Correct Answer:

Explanation:
Interactive ticket lookup: Delegated permissions from the signed-in employee; Approved article updates: A managed identity scoped to write production article records.
Ticket search must preserve the existing per-employee authorization model, so the downstream operation should execute in the signed-in employee ' s delegated identity context. That allows the ticket service to enforce the same user-level permissions it already uses. The knowledge-article update is an application- controlled write to one production repository, so a managed workload identity with only the required write permission is the least-privilege choice. Microsoft identity guidance distinguishes delegated/on-behalf-of access for user-context operations from managed or agent identities for service-to-service work. Both approaches also eliminate embedded secrets when configured with Microsoft Entra authentication. A broad Contributor assignment or shared API key would unnecessarily expand the blast radius and weaken audit attribution. Therefore the mixed model in the answer is intentional: delegated permissions for per-user reads, and a narrowly scoped managed identity for controlled production writes. From a security and governance perspective, the control should be enforced at the narrowest platform boundary that can deterministically block or constrain the action. Relying only on prompt text is weaker because the model can still be induced to behave unexpectedly.
Official Microsoft reference: Microsoft Foundry agent identity

Correct Answer:

Explanation:
Initial example source: Synthetic generation; Generation task type: Tool use.
The historical chat logs are both sparse and privacy-sensitive, so using them directly as the initial fine-tuning corpus creates unnecessary PII exposure. Microsoft Foundry synthetic-data generation is intended to create diverse training examples when production data is limited and can avoid carrying customer identifiers into the dataset. The specific behavior that needs improvement is structured API/function calling, so the generator should use the Tool use task type. Tool-use generation can use an API/OpenAPI definition to create conversations that include valid tool selection and parameter construction, which directly trains the failure mode described. A general Q & A generator would not systematically teach function-call schemas. Therefore Synthetic generation plus Tool use is the configuration that both targets invalid arguments and minimizes privacy exposure. At implementation time, the same rule should be expressed through the framework or service configuration rather than left only as a natural-language convention. That makes the behavior repeatable across runs, easier to test, and less sensitive to model variability.
Official Microsoft reference: Microsoft Foundry - synthetic fine-tuning data generation

Correct Answer:

Explanation:
Mandated evidence/tool behavior: Replay cases and assert tool-call and citation presence; Agent responsibility boundaries: Test in-scope responses and out-of-scope refusals; Workflow handoff contract: Replay fixtures and validate schema-conforming payloads.
The three regressions target different interfaces and should be evaluated with tests that directly observe those interfaces. Required evidence gathering is a process behavior, so replayed cases should assert that mandated tool calls and citations occur. Responsibility boundaries are best tested with both positive and negative prompts: in-scope cases must be handled, while out-of-scope work should be refused or redirected. The JSON handoff is an interface contract, so replay fixtures should be validated against the expected schema to catch missing fields, renamed properties, and type changes before downstream agents fail. Microsoft Foundry ' s agent evaluators and evaluation datasets support process-level tool checks, task-adherence checks, and structured regression testing. The supplied mappings therefore correctly align each evaluation technique with the failure it is intended to detect. For operational use, the measurement should be captured in a repeatable dataset, trace, or automated gate so that the same criterion can be compared across versions. That is more useful than a one-off manual observation and makes regressions visible before they become production incidents.
Official Microsoft reference: Microsoft Foundry - built-in evaluators and evaluation datasets

Correct Answer:

Explanation:
Identity structure: Separate blueprints for the orchestrator agent and each worker group; Permission assignment: Assign role-specific downstream permissions to each agent identity.
The supplier and finance workers operate in different trust domains: one reaches external supplier APIs while the other has confidential ERP access. Microsoft Entra Agent ID guidance recommends separating blueprint
/identity trust boundaries when compromise of one agent must not expose unrelated credentials or permissions. Each logical agent identity should then receive only the downstream roles required for its own function. This creates clear audit attribution and limits lateral movement. Giving all workers the same role or routing every privileged operation through an overly powerful orchestrator would expand the blast radius.
Creating an identity for every runtime replica is unnecessary when replicas represent the same logical agent role. The correct structure therefore separates the orchestrator and worker trust domains and assigns role- specific permissions to each identity rather than sharing a common authorization envelope. The same configuration should be paired with auditable identity, trace, and evaluation data so reviewers can prove which principal acted, which policy was applied, and why a request was allowed or blocked. That is particularly important for production multi-agent systems with external tools.
Official Microsoft reference: Microsoft Entra Agent ID - plan agent identity architecture

Correct Answer: D  🗳️

Explanation: Only visible for ValidVCE members. You can sign-up / login (it's free).