Question 290 A Copilot Studio agent serves both internal teams and external customers but shares an Azure AI Search index. Which architecture is necessary? Filter only by language and ignore confidentiality Assume external visitors inherit employee permissions Use one full-access query for every channel Apply channel-appropriate caller authorization and source filters before retrieval Explanation Correct Answer D:Shared indexes must not mix public and private results. Incorrect Answers A:Language is unrelated to data sensitivity. B:No such automatic trust relationship exists. C:External users could see internal content.
Question 291 A product team wants prompt results from a Foundry model to be stable JSON for a downstream workflow. What should be done? Treat a JSON-looking string as automatically schema-correct Constrain the prompt with structured output and validate returned JSON Omit error handling if the model is managed Accept arbitrary prose as an API payload Explanation Correct Answer B:A model can produce invalid or incomplete data without validation. Incorrect Answers A:Syntax does not prove required fields or types. C:Managed models can still produce unexpected responses. D:Downstream tools may fail to parse it.
Question 292 A platform team uses environment-level Application Insights telemetry to investigate nested agents. Which field type matters? Trace and span correlation across invocation and tool dependencies A manually guessed connection between unrelated log rows Only a static count of published agent logos Only the agent's display language Explanation Correct Answer A:Distributed telemetry connects parent and specialist work. Incorrect Answers B:Unrelated events should not be treated as one operation. C:A count gives no call-level sequence. D:Locale is not a trace identity.
Question 293 A company wants alerts when external search dependency failures spike. Which monitoring design fits? Use only the number of helpdesk icons Wait for end users to report every failure Create appropriate query/alert rules on Application Insights dependency failures Block all errors from logs to improve dashboards Explanation Correct Answer C:Aggregated operational telemetry can detect rising error rates. Incorrect Answers A:UI components do not measure API failures. B:Manual reports may miss widespread outages. D:Suppression hides the signal needed for alerts.
Question 294 A developer wants to measure agent answer latency but not capture private message bodies. What configuration approach fits? Enable raw transcripts in all environments by default Publicly post every failure trace Remove all timing information to avoid content capture Collect necessary timing and event metadata with conversation-detail logging constrained Explanation Correct Answer D:Observability and data minimization can coexist. Incorrect Answers A:Full-body logging increases privacy exposure. B:Public traces can reveal internal systems. C:Duration metrics can be collected without body details.
Question 295 An agent uses an Azure AI Search service with semantic retrieval but citations point to wrong document versions. Which part should be inspected? Disable citation generation permanently Ask the model to guess a newer document URL Adjust the citation display format without correcting index document-version mappings Index document keys, stored metadata and source link mappings Explanation Correct Answer D:Incorrect version references can result from bad index identity or metadata. Incorrect Answers A:Hiding links loses provenance rather than fixing it. B:Guessed links are not source evidence. C:A formatting update will preserve the wrong underlying source target.
Question 296 A custom prompt produces output from a Foundry model with usage quotas. What operational failure should the agent handle? Fail over to a model deployment in a different region without data-policy validation Throttling or quota exhaustion with bounded retry and honest status Invent results during quota outages Assume every request runs even after a 429 Explanation Correct Answer B:Model requests can be rejected due to service limits. Incorrect Answers A:Unapproved regional fallback can breach the model-use policy. C:Fabricated outputs are not model execution. D:Throttled calls did not succeed.
Question 297 An agent should retrieve internal contracts and compare clauses. Which search prompt strategy helps? Ignore version and effective date of contracts Ask a language model to invent contract excerpts Use semantic similarity alone without contract-party and clause filters Query for the specified contract and clause, with authorized filters and attributable passages Explanation Correct Answer D:Specific retrieval targets reduce unrelated source mixing. Incorrect Answers A:Superseded contracts may have different clauses. B:Generated quotations are not source text. C:A similar clause from the wrong agreement does not answer the authorized question.
Question 298 An Azure AI Search instance has broad indexing permissions, but the agent's audience is restricted. What must still happen? Trust that the search service enforces a filter never configured Validate runtime query authorization and restrict retrieved records Allow everyone to see what the indexer sees Hide records visually while the model still receives all Explanation Correct Answer B:Indexer permissions do not grant end users access to all indexed items. Incorrect Answers A:Security controls must be explicit and tested. C:That conflates service indexing rights and reader rights. D:Raw context may leak sensitive information.
Question 299 A model produces a good summary but an automated quality evaluation flags missing required fields. What should be changed? Tighten the prompt contract and validate structured output against the schema Disable checks only for production Dismiss schema failures because prose sounds natural Ignore missing fields and ask downstream tools to guess their values Explanation Correct Answer A:Downstream integration requires format as well as semantic quality. Incorrect Answers B:That would conceal a known regression. C:Machine consumers still need required fields. D:Downstream guesses can generate invalid or unauthorized business records.
Question 300 An agent is using Microsoft Foundry for sensitive document classification. Which test distinguishes prompt quality from access problems? Assume a model failure proves a missing SharePoint document Run only as a global administrator Ignore cases where restricted documents appear Separately test model classifications on authorized inputs and connector access under restricted identities Explanation Correct Answer D:A correct classifier cannot compensate for unauthorized input retrieval. Incorrect Answers A:Classification and retrieval are separate stages. B:Admin rights hide permission defects. C:Leaks require access enforcement review.