答案：

Explanation:
Refund specialist agent: Agent-as-tools; Compliance review agent: Agent-to-Agent (A2A) protocol.
The refund specialist runs in the same process as the triage agent, so exposing it as an agent-as-tool is the lightweight composition pattern. The outer triage agent can decide when to invoke the specialist while retaining overall task ownership. The compliance reviewer is different: it is a remote service owned by a partner team. Agent-to-Agent (A2A) is the interoperable protocol intended for independently hosted agents that may use a different runtime or framework. It supports capability discovery and remote task interaction without coupling the caller to the partner ' s implementation. Using A2A for the in-process specialist would add unnecessary network/protocol complexity, while treating the partner agent as an in-process tool would not match the deployment boundary. The two selections therefore reflect Microsoft ' s distinction between local agent composition and remote agent interoperability. The implementation should also preserve clear inputs and outputs around this step so that later agents receive only the information they require. This improves debuggability and keeps token, permission, and state growth under control as the workflow becomes more complex.
Official Microsoft reference: Microsoft Agent Framework - agents as tools and A2A

答案：

Explanation:
Yes / Yes / No
The first statement is true because the tool-integration contract returns only the explicitly defined `final_text` from FraudReview to ClaimsPlanner; the planner does not automatically receive the specialist ' s entire internal execution history. The second statement is also true because an external REST service such as PremiumRates can be exposed as an OpenAPI-defined tool. The OpenAPI schema describes operations and parameters, while the surrounding agent system retains conversational state; the REST tool itself does not need a developer-managed chat transcript. The third statement is false under the shown configuration because the partner publishes tool descriptors as published while the planner defines which schemas it accepts or rejects. Compatibility handling therefore occurs at the consumer/tool boundary rather than requiring PartnerIntegrations to rewrite every unsupported schema before invocation. The correct sequence is Yes, Yes, No. 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 agents - OpenAPI tools

答案：

Explanation:
Support agent: Document Retrieval; Policy agent: Document Retrieval.
Both failures are retrieval failures and should therefore be diagnosed with the Document Retrieval evaluator.
The dataset includes expected source IDs and ranked retrieved chunks, which provides the retrieval ground truth required by that evaluator. For the support agent, the expected source appears too low in the ranking after unrelated chunks. For the policy agent, the required exception-table source is missing entirely.
Groundedness would evaluate whether the final response is supported by the context it actually received; it could therefore pass even when the retrieval stage failed to fetch a required source. Microsoft Foundry ' s Document Retrieval evaluator compares retrieved documents against labeled ground truth and reports ranking
/search metrics such as NDCG and Fidelity. The corrected answer is Document Retrieval for both agents. 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 - RAG evaluators

答案：

Explanation:
Maximum agent concurrency: Two agents; Dataverse retry behavior: Respect the Retry-After duration returned by Dataverse.
Each analysis agent consumes 500 TPM and the deployment permits 1,000 TPM, so at most two agents can run simultaneously without exceeding the stated model limit. Running only one wastes available parallelism; running three or four violates the limit. The separate Dataverse problem is HTTP 429 service-protection throttling. Microsoft Dataverse returns a `Retry-After` duration that tells clients how long to wait before retrying. Respecting that value prevents premature retries from worsening the overload condition. A fixed retry interval or immediate retry ignores the server ' s current capacity signal. The orchestration should therefore cap model-side concurrency at two and, after each agent finishes, pace Dataverse writes according to the returned retry guidance. This combination minimizes task duration while honoring both provider and platform throttling constraints. 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: Dataverse service protection API limits