100% Latest & Updated Microsoft AB-620 Practice Test Questions, Exam Dumps & Verified Answers!
30 Days Free Updates, Instant Download!
Question 520
In a research organization, a new agent model responds inconsistently to the same deterministic calculation prompt. Which solution design decision BEST addresses this requirement?
- Evaluate out-of-scope safety and escalation behavior rather than only response fluency.
- Separate task correctness evaluation from user-experience measures.
- Evaluate freshness requirements using time-bounded test data and live source checks.
- Measure repeatability and validate numerical correctness across repeated runs.
Explanation
Correct Answer
D:Measure repeatability and validate numerical correctness across repeated runs. Implement this design by running seeded or repeated representative cases and comparing computed results against known outcomes.
Incorrect Answers
A:A test asks a symptom question and the agent responds as if issuing a clinical diagnosis. That condition could justify “Evaluate out-of-scope safety and escalation behavior rather than only response fluency”. Here, “Measure repeatability and validate numerical correctness across repeated runs” targets the needed outcome.
B:“Separate task correctness evaluation from user-experience measures” could be appropriate when an agent evaluator is asked whether users are satisfied, but only internal correctness tests exist. The present task instead requires “Measure repeatability and validate numerical correctness across repeated runs”.
C:The agent occasionally returns a stale shipment status after a connector refresh. In that setting, “Evaluate freshness requirements using time-bounded test data and live source checks” is relevant; the required choice here is “Measure repeatability and validate numerical correctness across repeated runs”.
Question 521
In a research organization, a new agent model responds inconsistently to the same deterministic calculation prompt. The team selected this design: Measure repeatability and validate numerical correctness across repeated runs. Which implementation action BEST puts it into practice?
- Run both candidates on the same labeled inputs and compare per-case and aggregate outcomes.
- Add representative non-English prompts, expected intent and grounding checks.
- Run seeded or repeated representative cases and compare computed results against known outcomes.
- Replace identifiers with synthetic records and minimize retained raw transcripts.
Explanation
Correct Answer
C:Run seeded or repeated representative cases and compare computed results against known outcomes. The expected result is that flaky numeric responses are identified instead of being hidden by a single pass.
Incorrect Answers
A:“Run both candidates on the same labeled inputs and compare per-case and aggregate outcomes” could be appropriate when the team wants to compare two agent versions on the same customer scenarios. The present task instead requires “Run seeded or repeated representative cases and compare computed results against known outcomes”.
B:An agent fails in multilingual conversations but all current tests use English. In that setting, “Add representative non-English prompts, expected intent and grounding checks” is relevant; the required choice here is “Run seeded or repeated representative cases and compare computed results against known outcomes”.
D:“Replace identifiers with synthetic records and minimize retained raw transcripts” could be appropriate when the test set uses real employee data that should not appear in evaluation logs. The present task instead requires “Run seeded or repeated representative cases and compare computed results against known outcomes”.
Question 522
In a research organization, a new agent model responds inconsistently to the same deterministic calculation prompt. After the team implements this action: Run seeded or repeated representative cases and compare computed results against known outcomes. Which evidence BEST verifies the requirement?
- Flaky numeric responses are identified instead of being hidden by a single pass.
- A malformed response fails even if its narrative text is reasonable.
- The release is blocked if access-control tests fail despite a strong average score.
- Test runs reveal exception failures instead of reporting only successful easy cases.
Explanation
Correct Answer
A:Flaky numeric responses are identified instead of being hidden by a single pass. This result tests the effect of running seeded or repeated representative cases and comparing computed results against known outcomes.
Incorrect Answers
B:A new prompt helps ordinary cases but breaks structured output parsing. That condition could justify “A malformed response fails even if its narrative text is reasonable”. Here, “Flaky numeric responses are identified instead of being hidden by a single pass” targets the needed outcome.
C:“The release is blocked if access-control tests fail despite a strong average score” could be appropriate when a candidate agent release has lower accuracy on rare but high-impact authorization requests. The present task instead requires “Flaky numeric responses are identified instead of being hidden by a single pass”.
D:An agent passes happy-path tests but fails on policy exceptions that occur in real conversations. In that setting, “Test runs reveal exception failures instead of reporting only successful easy cases” is relevant; the required choice here is “Flaky numeric responses are identified instead of being hidden by a single pass”.
Question 523
In an insurance claims team, an agent test set contains only valid claims, ignoring missing attachments and corrupt input files. Which solution design decision BEST addresses this requirement?
- Evaluate tool trace and intermediate decision evidence as well as response wording.
- Test validation and graceful failure on malformed inputs.
- Evaluate out-of-scope safety and escalation behavior rather than only response fluency.
- Expand the evaluation set to supported languages and relevant local phrasing.
Explanation
Correct Answer
B:Test validation and graceful failure on malformed inputs. Implement this design by adding missing-file, invalid-format and partial-record scenarios with expected error behavior.
Incorrect Answers
A:A test suite checks only the final answer, not whether the correct backend was invoked. That condition could justify “Evaluate tool trace and intermediate decision evidence as well as response wording”. Here, “Test validation and graceful failure on malformed inputs” targets the needed outcome.
C:A test asks a symptom question and the agent responds as if issuing a clinical diagnosis. In that setting, “Evaluate out-of-scope safety and escalation behavior rather than only response fluency” is relevant; the required choice here is “Test validation and graceful failure on malformed inputs”.
D:An agent fails in multilingual conversations but all current tests use English. That condition could justify “Expand the evaluation set to supported languages and relevant local phrasing”. Here, “Test validation and graceful failure on malformed inputs” targets the needed outcome.
Question 524
In an insurance claims team, an agent test set contains only valid claims, ignoring missing attachments and corrupt input files. The team selected this design: Test validation and graceful failure on malformed inputs. Which implementation action BEST puts it into practice?
- Add missing-file, invalid-format and partial-record scenarios with expected error behavior.
- Remove repeated scenarios and add independent booking intents and failure modes.
- Add labeled cases for normal questions, policy exceptions and adversarial input with expected behavior.
- Record expected tool/agent calls and compare them with execution traces.
Explanation
Correct Answer
A:Add missing-file, invalid-format and partial-record scenarios with expected error behavior. The expected result is that the agent asks for corrections rather than inventing claim details.
Incorrect Answers
B:An evaluation set contains duplicate questions and yields an overly optimistic success rate. In that setting, “Remove repeated scenarios and add independent booking intents and failure modes” is relevant; the required choice here is “Add missing-file, invalid-format and partial-record scenarios with expected error behavior”.
C:An agent passes happy-path tests but fails on policy exceptions that occur in real conversations. That condition could justify “Add labeled cases for normal questions, policy exceptions and adversarial input with expected behavior”. Here, “Add missing-file, invalid-format and partial-record scenarios with expected error behavior” targets the needed outcome.
D:“Record expected tool/agent calls and compare them with execution traces” could be appropriate when a test suite checks only the final answer, not whether the correct backend was invoked. The present task instead requires “Add missing-file, invalid-format and partial-record scenarios with expected error behavior”.
Question 525
In an insurance claims team, an agent test set contains only valid claims, ignoring missing attachments and corrupt input files. After the team implements this action: Add missing-file, invalid-format and partial-record scenarios with expected error behavior. Which evidence BEST verifies the requirement?
- Independent reviewers assign consistent outcomes to the same result.
- A measured difference reflects the tested versions rather than changing question sets.
- The agent asks for corrections rather than inventing claim details.
- The agent ignores the injected commands and follows the authorized user task.
Explanation
Correct Answer
C:The agent asks for corrections rather than inventing claim details. This result tests the effect of adding missing-file, invalid-format and partial-record scenarios with expected error behavior.
Incorrect Answers
A:Testers disagree on whether a partially correct answer should pass. In that setting, “Independent reviewers assign consistent outcomes to the same result” is relevant; the required choice here is “The agent asks for corrections rather than inventing claim details”.
B:The team wants to compare two agent versions on the same customer scenarios. That condition could justify “A measured difference reflects the tested versions rather than changing question sets”. Here, “The agent asks for corrections rather than inventing claim details” targets the needed outcome.
D:A model update changes the way an agent handles malicious instructions embedded in documents. In that setting, “The agent ignores the injected commands and follows the authorized user task” is relevant; the required choice here is “The agent asks for corrections rather than inventing claim details”.
Question 526
In a data privacy office, the team cannot prove what changed between agent evaluations across releases. Which solution design decision BEST addresses this requirement?
- Choose an evaluation method that checks factual grounding and citation provenance.
- Test permission-boundary and approval behavior explicitly.
- Measure repeatability and validate numerical correctness across repeated runs.
- Record version identifiers, test-set revision and scoring rubric with each run.
Explanation
Correct Answer
D:Record version identifiers, test-set revision and scoring rubric with each run. Implement this design by save immutable evaluation metadata and comparable per-case results.
Incorrect Answers
A:Agent answers must be correct and cite controlling regulatory documents. That condition could justify “Choose an evaluation method that checks factual grounding and citation provenance”. Here, “Record version identifiers, test-set revision and scoring rubric with each run” targets the needed outcome.
B:“Test permission-boundary and approval behavior explicitly” could be appropriate when a draft agent recommends running destructive commands when only read-only troubleshooting is permitted. The present task instead requires “Record version identifiers, test-set revision and scoring rubric with each run”.
C:A new agent model responds inconsistently to the same deterministic calculation prompt. In that setting, “Measure repeatability and validate numerical correctness across repeated runs” is relevant; the required choice here is “Record version identifiers, test-set revision and scoring rubric with each run”.
Question 527
In a data privacy office, the team cannot prove what changed between agent evaluations across releases. The team selected this design: Record version identifiers, test-set revision and scoring rubric with each run. Which implementation action BEST puts it into practice?
- Run seeded or repeated representative cases and compare computed results against known outcomes.
- Remove repeated scenarios and add independent booking intents and failure modes.
- Include tests that validate required output fields, types and allowed values.
- Save immutable evaluation metadata and comparable per-case results.
Explanation
Correct Answer
D:Save immutable evaluation metadata and comparable per-case results. The expected result is that auditors can reproduce which agent build and test definitions produced each score.
Incorrect Answers
A:“Run seeded or repeated representative cases and compare computed results against known outcomes” could be appropriate when a new agent model responds inconsistently to the same deterministic calculation prompt. The present task instead requires “Save immutable evaluation metadata and comparable per-case results”.
B:An evaluation set contains duplicate questions and yields an overly optimistic success rate. In that setting, “Remove repeated scenarios and add independent booking intents and failure modes” is relevant; the required choice here is “Save immutable evaluation metadata and comparable per-case results”.
C:A new prompt helps ordinary cases but breaks structured output parsing. That condition could justify “Include tests that validate required output fields, types and allowed values”. Here, “Save immutable evaluation metadata and comparable per-case results” targets the needed outcome.
Question 528
In a data privacy office, the team cannot prove what changed between agent evaluations across releases. After the team implements this action: Save immutable evaluation metadata and comparable per-case results. Which evidence BEST verifies the requirement?
- Auditors can reproduce which agent build and test definitions produced each score.
- Evaluation output reveals no actual employee personal information.
- A test cannot cause an unapproved production change.
- The agent ignores the injected commands and follows the authorized user task.
Explanation
Correct Answer
A:Auditors can reproduce which agent build and test definitions produced each score. This result tests the effect of save immutable evaluation metadata and comparable per-case results.
Incorrect Answers
B:The test set uses real employee data that should not appear in evaluation logs. That condition could justify “Evaluation output reveals no actual employee personal information”. Here, “Auditors can reproduce which agent build and test definitions produced each score” targets the needed outcome.
C:“A test cannot cause an unapproved production change” could be appropriate when a draft agent recommends running destructive commands when only read-only troubleshooting is permitted. The present task instead requires “Auditors can reproduce which agent build and test definitions produced each score”.
D:A model update changes the way an agent handles malicious instructions embedded in documents. In that setting, “The agent ignores the injected commands and follows the authorized user task” is relevant; the required choice here is “Auditors can reproduce which agent build and test definitions produced each score”.
Question 529
In an enterprise developer team, an agent was built in an unmanaged default solution but needs controlled deployment. Which solution design decision BEST addresses this requirement?
- Replace maker-only dependencies with properly governed production connection identities.
- Place the agent and dependencies in a custom Power Platform solution for ALM.
- Include and configure connection references as ALM dependencies.
- Add automated evaluation and release quality gates to the ALM pipeline.
Explanation
Correct Answer
B:Place the agent and dependencies in a custom Power Platform solution for ALM. Implement this design by creating an appropriate publisher/solution and adding the agent components.
Incorrect Answers
A:A deployed agent unexpectedly uses a developer's personal connector credentials. That condition could justify “Replace maker-only dependencies with properly governed production connection identities”. Here, “Place the agent and dependencies in a custom Power Platform solution for ALM” targets the needed outcome.
C:The agent works in development but fails in testing because its connector's connection reference was not mapped. In that setting, “Include and configure connection references as ALM dependencies” is relevant; the required choice here is “Place the agent and dependencies in a custom Power Platform solution for ALM”.
D:A pipeline promotes agents without checking response quality or safety regressions. That condition could justify “Add automated evaluation and release quality gates to the ALM pipeline”. Here, “Place the agent and dependencies in a custom Power Platform solution for ALM” targets the needed outcome.
Question 530
In an enterprise developer team, an agent was built in an unmanaged default solution but needs controlled deployment. The team selected this design: Place the agent and dependencies in a custom Power Platform solution for ALM. Which implementation action BEST puts it into practice?
- Review the solution inventory and remove draft dependencies before export.
- Keep a verified prior solution version and a documented recovery procedure.
- Create an appropriate publisher/solution and add the agent components.
- Run representative test sets and enforce pass thresholds before promotion.
Explanation
Correct Answer
C:Create an appropriate publisher/solution and add the agent components. The expected result is that the managed release package includes the intended agent and dependencies.
Incorrect Answers
A:“Review the solution inventory and remove draft dependencies before export” could be appropriate when a solution build includes unfinished experimental tools not authorized for release. The present task instead requires “Create an appropriate publisher/solution and add the agent components”.
B:An agent's latest version fails user testing after deployment but the team has no rollback plan. In that setting, “Keep a verified prior solution version and a documented recovery procedure” is relevant; the required choice here is “Create an appropriate publisher/solution and add the agent components”.
D:“Run representative test sets and enforce pass thresholds before promotion” could be appropriate when a pipeline promotes agents without checking response quality or safety regressions. The present task instead requires “Create an appropriate publisher/solution and add the agent components”.
Designing and Building Integrated AI Agent Solutions in Copilot Studio (AB-620) Practice Tests