Microsoft AB-731 Practice Test 68
Question 670
At Elmwood Banking, an audit department queries a policy assistant that cites a superseded manual. Evidence: The knowledge index was last refreshed before the recent policy revision. The business goal is to maintain fresh retrieval sources rather than changing model weights. Only non-sensitive synthetic data and approved sandbox systems may be used. Which bounded trial with synthetic records best tests the proposed approach?
- Simulate a bad source date and verify sends are blocked
- Test retrieval of mock credentials by authorized and denied roles
- Run a mock benchmark with intentionally overlapping cases
- Replace a fictional policy version and rerun the same question
Explanation
Correct Answer
D:
The knowledge index was last refreshed before the recent policy revision. Replace a fictional policy version and rerun the same question. A policy that changes after indexing will not be reflected in retrieved passages until the source is refreshed. Track source versions, ingestion success and index freshness instead of retraining the model for factual updates.
Incorrect Answers
A:
“Simulate a bad source date and verify sends are blocked” tests a proposed change for the condition: One incorrect dispatch date could be repeated to thousands of customers. This question instead requires an action that can maintain fresh retrieval sources rather than changing model weights. The knowledge index was last refreshed before the recent policy revision. Replace a fictional policy version and rerun the same question. A policy that changes after indexing will not be reflected in retrieved passages until the source is refreshed. Track source versions, ingestion success and index freshness instead of retraining the model for factual updates.
B:
“Test retrieval of mock credentials by authorized and denied roles” tests a proposed change for the condition: The credentials enable unauthorized paid inference requests. This question instead requires an action that can maintain fresh retrieval sources rather than changing model weights. The knowledge index was last refreshed before the recent policy revision. Replace a fictional policy version and rerun the same question. A policy that changes after indexing will not be reflected in retrieved passages until the source is refreshed. Track source versions, ingestion success and index freshness instead of retraining the model for factual updates.
C:
“Run a mock benchmark with intentionally overlapping cases” tests a proposed change for the condition: The same customer records appear in tuning and testing data. This question instead requires an action that can maintain fresh retrieval sources rather than changing model weights. The knowledge index was last refreshed before the recent policy revision. Replace a fictional policy version and rerun the same question. A policy that changes after indexing will not be reflected in retrieved passages until the source is refreshed. Track source versions, ingestion success and index freshness instead of retraining the model for factual updates.
Question 671
At Crownridge Travel, a service desk wants an agent grounded in approved return-policy articles. Evidence: The questions recur and answers must reflect maintained policy knowledge. The business goal is to use Copilot Studio for a governed agent when a standard chat is insufficient. Only non-sensitive synthetic data and approved sandbox systems may be used. Which bounded trial with synthetic records best tests the proposed approach?
- Test fictitious return requests including unclear exceptions
- Trial mock support questions with known correct resolutions
- Test mock restricted files under permitted and denied identities
- Test mock routine clarification and disputed exception requests
Explanation
Correct Answer
A:
The questions recur and answers must reflect maintained policy knowledge. Test fictitious return requests including unclear exceptions.
Incorrect Answers
B:
“Trial mock support questions with known correct resolutions” tests a proposed change for the condition: Usage metrics do not reveal whether questions were resolved correctly. This question instead requires an action that can use Copilot Studio for a governed agent when a standard chat is insufficient. The questions recur and answers must reflect maintained policy knowledge. Test fictitious return requests including unclear exceptions.
C:
“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 use Copilot Studio for a governed agent when a standard chat is insufficient. The questions recur and answers must reflect maintained policy knowledge. Test fictitious return requests including unclear exceptions.
D:
“Test mock routine clarification and disputed exception requests” tests a proposed change for the condition: Regulated interpretation requires designated decision-makers. This question instead requires an action that can use Copilot Studio for a governed agent when a standard chat is insufficient. The questions recur and answers must reflect maintained policy knowledge. Test fictitious return requests including unclear exceptions.
Question 672
At Silverpine Group, 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. The project owner requires a measurable acceptance criterion before signoff. Which evidence would most convincingly demonstrate that the requirement has been met?
- The final explanation uses corrected aggregates and explicit assumptions
- The chosen mode minimizes total cost subject to latency needs
- The chosen option meets the defined internal metric gates
- Critical amounts and identifiers match the original images
Explanation
Correct Answer
B:
Dedicated capacity may incur baseline costs even at low traffic. The chosen mode minimizes total cost subject to latency needs.
Incorrect Answers
A:
“The final explanation uses corrected aggregates and explicit assumptions” validates an outcome in a case where the evidence is: Incorrect join keys could double-count expense records. 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. The chosen mode minimizes total cost subject to latency needs.
C:
“The chosen option meets the defined internal metric gates” validates an outcome in a case where the evidence is: The published scores do not measure the company's actual document tasks. 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. The chosen mode minimizes total cost subject to latency needs.
D:
“Critical amounts and identifiers match the original images” validates an outcome in a case where the evidence is: The original files are scans and require text extraction. 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. The chosen mode minimizes total cost subject to latency needs.
Question 673
At Coral Bay Systems, a department moves from a prototype directly to paid production. Evidence: There is no documented validation, deployment or monitoring plan. The business goal is to apply a lifecycle from data preparation through ongoing evaluation. The project owner requires a measurable acceptance criterion before signoff. Which evidence would most convincingly demonstrate that the requirement has been met?
- The model cannot ship without required signoffs and rollback plan
- No confidential source leaves the authorized processing boundary
- Performance and fairness meet gates across all intended groups
- Generated descriptions satisfy brand criteria while forecasting remains separately measured
Explanation
Correct Answer
A:
There is no documented validation, deployment or monitoring plan. The model cannot ship without required signoffs and rollback plan.
Incorrect Answers
B:
“No confidential source leaves the authorized processing boundary” validates an outcome in a case where the evidence is: The vendor retention and training terms were not evaluated. This question instead requires an action that can apply a lifecycle from data preparation through ongoing evaluation. There is no documented validation, deployment or monitoring plan. The model cannot ship without required signoffs and rollback plan.
C:
“Performance and fairness meet gates across all intended groups” validates an outcome in a case where the evidence is: The target deployment covers regions with very different job families. This question instead requires an action that can apply a lifecycle from data preparation through ongoing evaluation. There is no documented validation, deployment or monitoring plan. The model cannot ship without required signoffs and rollback plan.
D:
“Generated descriptions satisfy brand criteria while forecasting remains separately measured” validates an outcome in a case where the evidence is: The deliverable is new natural-language copy, not a numeric sales prediction. This question instead requires an action that can apply a lifecycle from data preparation through ongoing evaluation. There is no documented validation, deployment or monitoring plan. The model cannot ship without required signoffs and rollback plan.
Question 674
At Monarch Energy, a field manager needs short approved Copilot guidance while traveling. Evidence: The desktop-only instructions may not describe mobile access or capabilities. The business goal is to match the task to supported mobile features and company controls. The project owner requires a measurable acceptance criterion before signoff. Which evidence would most convincingly demonstrate that the requirement has been met?
- The result matches approved arithmetic and labels units correctly
- The mobile experience works without exposing enterprise secrets
- Critical tasks pass before moving production traffic
- The final explanation uses corrected aggregates and explicit assumptions
Explanation
Correct Answer
B:
The desktop-only instructions may not describe mobile access or capabilities. The mobile experience works without exposing enterprise secrets. A desktop feature may not behave identically in mobile clients or under corporate controls. Test the supported mobile workflow and account entitlement rather than assuming parity.
Incorrect Answers
A:
“The result matches approved arithmetic and labels units correctly” validates an outcome in a case where the evidence is: The original workbook contains numbers and formulas requiring verification. This question instead requires an action that can match the task to supported mobile features and company controls. The desktop-only instructions may not describe mobile access or capabilities. The mobile experience works without exposing enterprise secrets. A desktop feature may not behave identically in mobile clients or under corporate controls. Test the supported mobile workflow and account entitlement rather than assuming parity.
C:
“Critical tasks pass before moving production traffic” validates an outcome in a case where the evidence is: The successor may change behavior and cost. This question instead requires an action that can match the task to supported mobile features and company controls. The desktop-only instructions may not describe mobile access or capabilities. The mobile experience works without exposing enterprise secrets. A desktop feature may not behave identically in mobile clients or under corporate controls. Test the supported mobile workflow and account entitlement rather than assuming parity.
D:
“The final explanation uses corrected aggregates and explicit assumptions” validates an outcome in a case where the evidence is: Incorrect join keys could double-count expense records. This question instead requires an action that can match the task to supported mobile features and company controls. The desktop-only instructions may not describe mobile access or capabilities. The mobile experience works without exposing enterprise secrets. A desktop feature may not behave identically in mobile clients or under corporate controls. Test the supported mobile workflow and account entitlement rather than assuming parity.
Question 675
At Harborline Products, 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 project owner requires a measurable acceptance criterion before signoff. Which evidence would most convincingly demonstrate that the requirement has been met?
- No unauthorized action follows the retrieved directive
- Net benefit remains positive after all incremental costs and errors are included
- Extracted amounts and row associations match source documents
- Answers switch to the new approved clause with version citation
Explanation
Correct Answer
C:
Corrupt source extraction can make even a grounded answer wrong. Extracted amounts and row associations match source documents. 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:
“No unauthorized action follows the retrieved directive” validates an outcome in a case where the evidence is: The embedded directive came from lower-trust document content. 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. Extracted amounts and row associations match source documents. 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.
B:
“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 validate document extraction before retrieval and generation. Corrupt source extraction can make even a grounded answer wrong. Extracted amounts and row associations match source documents. 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:
“Answers switch to the new approved clause with version citation” validates an outcome in a case where the evidence is: The knowledge index was last refreshed before the recent policy revision. 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. Extracted amounts and row associations match source documents. 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 676
At Blue Ridge Technology, a marketing team requests AI-generated graphics resembling a competitor trademark. Evidence: Similar branded content may create legal or reputational risks. The business goal is to review IP and brand constraints in generative outputs. 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 copied-looking outputs and reuse-policy changes
- Monitor permission inheritance changes and cross-user leakage
- Monitor overfitting to exemplar content and stale template wording
- Monitor per-language retrieval quality and translation drift
Explanation
Correct Answer
A:
Similar branded content may create legal or reputational risks. Monitor copied-looking outputs and reuse-policy changes.
Incorrect Answers
B:
“Monitor permission inheritance changes and cross-user leakage” monitors this signal: Retrieval happens before checking the caller's permissions. This question instead requires an action that can review IP and brand constraints in generative outputs. Similar branded content may create legal or reputational risks. Monitor copied-looking outputs and reuse-policy changes.
C:
“Monitor overfitting to exemplar content and stale template wording” monitors this signal: Written instructions alone yield inconsistent section order. This question instead requires an action that can review IP and brand constraints in generative outputs. Similar branded content may create legal or reputational risks. Monitor copied-looking outputs and reuse-policy changes.
D:
“Monitor per-language retrieval quality and translation drift” monitors this signal: Vector similarity may degrade for underrepresented languages. This question instead requires an action that can review IP and brand constraints in generative outputs. Similar branded content may create legal or reputational risks. Monitor copied-looking outputs and reuse-policy changes.
Question 677
At Riverbend Publishing, a procurement team gets different classifications for the same standardized request. Evidence: Small phrasing changes cause inconsistent answers in a business-critical step. The business goal is to test consistency rather than trusting one convincing example. Only non-sensitive synthetic data and approved sandbox systems may be used. Which bounded trial with synthetic records best tests the proposed approach?
- Test hypothetical jobs from the omitted regions
- Run paraphrased versions of the same fictional request repeatedly
- Run repeat extraction on invoices with the same hidden reference
- Test an unmatched balance that must fail arithmetic checks
Explanation
Correct Answer
B:
Small phrasing changes cause inconsistent answers in a business-critical step. Run paraphrased versions of the same fictional request repeatedly.
Incorrect Answers
A:
“Test hypothetical jobs from the omitted regions” tests a proposed change for the condition: The target deployment covers regions with very different job families. This question instead requires an action that can test consistency rather than trusting one convincing example. Small phrasing changes cause inconsistent answers in a business-critical step. Run paraphrased versions of the same fictional request repeatedly.
C:
“Run repeat extraction on invoices with the same hidden reference” tests a proposed change for the condition: Creative variation adds no value and can introduce mismatches. This question instead requires an action that can test consistency rather than trusting one convincing example. Small phrasing changes cause inconsistent answers in a business-critical step. Run paraphrased versions of the same fictional request repeatedly.
D:
“Test an unmatched balance that must fail arithmetic checks” tests a proposed change for the condition: The reconciliation is deterministic arithmetic with known control rules. This question instead requires an action that can test consistency rather than trusting one convincing example. Small phrasing changes cause inconsistent answers in a business-critical step. Run paraphrased versions of the same fictional request repeatedly.
Question 678
At Evergreen Research, an AI outreach draft consistently favors one demographic group. Evidence: Training examples omit important customer contexts. The business goal is to evaluate fairness across representative groups before use. The project owner requires a measurable acceptance criterion before signoff. Which evidence would most convincingly demonstrate that the requirement has been met?
- Both task classes meet their own latency and quality gates
- Results report uncertainty and fail safely when data is insufficient
- Generated descriptions satisfy brand criteria while forecasting remains separately measured
- Similar eligible customers receive comparable recommendations
Explanation
Correct Answer
D:
Training examples omit important customer contexts. Similar eligible customers receive comparable recommendations. Representative test cases are needed to expose subgroup performance gaps and avoid unfair conclusions from narrow training examples.
Incorrect Answers
A:
“Both task classes meet their own latency and quality gates” validates an outcome in a case where the evidence is: The two tasks have different turnaround and explanation requirements. This question instead requires an action that can evaluate fairness across representative groups before use. Training examples omit important customer contexts. Similar eligible customers receive comparable recommendations. Representative test cases are needed to expose subgroup performance gaps and avoid unfair conclusions from narrow training examples.
B:
“Results report uncertainty and fail safely when data is insufficient” validates an outcome in a case where the evidence is: The sample cannot represent real volume and edge cases. This question instead requires an action that can evaluate fairness across representative groups before use. Training examples omit important customer contexts. Similar eligible customers receive comparable recommendations. Representative test cases are needed to expose subgroup performance gaps and avoid unfair conclusions from narrow training examples.
C:
“Generated descriptions satisfy brand criteria while forecasting remains separately measured” validates an outcome in a case where the evidence is: The deliverable is new natural-language copy, not a numeric sales prediction. This question instead requires an action that can evaluate fairness across representative groups before use. Training examples omit important customer contexts. Similar eligible customers receive comparable recommendations. Representative test cases are needed to expose subgroup performance gaps and avoid unfair conclusions from narrow training examples.
Question 679
At Pinecrest Services, a finance assistant answers a query and is also asked to transfer funds. Evidence: Retrieving an account statement does not authorize transactions. The business goal is to separate read-only grounding from high-impact actions. Only non-sensitive synthetic data and approved sandbox systems may be used. Which bounded trial with synthetic records best tests the proposed approach?
- Run a mock benchmark with intentionally overlapping cases
- Cluster fictional feedback including one emerging novel issue
- Plant a conflicting clause in a synthetic policy set
- Run a dummy transfer instruction against a read-only sandbox
Explanation
Correct Answer
D:
Retrieving an account statement does not authorize transactions. Run a dummy transfer instruction against a read-only sandbox.
Incorrect Answers
A:
“Run a mock benchmark with intentionally overlapping cases” tests a proposed change for the condition: The same customer records appear in tuning and testing data. This question instead requires an action that can separate read-only grounding from high-impact actions. Retrieving an account statement does not authorize transactions. Run a dummy transfer instruction against a read-only sandbox.
B:
“Cluster fictional feedback including one emerging novel issue” tests a proposed change for the condition: There are no predefined labels and the first goal is discovering patterns. This question instead requires an action that can separate read-only grounding from high-impact actions. Retrieving an account statement does not authorize transactions. Run a dummy transfer instruction against a read-only sandbox.
C:
“Plant a conflicting clause in a synthetic policy set” tests a proposed change for the condition: Retrieval produced a plausible paragraph but not supporting evidence. This question instead requires an action that can separate read-only grounding from high-impact actions. Retrieving an account statement does not authorize transactions. Run a dummy transfer instruction against a read-only sandbox.
Question 680
At Silverpine Group, a released agent sends misleading guidance to many employees. Evidence: There is no owner or process for AI-specific incidents. The business goal is to establish AI incident detection, containment and reporting. The project owner requires a measurable acceptance criterion before signoff. Which evidence would most convincingly demonstrate that the requirement has been met?
- Users can complete core tasks without inaccessible controls
- The team contains harm and documents corrective decisions
- Each wave meets adoption, quality and security gates before expansion
- Only proportionate approved data is processed
Explanation
Correct Answer
B:
There is no owner or process for AI-specific incidents. The team contains harm and documents corrective decisions.
Incorrect Answers
A:
“Users can complete core tasks without inaccessible controls” validates an outcome in a case where the evidence is: Keyboard navigation and alternative text were not tested. This question instead requires an action that can establish AI incident detection, containment and reporting. There is no owner or process for AI-specific incidents. The team contains harm and documents corrective decisions.
C:
“Each wave meets adoption, quality and security gates before expansion” validates an outcome in a case where the evidence is: Support capacity and user task quality have not been proven. This question instead requires an action that can establish AI incident detection, containment and reporting. There is no owner or process for AI-specific incidents. The team contains harm and documents corrective decisions.
D:
“Only proportionate approved data is processed” validates an outcome in a case where the evidence is: The proposed system analyzes communications and behavior at scale. This question instead requires an action that can establish AI incident detection, containment and reporting. There is no owner or process for AI-specific incidents. The team contains harm and documents corrective decisions.