100% Latest & Updated Microsoft AB-100 Practice Test Questions, Exam Dumps & Verified Answers!
30 Days Free Updates, Instant Download!
Question 320
A service demands progressive rollout to a small user group before enabling a new model for everyone. What should the solution architect recommend for this specific need?
- Add automated policy/static checks and review gates to the release pipeline.
- Use a repeatable governed deployment pipeline for solutions and dependencies.
- Use staged/ring deployment with measured rollback criteria.
- Store a release manifest tying artifact versions, tests and approver identity.
Explanation
Correct Answer
C:Canary release limits blast radius and supplies evidence before broad exposure.
Incorrect Answers
A:Release automation should include security gates rather than bypass them, whereas canary release limits blast radius and supplies evidence before broad exposure.
B:Automated promotion reduces inconsistent manual copying while retaining approval gates. Canary release limits blast radius and supplies evidence before broad exposure.
D:A coherent release record supports audit and later incident reconstruction. Canary release limits blast radius and supplies evidence before broad exposure.
Question 321
A prompt change alters refund interpretation, but the team cannot reproduce yesterday’s result. Which approach best satisfies the actual business constraint?
- Version the prompt together with model, tool and grounding configuration identifiers.
- Reuse an approved policy instruction component with controlled ownership.
- Change the source solution in development and promote a tested release.
- Revert to a previously validated prompt version under the normal release process.
Explanation
Correct Answer
A:A reproducible run needs the complete instruction and runtime context, not just a prompt title.
Incorrect Answers
B:Shared policy definitions reduce drift across separately maintained agents, but a reproducible run needs the complete instruction and runtime context, not just a prompt title.
C:Controlled solution lifecycle keeps production changes reproducible and reviewable. A reproducible run needs the complete instruction and runtime context, not just a prompt title.
D:Versioned artifacts make a safe targeted rollback possible; a reproducible run needs the complete instruction and runtime context, not just a prompt title.
Question 322
A deployed agent starts returning worse outputs after a prompt edit. Which specific remediation should the business adopt?
- Reuse an approved policy instruction component with controlled ownership.
- Revert to a previously validated prompt version under the normal release process.
- Define an automatic hold or rollback trigger based on monitored release metrics.
- Define trust-level boundaries and review changes to privileged prompt text.
Explanation
Correct Answer
B:Versioned artifacts make a safe targeted rollback possible.
Incorrect Answers
A:Shared policy definitions reduce drift across separately maintained agents, but versioned artifacts make a safe targeted rollback possible.
C:Rollback criteria make response to a damaging canary outcome timely and auditable; versioned artifacts make a safe targeted rollback possible.
D:System instructions must remain controlled and separate from untrusted customer content, whereas versioned artifacts make a safe targeted rollback possible.
Question 323
Different teams use incompatible policy wording in prompts for the same entitlement rule. Which solution approach best meets the stated requirement?
- Define trust-level boundaries and review changes to privileged prompt text.
- Version the prompt together with model, tool and grounding configuration identifiers.
- Reuse an approved policy instruction component with controlled ownership.
- Coordinate ERP schema and agent connector release compatibility.
Explanation
Correct Answer
C:Shared policy definitions reduce drift across separately maintained agents.
Incorrect Answers
A:System instructions must remain controlled and separate from untrusted customer content. Shared policy definitions reduce drift across separately maintained agents.
B:A reproducible run needs the complete instruction and runtime context, not just a prompt title; shared policy definitions reduce drift across separately maintained agents.
D:Cross-component ALM avoids runtime failures from incompatible data contracts. Shared policy definitions reduce drift across separately maintained agents.
Question 324
A maker embeds an unreviewed customer message as a system instruction in a prompt template. Which option most directly closes the documented gap?
- Version the prompt together with model, tool and grounding configuration identifiers.
- Revert to a previously validated prompt version under the normal release process.
- Define an environment strategy and maker access policies centrally.
- Define trust-level boundaries and review changes to privileged prompt text.
Explanation
Correct Answer
D:System instructions must remain controlled and separate from untrusted customer content.
Incorrect Answers
A:A reproducible run needs the complete instruction and runtime context, not just a prompt title; system instructions must remain controlled and separate from untrusted customer content.
B:Versioned artifacts make a safe targeted rollback possible, whereas system instructions must remain controlled and separate from untrusted customer content.
C:Provisioning and role boundaries prevent uncontrolled proliferation of production workloads. System instructions must remain controlled and separate from untrusted customer content.
Question 325
A support agent still retrieves a revoked policy PDF after the document was withdrawn. Which concrete solution decision should the team make?
- Use a policy-compliant sanitized dataset for nonproduction test grounding.
- Remove or deactivate the revoked knowledge version and verify index refresh.
- Coordinate release notes, support readiness and owner communication with promotion.
- Version retrieval indexes and retain document-lineage metadata.
Explanation
Correct Answer
B:Retrieval lifecycle must reflect authoritative source withdrawal.
Incorrect Answers
A:Testing data should not unnecessarily expose production identities, whereas retrieval lifecycle must reflect authoritative source withdrawal.
C:Operational change management ensures the organization can support newly deployed behavior, but retrieval lifecycle must reflect authoritative source withdrawal.
D:Snapshot provenance enables accurate investigation of historical grounded responses. Retrieval lifecycle must reflect authoritative source withdrawal.
Question 326
A regulated team needs to reproduce which knowledge snapshot supported a decision last month. Which design decision should the architect prioritize?
- Use a policy-compliant sanitized dataset for nonproduction test grounding.
- Version templates and deploy with an approved notice-effective date.
- Version retrieval indexes and retain document-lineage metadata.
- Run schema contract tests and coordinate connector/index updates.
Explanation
Correct Answer
C:Snapshot provenance enables accurate investigation of historical grounded responses.
Incorrect Answers
A:Testing data should not unnecessarily expose production identities. Snapshot provenance enables accurate investigation of historical grounded responses.
B:Service communication consistency requires controlled version and effective-time management, but snapshot provenance enables accurate investigation of historical grounded responses.
D:Data schema changes can break grounding semantics even when a model deployment is unchanged; snapshot provenance enables accurate investigation of historical grounded responses.
Question 327
An agent uses a table copied from production into a test environment without masking real personal data. Which controlled change is most appropriate for this situation?
- Remove or deactivate the revoked knowledge version and verify index refresh.
- Develop in a sandbox environment and promote only after validation.
- Run schema contract tests and coordinate connector/index updates.
- Use a policy-compliant sanitized dataset for nonproduction test grounding.
Explanation
Correct Answer
D:Testing data should not unnecessarily expose production identities.
Incorrect Answers
A:Retrieval lifecycle must reflect authoritative source withdrawal, but testing data should not unnecessarily expose production identities.
B:Sandbox isolation separates experimental actions from production access and data. Testing data should not unnecessarily expose production identities.
C:Data schema changes can break grounding semantics even when a model deployment is unchanged; testing data should not unnecessarily expose production identities.
Question 328
The source system schema changes and an agent silently misreads a renamed account field. Which design choice targets the observed problem?
- Run schema contract tests and coordinate connector/index updates.
- Pin a test corpus and trace the tested code, tools and model versions.
- Remove or deactivate the revoked knowledge version and verify index refresh.
- Version retrieval indexes and retain document-lineage metadata.
Explanation
Correct Answer
A:Data schema changes can break grounding semantics even when a model deployment is unchanged.
Incorrect Answers
B:Versioned evaluation isolates behavior changes across the agent runtime; data schema changes can break grounding semantics even when a model deployment is unchanged.
C:Retrieval lifecycle must reflect authoritative source withdrawal, whereas data schema changes can break grounding semantics even when a model deployment is unchanged.
D:Snapshot provenance enables accurate investigation of historical grounded responses. Data schema changes can break grounding semantics even when a model deployment is unchanged.
Question 329
A fine-tuned model is updated but the organization cannot identify its training data version. Which architecture response most directly addresses the risk?
- Version retrieval indexes and retain document-lineage metadata.
- Restore the last validated model version and investigate with traced examples.
- Register the model artifact with training-data lineage and configuration metadata.
- Gate deployment on the preapproved evaluation threshold, not speed alone.
Explanation
Correct Answer
C:Model registry provenance supports repeatability and governance of updates.
Incorrect Answers
A:Snapshot provenance enables accurate investigation of historical grounded responses; model registry provenance supports repeatability and governance of updates.
B:Rollback and evidence preservation limit harm while engineers isolate a regression, whereas model registry provenance supports repeatability and governance of updates.
D:Model promotion needs measurable business and safety acceptance evidence, but model registry provenance supports repeatability and governance of updates.
Question 330
The previous model scores higher on a safety test but engineers prefer the faster new version. Which recommendation addresses the identified failure most directly?
- Verify actual destination version and a successful end-user smoke test.
- Perform model provenance, licensing and security assessment before adoption.
- Restore the last validated model version and investigate with traced examples.
- Gate deployment on the preapproved evaluation threshold, not speed alone.
Explanation
Correct Answer
D:Model promotion needs measurable business and safety acceptance evidence.
Incorrect Answers
A:Deployment success must be demonstrated in the target environment, whereas model promotion needs measurable business and safety acceptance evidence.
B:Unverified model supply-chain terms can create legal and security exposure. Model promotion needs measurable business and safety acceptance evidence.
C:Rollback and evidence preservation limit harm while engineers isolate a regression, but model promotion needs measurable business and safety acceptance evidence.
Agentic AI Business Solutions Architect (AB-100) Practice Tests