100% Latest & Updated Microsoft AB-100 Practice Test Questions, Exam Dumps & Verified Answers!
30 Days Free Updates, Instant Download!
Question 230
A warehouse agent has few complaints but most users manually replace its suggested supplier. Which design decision should the architect prioritize?
- Balance satisfaction with verified factual correctness and task success metrics.
- Triage feedback into defects, safety issues, usability gaps and approved enhancements.
- Measure suggestion acceptance and downstream edits rather than relying on complaint volume.
- Segment experience metrics by case type and examine high-risk cohorts.
Explanation
Correct Answer
C:Behavioral rejection is evidence of low utility even when explicit feedback is sparse.
Incorrect Answers
A:User enjoyment cannot replace objective safety and correctness measures, whereas behavioral rejection is evidence of low utility even when explicit feedback is sparse.
B:A categorized backlog prevents unverified feature requests from bypassing release risk controls. Behavioral rejection is evidence of low utility even when explicit feedback is sparse.
D:Cohort analysis exposes failures hidden by strong overall averages. Behavioral rejection is evidence of low utility even when explicit feedback is sparse.
Question 231
A support bot solves common questions but escalation rates for rare billing cases keep increasing. Which controlled change is most appropriate for this situation?
- Collect structured feedback reasons tied to the specific completed agent task.
- Track authorization-specific error rates by role and release version.
- Balance satisfaction with verified factual correctness and task success metrics.
- Segment experience metrics by case type and examine high-risk cohorts.
Explanation
Correct Answer
D:Cohort analysis exposes failures hidden by strong overall averages.
Incorrect Answers
A:Issue categories reveal whether the problem is grounding, tool use, usability or policy. Cohort analysis exposes failures hidden by strong overall averages.
B:Operational reliability includes scoped access regressions, not only HTTP uptime, but cohort analysis exposes failures hidden by strong overall averages.
C:User enjoyment cannot replace objective safety and correctness measures. Cohort analysis exposes failures hidden by strong overall averages.
Question 232
New prompts produce better positive ratings but also higher factual errors on a test set. Which design choice targets the observed problem?
- Balance satisfaction with verified factual correctness and task success metrics.
- Measure suggestion acceptance and downstream edits rather than relying on complaint volume.
- Collect structured feedback reasons tied to the specific completed agent task.
- Include reopens and first-contact resolution in the service scorecard.
Explanation
Correct Answer
A:User enjoyment cannot replace objective safety and correctness measures.
Incorrect Answers
B:Behavioral rejection is evidence of low utility even when explicit feedback is sparse. User enjoyment cannot replace objective safety and correctness measures.
C:Issue categories reveal whether the problem is grounding, tool use, usability or policy; user enjoyment cannot replace objective safety and correctness measures.
D:Quality-adjusted service metrics detect rework hidden by faster handling time, whereas user enjoyment cannot replace objective safety and correctness measures.
Question 233
An agent stops completing orders when the ERP connector returns HTTP 429 during business peaks. Which architecture response most directly addresses the risk?
- Make business success depend on authoritative API confirmation rather than response wording.
- Track authorization-specific error rates by role and release version.
- Monitor throttling, implement bounded backoff and investigate connector capacity.
- Correlate journey failures with retrieval dependency timeout traces.
Explanation
Correct Answer
C:Rate-limit telemetry and safe retry behavior directly address temporary downstream saturation.
Incorrect Answers
A:Outcome truth must come from the system of record, not the model statement. Rate-limit telemetry and safe retry behavior directly address temporary downstream saturation.
B:Operational reliability includes scoped access regressions, not only HTTP uptime; rate-limit telemetry and safe retry behavior directly address temporary downstream saturation.
D:Dependency-focused tracing distinguishes retrieval failure from model generation problems. Rate-limit telemetry and safe retry behavior directly address temporary downstream saturation.
Question 234
The LLM endpoint is healthy but some agent journeys fail after a knowledge service timeout. Which recommendation addresses the identified failure most directly?
- Correlate customer intent with transaction keys and detect already-completed actions.
- Monitor knowledge version lag and retrieval provenance.
- Make business success depend on authoritative API confirmation rather than response wording.
- Correlate journey failures with retrieval dependency timeout traces.
Explanation
Correct Answer
D:Dependency-focused tracing distinguishes retrieval failure from model generation problems.
Incorrect Answers
A:Cross-channel idempotency prevents duplicate side effects; dependency-focused tracing distinguishes retrieval failure from model generation problems.
B:Stale retrieval is a grounding drift issue, not necessarily model quality regression, whereas dependency-focused tracing distinguishes retrieval failure from model generation problems.
C:Outcome truth must come from the system of record, not the model statement. Dependency-focused tracing distinguishes retrieval failure from model generation problems.
Question 235
A refund agent reports success despite the transaction API returning a rejected payment status. What should the solution architect recommend for this specific need?
- Make business success depend on authoritative API confirmation rather than response wording.
- Monitor throttling, implement bounded backoff and investigate connector capacity.
- Track transaction and idempotency identifiers through agent and tool telemetry.
- Correlate customer intent with transaction keys and detect already-completed actions.
Explanation
Correct Answer
A:Outcome truth must come from the system of record, not the model statement.
Incorrect Answers
B:Rate-limit telemetry and safe retry behavior directly address temporary downstream saturation. Outcome truth must come from the system of record, not the model statement.
C:Cross-component correlation exposes duplicate side effects caused by retried actions, but outcome truth must come from the system of record, not the model statement.
D:Cross-channel idempotency prevents duplicate side effects. Outcome truth must come from the system of record, not the model statement.
Question 236
A customer-facing agent repeats the same action after a user retries a request from another channel. What is the most defensible intervention for this case?
- Monitor throttling, implement bounded backoff and investigate connector capacity.
- Correlate customer intent with transaction keys and detect already-completed actions.
- Break adoption and successful-use metrics down by employee role and channel.
- Correlate journey failures with retrieval dependency timeout traces.
Explanation
Correct Answer
B:Cross-channel idempotency prevents duplicate side effects. This choice resolves the cross-channel action deduplication concern illustrated by the scenario.
Incorrect Answers
A:Rate-limit telemetry and safe retry behavior directly address temporary downstream saturation. Cross-channel idempotency prevents duplicate side effects.
C:Segmented adoption highlights who does not benefit from the rollout. Cross-channel idempotency prevents duplicate side effects.
D:Dependency-focused tracing distinguishes retrieval failure from model generation problems; cross-channel idempotency prevents duplicate side effects.
Question 237
A model worked during the pilot but accuracy falls when a new industry terminology appears in customer requests. Which specific remediation should the business adopt?
- Measure cost and accuracy per task class and route eligible classes to an appropriate model.
- Monitor knowledge version lag and retrieval provenance.
- Compare pre/post model decisions on a stable borderline-case regression set.
- Monitor performance by new terminology cohort and refresh representative evaluations.
Explanation
Correct Answer
D:Input distribution changes can degrade answer accuracy without a platform outage.
Incorrect Answers
A:Cost-aware routing requires task-level benchmarks rather than universal model substitution, but input distribution changes can degrade answer accuracy without a platform outage.
B:Stale retrieval is a grounding drift issue, not necessarily model quality regression. Input distribution changes can degrade answer accuracy without a platform outage.
C:Paired cases expose changed decision boundaries that averages can conceal; input distribution changes can degrade answer accuracy without a platform outage.
Question 238
The source policy changed but agent responses still cite the old policy after a successful model deployment. Which solution approach best meets the stated requirement?
- Monitor knowledge version lag and retrieval provenance.
- Compare pre/post model decisions on a stable borderline-case regression set.
- Monitor out-of-scope intent frequency and escalate scope-change proposals for governance.
- Track accepted policy-compliant recommendations and net reviewer effort.
Explanation
Correct Answer
A:Stale retrieval is a grounding drift issue, not necessarily model quality regression.
Incorrect Answers
B:Paired cases expose changed decision boundaries that averages can conceal; stale retrieval is a grounding drift issue, not necessarily model quality regression.
C:Scope drift should be measured and reviewed, not silently answered with new privileges, whereas stale retrieval is a grounding drift issue, not necessarily model quality regression.
D:Useful task completion and reviewer effort establish realizable benefits. Stale retrieval is a grounding drift issue, not necessarily model quality regression.
Question 239
A model update changes how borderline claims are classified even though overall accuracy stays similar. Which option most directly closes the documented gap?
- Monitor performance by new terminology cohort and refresh representative evaluations.
- Compare pre/post model decisions on a stable borderline-case regression set.
- Measure cost and accuracy per task class and route eligible classes to an appropriate model.
- Monitor out-of-scope intent frequency and escalate scope-change proposals for governance.
Explanation
Correct Answer
B:Paired cases expose changed decision boundaries that averages can conceal.
Incorrect Answers
A:Input distribution changes can degrade answer accuracy without a platform outage; paired cases expose changed decision boundaries that averages can conceal.
C:Cost-aware routing requires task-level benchmarks rather than universal model substitution. Paired cases expose changed decision boundaries that averages can conceal.
D:Scope drift should be measured and reviewed, not silently answered with new privileges, but paired cases expose changed decision boundaries that averages can conceal.
Question 240
Users increasingly ask an agent to perform actions outside its original approved scope. Which approach best satisfies the actual business constraint?
- Correlate customer intent with transaction keys and detect already-completed actions.
- Monitor performance by new terminology cohort and refresh representative evaluations.
- Monitor out-of-scope intent frequency and escalate scope-change proposals for governance.
- Monitor knowledge version lag and retrieval provenance.
Explanation
Correct Answer
C:Scope drift should be measured and reviewed, not silently answered with new privileges.
Incorrect Answers
A:Cross-channel idempotency prevents duplicate side effects, whereas scope drift should be measured and reviewed, not silently answered with new privileges.
B:Input distribution changes can degrade answer accuracy without a platform outage. Scope drift should be measured and reviewed, not silently answered with new privileges.
D:Stale retrieval is a grounding drift issue, not necessarily model quality regression. Scope drift should be measured and reviewed, not silently answered with new privileges.
Agentic AI Business Solutions Architect (AB-100) Practice Tests