Microsoft AB-731 Practice Test 18
Question 170
At Cedar Analytics, a pilot reports rising user complaints but lacks request identifiers. Evidence: There is no trace from answer to model version and retrieved sources. The business goal is to make AI outcomes diagnosable without exposing secrets. The proposed solution must respect documented authority and organizational controls. Which approach most directly meets the stated business objective within the approved authority?
- Retrieve approved current policy passages at query time with citations
- Apply an appropriate re-ranking stage with measured latency impact
- Remove unrelated sensitive fields and apply purpose-specific access
- Log sanitized traces, model version, retrieval IDs and quality outcomes
Explanation
Correct Answer
D:
There is no trace from answer to model version and retrieved sources. Log sanitized traces, model version, retrieval IDs and quality outcomes.
Incorrect Answers
A:
“Retrieve approved current policy passages at query time with citations” is a design or implementation response to the condition: A static model snapshot cannot reliably know the latest approved document. This question instead requires an action that can make AI outcomes diagnosable without exposing secrets. There is no trace from answer to model version and retrieved sources. Log sanitized traces, model version, retrieval IDs and quality outcomes.
B:
“Apply an appropriate re-ranking stage with measured latency impact” is a design or implementation response to the condition: Initial keyword matching is broad and top result quality is poor. This question instead requires an action that can make AI outcomes diagnosable without exposing secrets. There is no trace from answer to model version and retrieved sources. Log sanitized traces, model version, retrieval IDs and quality outcomes.
C:
“Remove unrelated sensitive fields and apply purpose-specific access” is a design or implementation response to the condition: Only scheduling constraints and contact details are needed. This question instead requires an action that can make AI outcomes diagnosable without exposing secrets. There is no trace from answer to model version and retrieved sources. Log sanitized traces, model version, retrieval IDs and quality outcomes.
Question 171
At Lakeshore Commerce, a legal search returns relevant documents below older irrelevant results. Evidence: Initial keyword matching is broad and top result quality is poor. The business goal is to consider semantic ranking after initial retrieval. 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 active template version and approval history
- Review ranking failures with labeled relevance judgments
- Inspect language distribution of indexes and failed queries
- List missing decision criteria, sources and intended recipients
Explanation
Correct Answer
B:
Initial keyword matching is broad and top result quality is poor. Review ranking failures with labeled relevance judgments.
Incorrect Answers
A:
“Identify active template version and approval history” is a diagnostic step for the reported condition: Prompt templates have no version owner or change record. This question instead requires an action that can consider semantic ranking after initial retrieval. Initial keyword matching is broad and top result quality is poor. Review ranking failures with labeled relevance judgments.
C:
“Inspect language distribution of indexes and failed queries” is a diagnostic step for the reported condition: Vector similarity may degrade for underrepresented languages. This question instead requires an action that can consider semantic ranking after initial retrieval. Initial keyword matching is broad and top result quality is poor. Review ranking failures with labeled relevance judgments.
D:
“List missing decision criteria, sources and intended recipients” is a diagnostic step for the reported condition: The request gives no objective, audience, boundaries or evidence. This question instead requires an action that can consider semantic ranking after initial retrieval. Initial keyword matching is broad and top result quality is poor. Review ranking failures with labeled relevance judgments.
Question 172
At Tamarack Logistics, an organization wants descriptions of training images for accessibility. Evidence: Some images contain diagrams whose meaning needs human review. The business goal is to use vision-assisted descriptions as reviewed accessibility drafts. The project owner requires a measurable acceptance criterion before signoff. Which evidence would most convincingly demonstrate that the requirement has been met?
- Both exact and semantic requirements retrieve correct approved text
- Routine answers cite current policy and exceptions escalate
- Alternative text conveys necessary chart meaning without inventions
- Critical amounts and identifiers match the original images
Explanation
Correct Answer
C:
Some images contain diagrams whose meaning needs human review. Alternative text conveys necessary chart meaning without inventions. Vision-generated alternative text is a draft interpretation, especially for diagrams with implied meaning. Human review should verify essential relationships, labels and omissions before publication.
Incorrect Answers
A:
“Both exact and semantic requirements retrieve correct approved text” validates an outcome in a case where the evidence is: Vector-only matching risks missing literal identifiers. This question instead requires an action that can use vision-assisted descriptions as reviewed accessibility drafts. Some images contain diagrams whose meaning needs human review. Alternative text conveys necessary chart meaning without inventions. Vision-generated alternative text is a draft interpretation, especially for diagrams with implied meaning. Human review should verify essential relationships, labels and omissions before publication.
B:
“Routine answers cite current policy and exceptions escalate” validates an outcome in a case where the evidence is: The questions recur and answers must reflect maintained policy knowledge. This question instead requires an action that can use vision-assisted descriptions as reviewed accessibility drafts. Some images contain diagrams whose meaning needs human review. Alternative text conveys necessary chart meaning without inventions. Vision-generated alternative text is a draft interpretation, especially for diagrams with implied meaning. Human review should verify essential relationships, labels and omissions before publication.
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 use vision-assisted descriptions as reviewed accessibility drafts. Some images contain diagrams whose meaning needs human review. Alternative text conveys necessary chart meaning without inventions. Vision-generated alternative text is a draft interpretation, especially for diagrams with implied meaning. Human review should verify essential relationships, labels and omissions before publication.
Question 173
At BrightPath Media, an AI steering group approves a chatbot without recording known limitations. Evidence: Users and auditors cannot see intended scope and failure conditions. The business goal is to document the system's purpose, data, evaluations and limitations. Only non-sensitive synthetic data and approved sandbox systems may be used. Which bounded trial with synthetic records best tests the proposed approach?
- Create a fictional system card for an FAQ assistant
- Submit a fictional erroneous benefit answer to the appeal path
- Route two fictional low- and high-risk proposals through governance
- Run a fictional workshop demonstrating assisted but reviewed work
Explanation
Correct Answer
A:
Users and auditors cannot see intended scope and failure conditions. Create a fictional system card for an FAQ assistant. Documenting intended users, data provenance, metrics and failure boundaries makes system claims auditable. Users must know when outputs require verification or escalation, not merely the feature list.
Incorrect Answers
B:
“Submit a fictional erroneous benefit answer to the appeal path” tests a proposed change for the condition: Users may be unable to correct harmful mistakes. This question instead requires an action that can document the system's purpose, data, evaluations and limitations. Users and auditors cannot see intended scope and failure conditions. Create a fictional system card for an FAQ assistant. Documenting intended users, data provenance, metrics and failure boundaries makes system claims auditable. Users must know when outputs require verification or escalation, not merely the feature list.
C:
“Route two fictional low- and high-risk proposals through governance” tests a proposed change for the condition: A low-risk note formatter and high-stakes hiring scorer carry different risks. This question instead requires an action that can document the system's purpose, data, evaluations and limitations. Users and auditors cannot see intended scope and failure conditions. Create a fictional system card for an FAQ assistant. Documenting intended users, data provenance, metrics and failure boundaries makes system claims auditable. Users must know when outputs require verification or escalation, not merely the feature list.
D:
“Run a fictional workshop demonstrating assisted but reviewed work” tests a proposed change for the condition: Leadership communication focuses only on headcount cost savings. This question instead requires an action that can document the system's purpose, data, evaluations and limitations. Users and auditors cannot see intended scope and failure conditions. Create a fictional system card for an FAQ assistant. Documenting intended users, data provenance, metrics and failure boundaries makes system claims auditable. Users must know when outputs require verification or escalation, not merely the feature list.
Question 174
At Evergreen Research, a company reports projected cost savings before any pilot has run. Evidence: Real costs include licenses, training, evaluation and human review. The business goal is to measure realized ROI against a credible baseline. The proposed solution must respect documented authority and organizational controls. Which approach most directly meets the stated business objective within the approved authority?
- Apply a cross-functional scoring rubric and stage-gate pilots
- Design least privilege, isolation, authentication and audit controls
- Build transparent ROI reporting with benefit attribution
- Compare pay-as-you-go with available commitments over realistic scenarios
Explanation
Correct Answer
C:
Real costs include licenses, training, evaluation and human review. Build transparent ROI reporting with benefit attribution.
Incorrect Answers
A:
“Apply a cross-functional scoring rubric and stage-gate pilots” is a design or implementation response to the condition: No process ranks ideas by value, readiness and risk. This question instead requires an action that can measure realized ROI against a credible baseline. Real costs include licenses, training, evaluation and human review. Build transparent ROI reporting with benefit attribution.
B:
“Design least privilege, isolation, authentication and audit controls” is a design or implementation response to the condition: The agent can call tools and access internal data repositories. This question instead requires an action that can measure realized ROI against a credible baseline. Real costs include licenses, training, evaluation and human review. Build transparent ROI reporting with benefit attribution.
D:
“Compare pay-as-you-go with available commitments over realistic scenarios” is a design or implementation response to the condition: Demand is uncertain and production volume may be seasonal. This question instead requires an action that can measure realized ROI against a credible baseline. Real costs include licenses, training, evaluation and human review. Build transparent ROI reporting with benefit attribution.
Question 175
At Sapphire Mobility, a user asks for today's public news and internal project milestones together. Evidence: Public web claims and workplace records have different provenance. The business goal is to explicitly distinguish evidence sources in mixed-context workflows. 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 false assertions about connected data access
- Monitor repeated single-source bias and unsupported certainty
- Monitor source labels and accidental confidential disclosure
- Monitor privilege creep and unreviewed role changes
Explanation
Correct Answer
C:
Public web claims and workplace records have different provenance. Monitor source labels and accidental confidential disclosure.
Incorrect Answers
A:
“Monitor false assertions about connected data access” monitors this signal: The service has not been authorized to access the requested private file. This question instead requires an action that can explicitly distinguish evidence sources in mixed-context workflows. Public web claims and workplace records have different provenance. Monitor source labels and accidental confidential disclosure.
B:
“Monitor repeated single-source bias and unsupported certainty” monitors this signal: The agent draws from a small biased sample of industry articles. This question instead requires an action that can explicitly distinguish evidence sources in mixed-context workflows. Public web claims and workplace records have different provenance. Monitor source labels and accidental confidential disclosure.
D:
“Monitor privilege creep and unreviewed role changes” monitors this signal: Excess access can lead to unreviewed deployments and data exposure. This question instead requires an action that can explicitly distinguish evidence sources in mixed-context workflows. Public web claims and workplace records have different provenance. Monitor source labels and accidental confidential disclosure.
Question 176
At Sapphire Mobility, an enterprise sponsor sees surprise charges for a hosted language model. Evidence: Input and output tokens and deployment type all affect billing. The business goal is to build a model-specific cost forecast before scaling. The project owner requires a measurable acceptance criterion before signoff. Which evidence would most convincingly demonstrate that the requirement has been met?
- The assistant refuses unsupported exceptions and cites real clauses
- Observed unit cost matches forecast within tolerance
- Both exact and semantic requirements retrieve correct approved text
- Final deck remains understandable using keyboard and screen reader
Explanation
Correct Answer
B:
Input and output tokens and deployment type all affect billing. Observed unit cost matches forecast within tolerance. Inference cost depends on both input and generated tokens as well as the deployment rate and throughput pattern. A scale estimate must include real workload distributions, not just the headline model price.
Incorrect Answers
A:
“The assistant refuses unsupported exceptions and cites real clauses” validates an outcome in a case where the evidence is: Citations exist but conclusions are not entailed by the passages. This question instead requires an action that can build a model-specific cost forecast before scaling. Input and output tokens and deployment type all affect billing. Observed unit cost matches forecast within tolerance. Inference cost depends on both input and generated tokens as well as the deployment rate and throughput pattern. A scale estimate must include real workload distributions, not just the headline model price.
C:
“Both exact and semantic requirements retrieve correct approved text” validates an outcome in a case where the evidence is: Vector-only matching risks missing literal identifiers. This question instead requires an action that can build a model-specific cost forecast before scaling. Input and output tokens and deployment type all affect billing. Observed unit cost matches forecast within tolerance. Inference cost depends on both input and generated tokens as well as the deployment rate and throughput pattern. A scale estimate must include real workload distributions, not just the headline model price.
D:
“Final deck remains understandable using keyboard and screen reader” validates an outcome in a case where the evidence is: Generated visuals may omit alt text or use low-contrast design. This question instead requires an action that can build a model-specific cost forecast before scaling. Input and output tokens and deployment type all affect billing. Observed unit cost matches forecast within tolerance. Inference cost depends on both input and generated tokens as well as the deployment rate and throughput pattern. A scale estimate must include real workload distributions, not just the headline model price.
Question 177
At Evergreen Research, an agent is halfway through preparing a multi-step report when a reviewer objects. Evidence: Restarting could duplicate work and bypass reviewed steps. The business goal is to use appropriate task state and human feedback rather than blind restart. Only non-sensitive synthetic data and approved sandbox systems may be used. Which bounded trial with synthetic records best tests the proposed approach?
- Test a mock brief combining outside articles and spreadsheet tables
- Edit a fictional plan with two sandbox colleagues
- Query fictional clauses by identifier and paraphrased concept
- Simulate a report plan with one deliberate input correction
Explanation
Correct Answer
D:
Restarting could duplicate work and bypass reviewed steps. Simulate a report plan with one deliberate input correction.
Incorrect Answers
A:
“Test a mock brief combining outside articles and spreadsheet tables” tests a proposed change for the condition: External research and numerical modeling are distinct activities. This question instead requires an action that can use appropriate task state and human feedback rather than blind restart. Restarting could duplicate work and bypass reviewed steps. Simulate a report plan with one deliberate input correction.
B:
“Edit a fictional plan with two sandbox colleagues” tests a proposed change for the condition: Parallel copies cause conflicting versions of one plan. This question instead requires an action that can use appropriate task state and human feedback rather than blind restart. Restarting could duplicate work and bypass reviewed steps. Simulate a report plan with one deliberate input correction.
C:
“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 use appropriate task state and human feedback rather than blind restart. Restarting could duplicate work and bypass reviewed steps. Simulate a report plan with one deliberate input correction.
Question 178
At Cedar Analytics, a service desk updates one instruction and later cannot explain changed answers. Evidence: Prompt templates have no version owner or change record. The business goal is to control prompt changes like other production artifacts. 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
- Cost and latency improve while answer grounding passes quality tests
- The assistant admits uncertainty rather than fabricating certainty
- A clean holdout provides a credible lower performance estimate
Explanation
Correct Answer
A:
Prompt templates have no version owner or change record. Only approved changes are deployed and reproducible.
Incorrect Answers
B:
“Cost and latency improve while answer grounding passes quality tests” validates an outcome in a case where the evidence is: Large retrieved context increases token spending and response time. This question instead requires an action that can control prompt changes like other production artifacts. Prompt templates have no version owner or change record. Only approved changes are deployed and reproducible.
C:
“The assistant admits uncertainty rather than fabricating certainty” validates an outcome in a case where the evidence is: Confidence language may conceal missing or stale evidence. This question instead requires an action that can control prompt changes like other production artifacts. Prompt templates have no version owner or change record. Only approved changes are deployed and reproducible.
D:
“A clean holdout provides a credible lower performance estimate” validates an outcome in a case where the evidence is: The same customer records appear in tuning and testing data. This question instead requires an action that can control prompt changes like other production artifacts. Prompt templates have no version owner or change record. Only approved changes are deployed and reproducible.
Question 179
At Blue Ridge Technology, an executive asks an AI agent to map private collaboration networks. Evidence: Graph relationships may expose sensitive work patterns and restricted files. The business goal is to apply authorized scope and privacy review to graph-derived insights. 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 throughput commitments, idle periods and hourly charges
- List which facts need public search versus work file access
- Inventory relevant sources and access scopes for continuous research
- Review purpose, access permissions and sensitive relationship categories
Explanation
Correct Answer
D:
Graph relationships may expose sensitive work patterns and restricted files. Review purpose, access permissions and sensitive relationship categories.
Incorrect Answers
A:
“Identify throughput commitments, idle periods and hourly charges” is a diagnostic step for the reported condition: Dedicated capacity may incur baseline costs even at low traffic. This question instead requires an action that can apply authorized scope and privacy review to graph-derived insights. Graph relationships may expose sensitive work patterns and restricted files. Review purpose, access permissions and sensitive relationship categories.
B:
“List which facts need public search versus work file access” is a diagnostic step for the reported condition: Public web claims and workplace records have different provenance. This question instead requires an action that can apply authorized scope and privacy review to graph-derived insights. Graph relationships may expose sensitive work patterns and restricted files. Review purpose, access permissions and sensitive relationship categories.
C:
“Inventory relevant sources and access scopes for continuous research” is a diagnostic step for the reported condition: Unorganized ad hoc chats repeatedly lose contextual references. This question instead requires an action that can apply authorized scope and privacy review to graph-derived insights. Graph relationships may expose sensitive work patterns and restricted files. Review purpose, access permissions and sensitive relationship categories.
Question 180
At Elmwood Banking, a claims AI pilot reports high answer accuracy but sometimes exposes private fields. Evidence: Data protection failures are not reflected in the accuracy metric. The business goal is to use multidimensional quality and risk gates. The project owner requires a measurable acceptance criterion before signoff. Which evidence would most convincingly demonstrate that the requirement has been met?
- Forecast error beats a defined naive baseline across key items
- Both answers cite the relevant sections without unsupported invented text
- Release is blocked if safety thresholds fail despite accurate answers
- Cost and latency improve while answer grounding passes quality tests
Explanation
Correct Answer
C:
Data protection failures are not reflected in the accuracy metric. Release is blocked if safety thresholds fail despite accurate answers. An answer-quality metric misses unauthorized disclosure and unsafe actions. Use separate privacy, authorization and safety thresholds alongside utility metrics before promotion.
Incorrect Answers
A:
“Forecast error beats a defined naive baseline across key items” validates an outcome in a case where the evidence is: Historical seasonality and promotions drive numerical forecasts. This question instead requires an action that can use multidimensional quality and risk gates. Data protection failures are not reflected in the accuracy metric. Release is blocked if safety thresholds fail despite accurate answers. An answer-quality metric misses unauthorized disclosure and unsafe actions. Use separate privacy, authorization and safety thresholds alongside utility metrics before promotion.
B:
“Both answers cite the relevant sections without unsupported invented text” validates an outcome in a case where the evidence is: Critical terms from the oldest section may be omitted or truncated. This question instead requires an action that can use multidimensional quality and risk gates. Data protection failures are not reflected in the accuracy metric. Release is blocked if safety thresholds fail despite accurate answers. An answer-quality metric misses unauthorized disclosure and unsafe actions. Use separate privacy, authorization and safety thresholds alongside utility metrics before promotion.
D:
“Cost and latency improve while answer grounding passes quality tests” validates an outcome in a case where the evidence is: Large retrieved context increases token spending and response time. This question instead requires an action that can use multidimensional quality and risk gates. Data protection failures are not reflected in the accuracy metric. Release is blocked if safety thresholds fail despite accurate answers. An answer-quality metric misses unauthorized disclosure and unsafe actions. Use separate privacy, authorization and safety thresholds alongside utility metrics before promotion.