100% Latest & Updated Microsoft AB-620 Practice Test Questions, Exam Dumps & Verified Answers!
30 Days Free Updates, Instant Download!
Question 440
In an automotive supplier, the monitoring dashboard tracks only successful requests, hiding tool timeouts and throttling. The team selected this design: Instrument failures and dependencies in Application Insights alongside successful operations. Which implementation action BEST puts it into practice?
- Run representative test cases on the new model and review result differences before promotion.
- Filter retrieval to active publication status and return version metadata.
- Record exception types, dependency durations and correlation IDs with redaction.
- Use correlated dependency telemetry to identify and optimize the slow API call.
Explanation
Correct Answer
C:Record exception types, dependency durations and correlation IDs with redaction. The expected result is that the failure-rate dashboard reflects throttled and timed-out calls, not only successes.
Incorrect Answers
A:“Run representative test cases on the new model and review result differences before promotion” could be appropriate when a custom Foundry model prompt is switched without checking safety and accuracy regression. The present task instead requires “Record exception types, dependency durations and correlation IDs with redaction”.
B:Search indexes contain multiple versions of content and the agent cites an obsolete policy. In that setting, “Filter retrieval to active publication status and return version metadata” is relevant; the required choice here is “Record exception types, dependency durations and correlation IDs with redaction”.
D:“Use correlated dependency telemetry to identify and optimize the slow API call” could be appropriate when application Insights shows long tool-call duration but almost no model-generation latency. The present task instead requires “Record exception types, dependency durations and correlation IDs with redaction”.
Question 441
In an automotive supplier, the monitoring dashboard tracks only successful requests, hiding tool timeouts and throttling. After the team implements this action: Record exception types, dependency durations and correlation IDs with redaction. Which evidence BEST verifies the requirement?
- The prompt executes on the approved model and returns a schema-compliant response.
- The failure-rate dashboard reflects throttled and timed-out calls, not only successes.
- A re-run retrieves the new document content rather than outdated text.
- Malformed outputs trigger a safe failure rather than silently updating records.
Explanation
Correct Answer
B:The failure-rate dashboard reflects throttled and timed-out calls, not only successes. This result tests the effect of recording exception types, dependency durations and correlation IDs with redaction.
Incorrect Answers
A:An agent needs a different model from the default for a custom structured summarization prompt. In that setting, “The prompt executes on the approved model and returns a schema-compliant response” is relevant; the required choice here is “The failure-rate dashboard reflects throttled and timed-out calls, not only successes”.
C:“A re-run retrieves the new document content rather than outdated text” could be appropriate when the search index uses a stale document snapshot after source pages are updated. The present task instead requires “The failure-rate dashboard reflects throttled and timed-out calls, not only successes”.
D:A Foundry prompt returns unrestricted free text to a step expecting named fields. In that setting, “Malformed outputs trigger a safe failure rather than silently updating records” is relevant; the required choice here is “The failure-rate dashboard reflects throttled and timed-out calls, not only successes”.
Question 442
In a publisher, search indexes contain multiple versions of content and the agent cites an obsolete policy. Which solution design decision BEST addresses this requirement?
- Improve grounded retrieval configuration and evidence relevance selection.
- Investigate downstream integration performance instead of tuning unrelated generation prompts.
- Apply version filtering and document metadata to Azure AI Search grounding.
- Plan environment-specific model resource configuration with a deployment binding.
Explanation
Correct Answer
C:Apply version filtering and document metadata to Azure AI Search grounding. Implement this design by filter retrieval to active publication status and return version metadata.
Incorrect Answers
A:The agent uses Azure AI Search, but responses include indexed passages that are not relevant to the current question. That condition could justify “Improve grounded retrieval configuration and evidence relevance selection”. Here, “Apply version filtering and document metadata to Azure AI Search grounding” targets the needed outcome.
B:“Investigate downstream integration performance instead of tuning unrelated generation prompts” could be appropriate when application Insights shows long tool-call duration but almost no model-generation latency. The present task instead requires “Apply version filtering and document metadata to Azure AI Search grounding”.
D:A Foundry model deployment used in development does not exist in the production subscription. That condition could justify “Plan environment-specific model resource configuration with a deployment binding”. Here, “Apply version filtering and document metadata to Azure AI Search grounding” targets the needed outcome.
Question 443
In a publisher, search indexes contain multiple versions of content and the agent cites an obsolete policy. The team selected this design: Apply version filtering and document metadata to Azure AI Search grounding. Which implementation action BEST puts it into practice?
- Log source identifiers and response outcomes rather than full sensitive raw inputs.
- Filter retrieval to active publication status and return version metadata.
- Apply the user's authorized filter or permissions when retrieving indexed evidence.
- Exclude tokens and headers while retaining operation IDs, status and latency.
Explanation
Correct Answer
B:Filter retrieval to active publication status and return version metadata. The expected result is that questions cite current approved policy instead of archived drafts.
Incorrect Answers
A:“Log source identifiers and response outcomes rather than full sensitive raw inputs” could be appropriate when the agent must distinguish model hallucination from missing retrieval evidence when an answer is wrong. The present task instead requires “Filter retrieval to active publication status and return version metadata”.
C:A search-backed agent returns data from Azure AI Search but needs access control to follow each requester's identity. That condition could justify “Apply the user's authorized filter or permissions when retrieving indexed evidence”. Here, “Filter retrieval to active publication status and return version metadata” targets the needed outcome.
D:“Exclude tokens and headers while retaining operation IDs, status and latency” could be appropriate when Azure telemetry collection threatens to log raw authentication headers from connector calls. The present task instead requires “Filter retrieval to active publication status and return version metadata”.
Question 444
In a publisher, search indexes contain multiple versions of content and the agent cites an obsolete policy. After the team implements this action: Filter retrieval to active publication status and return version metadata. Which evidence BEST verifies the requirement?
- Each grounded answer can be traced to the retrieved article.
- Reviewers can see whether evidence was missing or the model ignored available evidence.
- Questions cite current approved policy instead of archived drafts.
- Application Insights supports diagnosis without storing credentials.
Explanation
Correct Answer
C:Questions cite current approved policy instead of archived drafts. This result tests the effect of filter retrieval to active publication status and return version metadata.
Incorrect Answers
A:An agent generates answers against Azure AI Search results but omits document attribution. In that setting, “Each grounded answer can be traced to the retrieved article” is relevant; the required choice here is “Questions cite current approved policy instead of archived drafts”.
B:The agent must distinguish model hallucination from missing retrieval evidence when an answer is wrong. That condition could justify “Reviewers can see whether evidence was missing or the model ignored available evidence”. Here, “Questions cite current approved policy instead of archived drafts” targets the needed outcome.
D:Azure telemetry collection threatens to log raw authentication headers from connector calls. In that setting, “Application Insights supports diagnosis without storing credentials” is relevant; the required choice here is “Questions cite current approved policy instead of archived drafts”.
Question 445
In an operations center, the agent must distinguish model hallucination from missing retrieval evidence when an answer is wrong. Which solution design decision BEST addresses this requirement?
- Capture safe retrieval and prompt traces for diagnosis while respecting privacy.
- Constrain model output and validate the structured response before downstream consumption.
- Instrument the solution with Application Insights telemetry for agent operations.
- Plan environment-specific model resource configuration with a deployment binding.
Explanation
Correct Answer
A:Capture safe retrieval and prompt traces for diagnosis while respecting privacy. Implement this design by logging source identifiers and response outcomes rather than full sensitive raw inputs.
Incorrect Answers
B:“Constrain model output and validate the structured response before downstream consumption” could be appropriate when a Foundry prompt returns unrestricted free text to a step expecting named fields. The present task instead requires “Capture safe retrieval and prompt traces for diagnosis while respecting privacy”.
C:An agent tool call intermittently fails and operators need service-level traces. In that setting, “Instrument the solution with Application Insights telemetry for agent operations” is relevant; the required choice here is “Capture safe retrieval and prompt traces for diagnosis while respecting privacy”.
D:A Foundry model deployment used in development does not exist in the production subscription. That condition could justify “Plan environment-specific model resource configuration with a deployment binding”. Here, “Capture safe retrieval and prompt traces for diagnosis while respecting privacy” targets the needed outcome.
Question 446
In an operations center, the agent must distinguish model hallucination from missing retrieval evidence when an answer is wrong. The team selected this design: Capture safe retrieval and prompt traces for diagnosis while respecting privacy. Which implementation action BEST puts it into practice?
- Schedule or trigger appropriate index updates and verify indexing completion.
- Specify the structured output contract and handle parse errors explicitly.
- Log source identifiers and response outcomes rather than full sensitive raw inputs.
- Tune the approved search index/query filtering and use retrieved passages as grounding context.
Explanation
Correct Answer
C:Log source identifiers and response outcomes rather than full sensitive raw inputs. The expected result is that reviewers can see whether evidence was missing or the model ignored available evidence.
Incorrect Answers
A:“Schedule or trigger appropriate index updates and verify indexing completion” could be appropriate when the search index uses a stale document snapshot after source pages are updated. The present task instead requires “Log source identifiers and response outcomes rather than full sensitive raw inputs”.
B:A Foundry prompt returns unrestricted free text to a step expecting named fields. In that setting, “Specify the structured output contract and handle parse errors explicitly” is relevant; the required choice here is “Log source identifiers and response outcomes rather than full sensitive raw inputs”.
D:“Tune the approved search index/query filtering and use retrieved passages as grounding context” could be appropriate when the agent uses Azure AI Search, but responses include indexed passages that are not relevant to the current question. The present task instead requires “Log source identifiers and response outcomes rather than full sensitive raw inputs”.
Question 447
In an operations center, the agent must distinguish model hallucination from missing retrieval evidence when an answer is wrong. After the team implements this action: Log source identifiers and response outcomes rather than full sensitive raw inputs. Which evidence BEST verifies the requirement?
- Each grounded answer can be traced to the retrieved article.
- After optimization, the dependency duration falls and end-to-end agent response improves.
- Malformed outputs trigger a safe failure rather than silently updating records.
- Reviewers can see whether evidence was missing or the model ignored available evidence.
Explanation
Correct Answer
D:Reviewers can see whether evidence was missing or the model ignored available evidence. This result tests the effect of logging source identifiers and response outcomes rather than full sensitive raw inputs.
Incorrect Answers
A:An agent generates answers against Azure AI Search results but omits document attribution. In that setting, “Each grounded answer can be traced to the retrieved article” is relevant; the required choice here is “Reviewers can see whether evidence was missing or the model ignored available evidence”.
B:Application Insights shows long tool-call duration but almost no model-generation latency. That condition could justify “After optimization, the dependency duration falls and end-to-end agent response improves”. Here, “Reviewers can see whether evidence was missing or the model ignored available evidence” targets the needed outcome.
C:“Malformed outputs trigger a safe failure rather than silently updating records” could be appropriate when a Foundry prompt returns unrestricted free text to a step expecting named fields. The present task instead requires “Reviewers can see whether evidence was missing or the model ignored available evidence”.
Question 448
In an energy utility, azure telemetry collection threatens to log raw authentication headers from connector calls. Which solution design decision BEST addresses this requirement?
- Capture safe retrieval and prompt traces for diagnosis while respecting privacy.
- Establish a controlled refresh or reindex process for the Azure AI Search knowledge source.
- Instrument the solution with Application Insights telemetry for agent operations.
- Redact secrets and limit telemetry capture to approved diagnostic metadata.
Explanation
Correct Answer
D:Redact secrets and limit telemetry capture to approved diagnostic metadata. Implement this design by excluding tokens and headers while retaining operation IDs, status and latency.
Incorrect Answers
A:The agent must distinguish model hallucination from missing retrieval evidence when an answer is wrong. That condition could justify “Capture safe retrieval and prompt traces for diagnosis while respecting privacy”. Here, “Redact secrets and limit telemetry capture to approved diagnostic metadata” targets the needed outcome.
B:“Establish a controlled refresh or reindex process for the Azure AI Search knowledge source” could be appropriate when the search index uses a stale document snapshot after source pages are updated. The present task instead requires “Redact secrets and limit telemetry capture to approved diagnostic metadata”.
C:An agent tool call intermittently fails and operators need service-level traces. In that setting, “Instrument the solution with Application Insights telemetry for agent operations” is relevant; the required choice here is “Redact secrets and limit telemetry capture to approved diagnostic metadata”.
Question 449
In an energy utility, azure telemetry collection threatens to log raw authentication headers from connector calls. The team selected this design: Redact secrets and limit telemetry capture to approved diagnostic metadata. Which implementation action BEST puts it into practice?
- Exclude tokens and headers while retaining operation IDs, status and latency.
- Send a redacted, task-specific field set and enforce input validation.
- Configure the prompt/model resource and validate output shape and usage permissions.
- Run representative test cases on the new model and review result differences before promotion.
Explanation
Correct Answer
A:Exclude tokens and headers while retaining operation IDs, status and latency. The expected result is that application Insights supports diagnosis without storing credentials.
Incorrect Answers
B:A custom Foundry prompt accepts raw patient data even though it needs only appointment type. In that setting, “Send a redacted, task-specific field set and enforce input validation” is relevant; the required choice here is “Exclude tokens and headers while retaining operation IDs, status and latency”.
C:An agent needs a different model from the default for a custom structured summarization prompt. That condition could justify “Configure the prompt/model resource and validate output shape and usage permissions”. Here, “Exclude tokens and headers while retaining operation IDs, status and latency” targets the needed outcome.
D:“Run representative test cases on the new model and review result differences before promotion” could be appropriate when a custom Foundry model prompt is switched without checking safety and accuracy regression. The present task instead requires “Exclude tokens and headers while retaining operation IDs, status and latency”.
Question 450
In an energy utility, azure telemetry collection threatens to log raw authentication headers from connector calls. After the team implements this action: Exclude tokens and headers while retaining operation IDs, status and latency. Which evidence BEST verifies the requirement?
- The failure-rate dashboard reflects throttled and timed-out calls, not only successes.
- The prompt executes on the approved model and returns a schema-compliant response.
- After optimization, the dependency duration falls and end-to-end agent response improves.
- Application Insights supports diagnosis without storing credentials.
Explanation
Correct Answer
D:Application Insights supports diagnosis without storing credentials. This result tests the effect of excluding tokens and headers while retaining operation IDs, status and latency.
Incorrect Answers
A:The monitoring dashboard tracks only successful requests, hiding tool timeouts and throttling. In that setting, “The failure-rate dashboard reflects throttled and timed-out calls, not only successes” is relevant; the required choice here is “Application Insights supports diagnosis without storing credentials”.
B:An agent needs a different model from the default for a custom structured summarization prompt. That condition could justify “The prompt executes on the approved model and returns a schema-compliant response”. Here, “Application Insights supports diagnosis without storing credentials” targets the needed outcome.
C:“After optimization, the dependency duration falls and end-to-end agent response improves” could be appropriate when application Insights shows long tool-call duration but almost no model-generation latency. The present task instead requires “Application Insights supports diagnosis without storing credentials”.
Designing and Building Integrated AI Agent Solutions in Copilot Studio (AB-620) Practice Tests