized Microsoft 365 work context. This question instead requires an action that can choose the simplest governed agent route. A prebuilt agent may already support search without bespoke code. Compare prebuilt answers against a mock custom prototype. A supported prebuilt agent may already supply retrieval and administration features. Verify its permissions, limitations and support terms before adding unnecessary custom integration code.
D:
“Create a harmless weekly sandbox summary and inspect next-run status” tests a proposed change for the condition: An ordinary chat response does not itself establish future recurring delivery. This question instead requires an action that can choose the simplest governed agent route. A prebuilt agent may already support search without bespoke code. Compare prebuilt answers against a mock custom prototype. A supported prebuilt agent may already supply retrieval and administration features. Verify its permissions, limitations and support terms before adding unnecessary custom integration code.
Question 346
At Orion Distribution, a board briefing lists a regulation and fabricated penalty figure. Evidence: The cited source does not contain the specific numerical assertion. The business goal is to require evidence verification before consequential use. The proposed solution must respect documented authority and organizational controls. Which approach most directly meets the stated business objective within the approved authority?
- Remove or qualify unsupported facts and request owner validation
- Create a scorecard with hard-stop risk thresholds
- Build a multi-year TCO comparison with sensitivity ranges
- Use validated forecasting analysis and label scenario assumptions
Explanation
Correct Answer
A:
The cited source does not contain the specific numerical assertion. Remove or qualify unsupported facts and request owner validation.
Incorrect Answers
B:
“Create a scorecard with hard-stop risk thresholds” is a design or implementation response to the condition: Data protection failures are not reflected in the accuracy metric. This question instead requires an action that can require evidence verification before consequential use. The cited source does not contain the specific numerical assertion. Remove or qualify unsupported facts and request owner validation.
C:
“Build a multi-year TCO comparison with sensitivity ranges” is a design or implementation response to the condition: Integration, human review, logging and employee training are omitted. This question instead requires an action that can require evidence verification before consequential use. The cited source does not contain the specific numerical assertion. Remove or qualify unsupported facts and request owner validation.
D:
“Use validated forecasting analysis and label scenario assumptions” is a design or implementation response to the condition: The model lacks an approved statistical forecast and uncertainty bounds. This question instead requires an action that can require evidence verification before consequential use. The cited source does not contain the specific numerical assertion. Remove or qualify unsupported facts and request owner validation.
Question 347
At Northlake Services, a retailer asks whether a generative chatbot should predict inventory demand. Evidence: Historical seasonality and promotions drive numerical forecasts. The business goal is to prefer evaluated forecasting models for quantitative predictions. Only non-sensitive synthetic data and approved sandbox systems may be used. Which bounded trial with synthetic records best tests the proposed approach?
- Test semantically related fictional cases with different wording
- Backtest fictional past quarters without future-data leakage
- Test a synthetic geography-constrained architecture review
- Compare retrieval on test questions before and after chunk changes
Explanation
Correct Answer
B:
Historical seasonality and promotions drive numerical forecasts. Backtest fictional past quarters without future-data leakage. Numerical forecasts require measured error against historical demand and known seasonal drivers. A fluent language-model narrative alone does not establish predictive accuracy or calibrated uncertainty.
Incorrect Answers
A:
“Test semantically related fictional cases with different wording” tests a proposed change for the condition: Similarity matching is distinct from composing a final recommendation. This question instead requires an action that can prefer evaluated forecasting models for quantitative predictions. Historical seasonality and promotions drive numerical forecasts. Backtest fictional past quarters without future-data leakage. Numerical forecasts require measured error against historical demand and known seasonal drivers. A fluent language-model narrative alone does not establish predictive accuracy or calibrated uncertainty.
C:
“Test a synthetic geography-constrained architecture review” tests a proposed change for the condition: Data location and transfer obligations may constrain service selection. This question instead requires an action that can prefer evaluated forecasting models for quantitative predictions. Historical seasonality and promotions drive numerical forecasts. Backtest fictional past quarters without future-data leakage. Numerical forecasts require measured error against historical demand and known seasonal drivers. A fluent language-model narrative alone does not establish predictive accuracy or calibrated uncertainty.
D:
“Compare retrieval on test questions before and after chunk changes” tests a proposed change for the condition: Large irrelevant text consumes context and obscures the answer. This question instead requires an action that can prefer evaluated forecasting models for quantitative predictions. Historical seasonality and promotions drive numerical forecasts. Backtest fictional past quarters without future-data leakage. Numerical forecasts require measured error against historical demand and known seasonal drivers. A fluent language-model narrative alone does not establish predictive accuracy or calibrated uncertainty.
Question 348
At Riverbend Publishing, a legal assistant needs current policy answers and consistent output formatting. Evidence: Fresh facts and style behavior are separate concerns. The business goal is to choose RAG for current facts and prompts or tuning for response style. The sponsor needs an evidence-based diagnosis rather than a guessed fix. Which initial investigation best identifies the controlling constraint before any solution is chosen?
- Distinguish stale source failures from tone or formatting failures
- Check chosen model, quota, region and data-processing terms
- Document gaps in available prebuilt capabilities and workflow needs
- Test sample queries by exact number and paraphrased policy terms
Explanation
Correct Answer
A:
Fresh facts and style behavior are separate concerns. Distinguish stale source failures from tone or formatting failures. Retrieval supplies changed factual sources at answer time, whereas prompt constraints or tuning address presentation behavior. Confusing these levers can leave a stale knowledge base untouched.
Incorrect Answers
B:
“Check chosen model, quota, region and data-processing terms” is a diagnostic step for the reported condition: Not every model and deployment option exists in all regions. This question instead requires an action that can choose RAG for current facts and prompts or tuning for response style. Fresh facts and style behavior are separate concerns. Distinguish stale source failures from tone or formatting failures. Retrieval supplies changed factual sources at answer time, whereas prompt constraints or tuning address presentation behavior. Confusing these levers can leave a stale knowledge base untouched.
C:
“Document gaps in available prebuilt capabilities and workflow needs” is a diagnostic step for the reported condition: A duplicate custom project adds governance and maintenance cost. This question instead requires an action that can choose RAG for current facts and prompts or tuning for response style. Fresh facts and style behavior are separate concerns. Distinguish stale source failures from tone or formatting failures. Retrieval supplies changed factual sources at answer time, whereas prompt constraints or tuning address presentation behavior. Confusing these levers can leave a stale knowledge base untouched.
D:
“Test sample queries by exact number and paraphrased policy terms” is a diagnostic step for the reported condition: Vector similarity alone is not enough to preserve identifiers. This question instead requires an action that can choose RAG for current facts and prompts or tuning for response style. Fresh facts and style behavior are separate concerns. Distinguish stale source failures from tone or formatting failures. Retrieval supplies changed factual sources at answer time, whereas prompt constraints or tuning address presentation behavior. Confusing these levers can leave a stale knowledge base untouched.
Question 349
At Sapphire Mobility, an international company wants a single AI adoption policy for all jurisdictions. Evidence: Availability, privacy law and employee consultation rules differ by region. The business goal is to adapt rollout controls to local obligations without lowering core safety. The project owner requires a measurable acceptance criterion before signoff. Which evidence would most convincingly demonstrate that the requirement has been met?
- No order is approved without delegated decision authority
- Employees accurately identify allowed inputs and escalation routes
- The workflow saves time without new quality defects
- Each regional plan follows its local requirements and business outcomes
Explanation
Correct Answer
D:
Availability, privacy law and employee consultation rules differ by region. Each regional plan follows its local requirements and business outcomes. A feature or data practice acceptable in one jurisdiction may require different notice, consultation or deployment controls elsewhere. Review local obligations while preserving consistent minimum safeguards.
Incorrect Answers
A:
“No order is approved without delegated decision authority” validates an outcome in a case where the evidence is: The agent's action scope could exceed its intended advisory role. This question instead requires an action that can adapt rollout controls to local obligations without lowering core safety. Availability, privacy law and employee consultation rules differ by region. Each regional plan follows its local requirements and business outcomes. A feature or data practice acceptable in one jurisdiction may require different notice, consultation or deployment controls elsewhere. Review local obligations while preserving consistent minimum safeguards.
B:
“Employees accurately identify allowed inputs and escalation routes” validates an outcome in a case where the evidence is: Conflicting messages describe data use and access controls. This question instead requires an action that can adapt rollout controls to local obligations without lowering core safety. Availability, privacy law and employee consultation rules differ by region. Each regional plan follows its local requirements and business outcomes. A feature or data practice acceptable in one jurisdiction may require different notice, consultation or deployment controls elsewhere. Review local obligations while preserving consistent minimum safeguards.
C:
“The workflow saves time without new quality defects” validates an outcome in a case where the evidence is: The assistant is not integrated into their ordinary work process. This question instead requires an action that can adapt rollout controls to local obligations without lowering core safety. Availability, privacy law and employee consultation rules differ by region. Each regional plan follows its local requirements and business outcomes. A feature or data practice acceptable in one jurisdiction may require different notice, consultation or deployment controls elsewhere. Review local obligations while preserving consistent minimum safeguards.
Question 350
At Meridian Imports, a marketing lead assumes public Copilot can read a protected SharePoint strategy. Evidence: The service has not been authorized to access the requested private file. The business goal is to distinguish public web answers from permissioned work grounding. The sponsor needs an evidence-based diagnosis rather than a guessed fix. Which initial investigation best identifies the controlling constraint before any solution is chosen?
- Identify source memo, requested sections and edit permissions
- List required read and write operations separately
- Determine which visual information is essential to the lesson
- Identify the app surface, license and actual connected sources
Explanation
Correct Answer
D:
The service has not been authorized to access the requested private file. Identify the app surface, license and actual connected sources.
Incorrect Answers
A:
“Identify source memo, requested sections and edit permissions” tests a proposed change for the condition: The task needs a working document with source-grounded language. This question instead requires an action that can distinguish public web answers from permissioned work grounding. The service has not been authorized to access the requested private file. Identify the app surface, license and actual connected sources.
B:
“List required read and write operations separately” is a diagnostic step for the reported condition: An unnecessary permission increases risk without improving the task. This question instead requires an action that can distinguish public web answers from permissioned work grounding. The service has not been authorized to access the requested private file. Identify the app surface, license and actual connected sources.
C:
“Determine which visual information is essential to the lesson” is a diagnostic step for the reported condition: Some images contain diagrams whose meaning needs human review. This question instead requires an action that can distinguish public web answers from permissioned work grounding. The service has not been authorized to access the requested private file. Identify the app surface, license and actual connected sources.