Microsoft AB-731 Practice Test 91
Question 900
At Marigold Education, a mortgage assistant answers complex legal questions without escalation. Evidence: Its intended scope is simple FAQ and required documents. The business goal is to define confidence and impact triggers for human handoff. The proposed solution must respect documented authority and organizational controls. Which approach most directly meets the stated business objective within the approved authority?
- Prototype an anomaly or prediction model with maintenance specialists
- Allow grounded answers but require explicit privileged payment authorization
- Publish a documented ranking and choose a feasible limited first trial
- Provide safe fallback answers and an authorized specialist route
Explanation
Correct Answer
D:
Its intended scope is simple FAQ and required documents. Provide safe fallback answers and an authorized specialist route. A FAQ assistant should route missing, ambiguous or high-impact cases to staff instead of inventing policy answers. Define escalation triggers and confirm that the handoff preserves essential context.
Incorrect Answers
A:
“Prototype an anomaly or prediction model with maintenance specialists” is a design or implementation response to the condition: Threshold rules miss complex multi-variable patterns. This question instead requires an action that can define confidence and impact triggers for human handoff. Its intended scope is simple FAQ and required documents. Provide safe fallback answers and an authorized specialist route. A FAQ assistant should route missing, ambiguous or high-impact cases to staff instead of inventing policy answers. Define escalation triggers and confirm that the handoff preserves essential context.
B:
“Allow grounded answers but require explicit privileged payment authorization” is a design or implementation response to the condition: Retrieving an account statement does not authorize transactions. This question instead requires an action that can define confidence and impact triggers for human handoff. Its intended scope is simple FAQ and required documents. Provide safe fallback answers and an authorized specialist route. A FAQ assistant should route missing, ambiguous or high-impact cases to staff instead of inventing policy answers. Define escalation triggers and confirm that the handoff preserves essential context.
C:
“Publish a documented ranking and choose a feasible limited first trial” tests a proposed change for the condition: Ideas differ in readiness, benefits and regulatory exposure. This question instead requires an action that can define confidence and impact triggers for human handoff. Its intended scope is simple FAQ and required documents. Provide safe fallback answers and an authorized specialist route. A FAQ assistant should route missing, ambiguous or high-impact cases to staff instead of inventing policy answers. Define escalation triggers and confirm that the handoff preserves essential context.
Question 901
At Meridian Imports, an assistant reads OCR scans where table columns merge incorrectly. Evidence: Corrupt source extraction can make even a grounded answer wrong. The business goal is to validate document extraction before retrieval and generation. The operating team needs an ongoing check after the initial rollout. Which post-rollout signal is most important for detecting regression or governance drift?
- Track precision for paraphrased and exact-match queries
- Monitor OCR failures and source format changes
- Monitor unusual source instructions and attempted tool misuse
- Monitor opaque rankings and unexplained score swings
Explanation
Correct Answer
B:
Corrupt source extraction can make even a grounded answer wrong. Monitor OCR failures and source format changes. If extraction drops table columns or corrupts words, even a faithfully cited answer can be false. Validate extraction fidelity and parse failures before indexing the content.
Incorrect Answers
A:
“Track precision for paraphrased and exact-match queries” monitors this signal: Matching literal keywords alone fails for synonymous wording. This question instead requires an action that can validate document extraction before retrieval and generation. Corrupt source extraction can make even a grounded answer wrong. Monitor OCR failures and source format changes. If extraction drops table columns or corrupts words, even a faithfully cited answer can be false. Validate extraction fidelity and parse failures before indexing the content.
C:
“Monitor unusual source instructions and attempted tool misuse” monitors this signal: The page is untrusted external content rather than a user-approved command. This question instead requires an action that can validate document extraction before retrieval and generation. Corrupt source extraction can make even a grounded answer wrong. Monitor OCR failures and source format changes. If extraction drops table columns or corrupts words, even a faithfully cited answer can be false. Validate extraction fidelity and parse failures before indexing the content.
D:
“Monitor opaque rankings and unexplained score swings” monitors this signal: The decision-maker must justify awards to an audit committee. This question instead requires an action that can validate document extraction before retrieval and generation. Corrupt source extraction can make even a grounded answer wrong. Monitor OCR failures and source format changes. If extraction drops table columns or corrupts words, even a faithfully cited answer can be false. Validate extraction fidelity and parse failures before indexing the content.
Question 902
At Prairie Health, a claims classification model performs badly on unusual claim types. Evidence: Training labels are inconsistent and rare classes have few examples. The business goal is to improve labeling quality and representation before model tuning. The project owner requires a measurable acceptance criterion before signoff. Which evidence would most convincingly demonstrate that the requirement has been met?
- Only approved changes are deployed and reproducible
- Support staff can explain the failure without collecting sensitive prompts
- Net benefit remains positive after all incremental costs and errors are included
- Inter-annotator agreement and minority-class performance improve
Explanation
Correct Answer
D:
Training labels are inconsistent and rare classes have few examples. Inter-annotator agreement and minority-class performance improve. Inconsistent labels teach the model contradictory targets, while missing rare classes distort accuracy. Resolve annotation rules and sample coverage before changing architecture or tuning hyperparameters.
Incorrect Answers
A:
“Only approved changes are deployed and reproducible” validates an outcome in a case where the evidence is: Prompt templates have no version owner or change record. This question instead requires an action that can improve labeling quality and representation before model tuning. Training labels are inconsistent and rare classes have few examples. Inter-annotator agreement and minority-class performance improve. Inconsistent labels teach the model contradictory targets, while missing rare classes distort accuracy. Resolve annotation rules and sample coverage before changing architecture or tuning hyperparameters.
B:
“Support staff can explain the failure without collecting sensitive prompts” validates an outcome in a case where the evidence is: There is no trace from answer to model version and retrieved sources. This question instead requires an action that can improve labeling quality and representation before model tuning. Training labels are inconsistent and rare classes have few examples. Inter-annotator agreement and minority-class performance improve. Inconsistent labels teach the model contradictory targets, while missing rare classes distort accuracy. Resolve annotation rules and sample coverage before changing architecture or tuning hyperparameters.
C:
“Net benefit remains positive after all incremental costs and errors are included” validates an outcome in a case where the evidence is: No baseline handling time, error rate or deployment cost is recorded. This question instead requires an action that can improve labeling quality and representation before model tuning. Training labels are inconsistent and rare classes have few examples. Inter-annotator agreement and minority-class performance improve. Inconsistent labels teach the model contradictory targets, while missing rare classes distort accuracy. Resolve annotation rules and sample coverage before changing architecture or tuning hyperparameters.
Question 903
At Summit Manufacturing, a department refuses an AI assistant after one fabricated answer. Evidence: Users have no transparent examples of error correction or escalation. The business goal is to rebuild trust through evidence, safeguards and iterative feedback. The project owner requires a measurable acceptance criterion before signoff. Which evidence would most convincingly demonstrate that the requirement has been met?
- The selected supplier meets mandatory governance controls
- Users can identify limits and report errors safely
- Participants demonstrate accuracy and proper escalation
- Observed net value remains positive under sensitivity analysis
Explanation
Correct Answer
B:
Users have no transparent examples of error correction or escalation. Users can identify limits and report errors safely.
Incorrect Answers
A:
“The selected supplier meets mandatory governance controls” validates an outcome in a case where the evidence is: Contract data and prompts may be handled under unclear commitments. This question instead requires an action that can rebuild trust through evidence, safeguards and iterative feedback. Users have no transparent examples of error correction or escalation. Users can identify limits and report errors safely.
C:
“Participants demonstrate accuracy and proper escalation” validates an outcome in a case where the evidence is: Video completion does not show safe real-world ability. This question instead requires an action that can rebuild trust through evidence, safeguards and iterative feedback. Users have no transparent examples of error correction or escalation. Users can identify limits and report errors safely.
D:
“Observed net value remains positive under sensitivity analysis” validates an outcome in a case where the evidence is: Real costs include licenses, training, evaluation and human review. This question instead requires an action that can rebuild trust through evidence, safeguards and iterative feedback. Users have no transparent examples of error correction or escalation. Users can identify limits and report errors safely.
Question 904
At Cascade Utilities, an organization recruits champions solely by enthusiasm. Evidence: Some champions share risky shortcuts and outdated feature claims. The business goal is to train champions on safe usage and referral boundaries. The operating team needs an ongoing check after the initial rollout. Which post-rollout signal is most important for detecting regression or governance drift?
- Monitor rework, usage and changes in cost assumptions
- Audit sensitive-data leakage and collection creep
- Monitor champion misinformation and coaching consistency
- Monitor skill decay and recurrent misuse after training
Explanation
Correct Answer
C:
Some champions share risky shortcuts and outdated feature claims. Monitor champion misinformation and coaching consistency.
Incorrect Answers
A:
“Monitor rework, usage and changes in cost assumptions” monitors this signal: Real costs include licenses, training, evaluation and human review. This question instead requires an action that can train champions on safe usage and referral boundaries. Some champions share risky shortcuts and outdated feature claims. Monitor champion misinformation and coaching consistency.
B:
“Audit sensitive-data leakage and collection creep” monitors this signal: The project only needs agreed project decisions. This question instead requires an action that can train champions on safe usage and referral boundaries. Some champions share risky shortcuts and outdated feature claims. Monitor champion misinformation and coaching consistency.
D:
“Monitor skill decay and recurrent misuse after training” monitors this signal: Video completion does not show safe real-world ability. This question instead requires an action that can train champions on safe usage and referral boundaries. Some champions share risky shortcuts and outdated feature claims. Monitor champion misinformation and coaching consistency.
Question 905
At Northlake Services, a policy director needs a comparative industry briefing from many sources. Evidence: The question requires multi-step evidence gathering rather than quick lookup. The business goal is to select Researcher for deeper cited multi-source synthesis where available. The proposed solution must respect documented authority and organizational controls. Which approach most directly meets the stated business objective within the approved authority?
- Choose Analyst with reviewed files and independently validate results
- Generate draft alt text with manual accuracy and inclusiveness review
- Pilot renewal tiers based on observed task-level benefit and governance
- Use the supported Researcher workflow with explicit review criteria
Explanation
Correct Answer
D:
The question requires multi-step evidence gathering rather than quick lookup. Use the supported Researcher workflow with explicit review criteria.
Incorrect Answers
A:
“Choose Analyst with reviewed files and independently validate results” tests a proposed change for the condition: The task requires calculations and statistical checks beyond simple prose. This question instead requires an action that can select Researcher for deeper cited multi-source synthesis where available. The question requires multi-step evidence gathering rather than quick lookup. Use the supported Researcher workflow with explicit review criteria.
B:
“Generate draft alt text with manual accuracy and inclusiveness review” is a design or implementation response to the condition: Some images contain diagrams whose meaning needs human review. This question instead requires an action that can select Researcher for deeper cited multi-source synthesis where available. The question requires multi-step evidence gathering rather than quick lookup. Use the supported Researcher workflow with explicit review criteria.
C:
“Pilot renewal tiers based on observed task-level benefit and governance” is a design or implementation response to the condition: License usage is not equivalent to measurable task improvement. This question instead requires an action that can select Researcher for deeper cited multi-source synthesis where available. The question requires multi-step evidence gathering rather than quick lookup. Use the supported Researcher workflow with explicit review criteria.
Question 906
At Pinecrest Services, a sales officer presents generated future revenue to two decimal places. Evidence: The model lacks an approved statistical forecast and uncertainty bounds. The business goal is to avoid presenting generated guesses as quantified forecast evidence. The operating team needs an ongoing check after the initial rollout. Which post-rollout signal is most important for detecting regression or governance drift?
- Monitor response bias and sudden changes in survey participation
- Watch missing critical evidence and overly long prompts
- Monitor cluster stability, low-frequency issues and inappropriate labels
- Watch fabricated precision and overly confident executive narratives
Explanation
Correct Answer
D:
The model lacks an approved statistical forecast and uncertainty bounds. Watch fabricated precision and overly confident executive narratives.
Incorrect Answers
A:
“Monitor response bias and sudden changes in survey participation” monitors this signal: Feedback is self-selected and omits escalated failures. This question instead requires an action that can avoid presenting generated guesses as quantified forecast evidence. The model lacks an approved statistical forecast and uncertainty bounds. Watch fabricated precision and overly confident executive narratives.
B:
“Watch missing critical evidence and overly long prompts” monitors this signal: Retrieving many passages improves coverage while increasing noise. This question instead requires an action that can avoid presenting generated guesses as quantified forecast evidence. The model lacks an approved statistical forecast and uncertainty bounds. Watch fabricated precision and overly confident executive narratives.
C:
“Monitor cluster stability, low-frequency issues and inappropriate labels” monitors this signal: There are no predefined labels and the first goal is discovering patterns. This question instead requires an action that can avoid presenting generated guesses as quantified forecast evidence. The model lacks an approved statistical forecast and uncertainty bounds. Watch fabricated precision and overly confident executive narratives.
Question 907
At Riverbend Publishing, a hiring assistant pilot uses historical examples from one office only. Evidence: The target deployment covers regions with very different job families. The business goal is to check evaluation representativeness before organization-wide claims. 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 exact extraction from creative brainstorming needs
- Compare sample distributions to target users and workflows
- Identify regulated data locations and service-region terms
- Identify unnecessary health fields in each request payload
Explanation
Correct Answer
B:
The target deployment covers regions with very different job families. Compare sample distributions to target users and workflows. Accuracy measured on one region or job family may mask poor performance elsewhere. Sample intended deployment cohorts and report subgroup failures before claiming enterprise representativeness.
Incorrect Answers
A:
“Distinguish exact extraction from creative brainstorming needs” is a diagnostic step for the reported condition: Creative variation adds no value and can introduce mismatches. This question instead requires an action that can check evaluation representativeness before organization-wide claims. The target deployment covers regions with very different job families. Compare sample distributions to target users and workflows. Accuracy measured on one region or job family may mask poor performance elsewhere. Sample intended deployment cohorts and report subgroup failures before claiming enterprise representativeness.
C:
“Identify regulated data locations and service-region terms” is a diagnostic step for the reported condition: Data location and transfer obligations may constrain service selection. This question instead requires an action that can check evaluation representativeness before organization-wide claims. The target deployment covers regions with very different job families. Compare sample distributions to target users and workflows. Accuracy measured on one region or job family may mask poor performance elsewhere. Sample intended deployment cohorts and report subgroup failures before claiming enterprise representativeness.
D:
“Identify unnecessary health fields in each request payload” is a diagnostic step for the reported condition: Only scheduling constraints and contact details are needed. This question instead requires an action that can check evaluation representativeness before organization-wide claims. The target deployment covers regions with very different job families. Compare sample distributions to target users and workflows. Accuracy measured on one region or job family may mask poor performance elsewhere. Sample intended deployment cohorts and report subgroup failures before claiming enterprise representativeness.
Question 908
At Sapphire Mobility, a customer service sponsor wants a model for text classification. Evidence: Candidate models differ in modality, region, cost and accuracy. The business goal is to select models by measurable workload requirements. Only non-sensitive synthetic data and approved sandbox systems may be used. Which bounded trial with synthetic records best tests the proposed approach?
- Run fictional queries with deliberately distractor paragraphs
- Query fictional clauses by identifier and paraphrased concept
- Test a fictional slide deck containing unreadable chart labels
- Evaluate candidate models on identical fictional customer tickets
Explanation
Correct Answer
D:
Candidate models differ in modality, region, cost and accuracy. Evaluate candidate models on identical fictional customer tickets.
Incorrect Answers
A:
“Run fictional queries with deliberately distractor paragraphs” tests a proposed change for the condition: First-stage keyword search has poor top-result precision. This question instead requires an action that can select models by measurable workload requirements. Candidate models differ in modality, region, cost and accuracy. Evaluate candidate models on identical fictional customer tickets.
B:
“Query fictional clauses by identifier and paraphrased concept” tests a proposed change for the condition: Vector-only matching risks missing literal identifiers. This question instead requires an action that can select models by measurable workload requirements. Candidate models differ in modality, region, cost and accuracy. Evaluate candidate models on identical fictional customer tickets.
C:
“Test a fictional slide deck containing unreadable chart labels” tests a proposed change for the condition: Generated visuals may omit alt text or use low-contrast design. This question instead requires an action that can select models by measurable workload requirements. Candidate models differ in modality, region, cost and accuracy. Evaluate candidate models on identical fictional customer tickets.
Question 909
At Summit Manufacturing, a finance analyst compares serverless and provisioned inference. Evidence: Dedicated capacity may incur baseline costs even at low traffic. The business goal is to include fixed capacity and variable usage in deployment comparison. Only non-sensitive synthetic data and approved sandbox systems may be used. Which bounded trial with synthetic records best tests the proposed approach?
- Model simulated daily peak and overnight idle usage
- Search fake regulations using a code and a synonym
- Trial a fictional workload using a limited nonproduction project
- Compare mock source versions with a stale generated summary
Explanation
Correct Answer
A:
Dedicated capacity may incur baseline costs even at low traffic. Model simulated daily peak and overnight idle usage.
Incorrect Answers
B:
“Search fake regulations using a code and a synonym” tests a proposed change for the condition: Vector similarity alone is not enough to preserve identifiers. This question instead requires an action that can include fixed capacity and variable usage in deployment comparison. Dedicated capacity may incur baseline costs even at low traffic. Model simulated daily peak and overnight idle usage.
C:
“Trial a fictional workload using a limited nonproduction project” tests a proposed change for the condition: The team must compare models, tools and safety controls as one lifecycle. This question instead requires an action that can include fixed capacity and variable usage in deployment comparison. Dedicated capacity may incur baseline costs even at low traffic. Model simulated daily peak and overnight idle usage.
D:
“Compare mock source versions with a stale generated summary” tests a proposed change for the condition: Some underlying work records were revised after the first draft. This question instead requires an action that can include fixed capacity and variable usage in deployment comparison. Dedicated capacity may incur baseline costs even at low traffic. Model simulated daily peak and overnight idle usage.
Question 910
At BrightPath Media, a manufacturer wants to predict rare equipment failures from sensor histories. Evidence: Threshold rules miss complex multi-variable patterns. The business goal is to use predictive ML when labeled trends add real signal. The proposed solution must respect documented authority and organizational controls. Which approach most directly meets the stated business objective within the approved authority?
- Draft from approved clauses and route final language to legal review
- Use an approved redacted extract or seek authorized access
- Prototype an anomaly or prediction model with maintenance specialists
- Return traceable references and abstain where evidence is absent
Explanation
Correct Answer
C:
Threshold rules miss complex multi-variable patterns. Prototype an anomaly or prediction model with maintenance specialists. When multiple signals interact, a fixed threshold may lose useful predictive information. Evaluate a labeled model against the current rule baseline using task-relevant false-positive and false-negative costs.
Incorrect Answers
A:
“Draft from approved clauses and route final language to legal review” is a design or implementation response to the condition: The proposed wording could create commitments that the legal owner has not approved. This question instead requires an action that can use predictive ML when labeled trends add real signal. Threshold rules miss complex multi-variable patterns. Prototype an anomaly or prediction model with maintenance specialists. When multiple signals interact, a fixed threshold may lose useful predictive information. Evaluate a labeled model against the current rule baseline using task-relevant false-positive and false-negative costs.
B:
“Use an approved redacted extract or seek authorized access” is a design or implementation response to the condition: The user can access the approved summary but not full agreement text. This question instead requires an action that can use predictive ML when labeled trends add real signal. Threshold rules miss complex multi-variable patterns. Prototype an anomaly or prediction model with maintenance specialists. When multiple signals interact, a fixed threshold may lose useful predictive information. Evaluate a labeled model against the current rule baseline using task-relevant false-positive and false-negative costs.
D:
“Return traceable references and abstain where evidence is absent” is a design or implementation response to the condition: Retrieval produced a plausible paragraph but not supporting evidence. This question instead requires an action that can use predictive ML when labeled trends add real signal. Threshold rules miss complex multi-variable patterns. Prototype an anomaly or prediction model with maintenance specialists. When multiple signals interact, a fixed threshold may lose useful predictive information. Evaluate a labeled model against the current rule baseline using task-relevant false-positive and false-negative costs.