Question 370 A multi-tool agent is returning unpredictable answers because intermediate tool results are not linked to the generated text in its MLflow experiment. Which verification is strongest for this particular failure mode? Select a failed response in MLflow and navigate to the exact source/tool span responsible for its unsupported content. Use adjudicated summary cases to score citation support and hallucinations separately from n-gram overlap. Replay the invoice and inspect an end-to-end trace linking the request, SQL query span and returned amount. Verify a staging deployment with supported agent entry points, health checks and trace capture against current docs. ExplanationCorrect Answer A:Trace context provides causality between retrieval, tool execution and final generation; isolated logs cannot reconstruct the path. Other Answers B:Select a failed response in MLflow and navigate to the exact source/tool span responsible for its unsupported content. C:Select a failed response in MLflow and navigate to the exact source/tool span responsible for its unsupported content. D:Select a failed response in MLflow and navigate to the exact source/tool span responsible for its unsupported content.
Question 371 A multi-tool agent is returning unpredictable answers because intermediate tool results are not linked to the generated text in its MLflow experiment. What change to engineering governance would reduce this risk over time? Treat high-impact factual errors as hard acceptance criteria in experimental model comparisons. Maintain a platform-version crosswalk and migration test suite rather than hard-coding legacy sample APIs. Require reproducible trace IDs and stable experiment tagging for each agent revision. Standardize trace instrumentation and safe field redaction across agent tools. ExplanationCorrect Answer C:Trace context provides causality between retrieval, tool execution and final generation; isolated logs cannot reconstruct the path. Other Answers A:Trace context provides causality between retrieval, tool execution and final generation; isolated logs cannot reconstruct the path. B:This control addresses a different exposure: A custom Python agent uses an old deployment scaffold that no longer aligns with current Agent Runtime guidance. It would not prevent this failure, because trace context provides causality between retrieval, tool execution and final generation; isolated logs cannot reconstruct the path. D:That trace context provides causality between retrieval, tool execution and final generation; isolated logs cannot reconstruct the path.
Question 372 A project promotes agent code without recording dependency or prompt revision identifiers, making a rollback to known behavior impossible. Which engineering action most directly addresses this specific requirement? Pin compatible framework and tool package versions, validate tool schemas and deploy from a reproducible environment with traceable builds. Reconcile the March exam's Agent Framework terms with the October 2026 Agent Bricks CLI deployment guidance before migrating code. Enable MLflow tracing of the authorized tool invocation, input parameters and result metadata with sensitive values redacted. Register the model or agent artifact and configuration revision with MLflow and link deployed versions to their evaluation runs. ExplanationCorrect Answer D:The action targets the stated cause: Reproducible agent behavior depends on code, model, prompts and dependencies, not just the endpoint name. Check the repair by: restore the previous registered version in staging and verify trace metadata, dependencies and known-good acceptance tests. Other Answers A:Reproducible agent behavior depends on code, model, prompts and dependencies, not just the endpoint name. B:Reproducible agent behavior depends on code, model, prompts and dependencies, not just the endpoint name. C:Reproducible agent behavior depends on code, model, prompts and dependencies, not just the endpoint name.
Question 373 A project promotes agent code without recording dependency or prompt revision identifiers, making a rollback to known behavior impossible. What architectural constraint is most directly responsible? Agent behavior spans tool calls and intermediate messages; missing spans block root-cause attribution. Historical certification vocabulary and current agent serving products can describe different deployment lifecycles. Reproducible agent behavior depends on code, model, prompts and dependencies, not just the endpoint name. Library version drift can alter runnable interfaces or serialization even when model prompts are unchanged. ExplanationCorrect Answer C:Register the model or agent artifact and configuration revision with MLflow and link deployed versions to their evaluation runs. Other Answers A:Reproducible agent behavior depends on code, model, prompts and dependencies, not just the endpoint name. B:That mechanism describes another problem: A custom Python agent uses an old deployment scaffold that no longer aligns with current Agent Runtime guidance. It does not explain this case, where reproducible agent behavior depends on code, model, prompts and dependencies, not just the endpoint name. D:Reproducible agent behavior depends on code, model, prompts and dependencies, not just the endpoint name.
Question 374 A project promotes agent code without recording dependency or prompt revision identifiers, making a rollback to known behavior impossible. What validation evidence would most specifically confirm the resolution? Restore the previous registered version in staging and verify trace metadata, dependencies and known-good acceptance tests. Verify a staging deployment with supported agent entry points, health checks and trace capture against current docs. Run integration tests in the target runtime and compare tool-call payload schemas with the development environment. Replay the invoice and inspect an end-to-end trace linking the request, SQL query span and returned amount. ExplanationCorrect Answer A:Reproducible agent behavior depends on code, model, prompts and dependencies, not just the endpoint name. Other Answers B:Restore the previous registered version in staging and verify trace metadata, dependencies and known-good acceptance tests. C:Restore the previous registered version in staging and verify trace metadata, dependencies and known-good acceptance tests. D:Restore the previous registered version in staging and verify trace metadata, dependencies and known-good acceptance tests.
Question 375 A project promotes agent code without recording dependency or prompt revision identifiers, making a rollback to known behavior impossible. What prevention measure is most targeted to this underlying cause? Standardize trace instrumentation and safe field redaction across agent tools. Treat model and prompt artifact versioning as deployment inputs under change control. Maintain a platform-version crosswalk and migration test suite rather than hard-coding legacy sample APIs. Use dependency lockfiles, controlled upgrades and contract regression tests for all chain adapters. ExplanationCorrect Answer B:Reproducible agent behavior depends on code, model, prompts and dependencies, not just the endpoint name. Other Answers A:This control addresses a different exposure: An agent produces a wrong invoice answer, and its current logs show only the final chat message, not the SQL tool call used. It would not prevent this failure, because reproducible agent behavior depends on code, model, prompts and dependencies, not just the endpoint name. C:That reproducible agent behavior depends on code, model, prompts and dependencies, not just the endpoint name. D:Reproducible agent behavior depends on code, model, prompts and dependencies, not just the endpoint name.
Question 376 A new agent has not shipped, but engineers enable production anomaly dashboards rather than checking whether it answers benchmark questions correctly. Which implementation decision best restores the intended behavior? Use configurable sampled production monitoring with risk-based selection and targeted offline challenges for known severe cases. Add production sampling and MLflow scorer monitoring for groundedness and policy freshness while retaining periodic offline regression tests. Fail the safety and tone criterion independently of policy factuality and route the case through controlled moderation or rewrite review. Run offline evaluation on labeled or curated prompts before deployment, then establish production monitoring after release. ExplanationCorrect Answer D:The action targets the stated cause: Offline evaluation estimates expected quality in controlled conditions; live monitoring detects drift and operational failures after deployment. Check the repair by: confirm evaluation scores on the frozen set before production traffic exists, then validate live-trace scoring once deployed. Other Answers A:Offline evaluation estimates expected quality in controlled conditions; live monitoring detects drift and operational failures after deployment. B:Offline evaluation estimates expected quality in controlled conditions; live monitoring detects drift and operational failures after deployment. C:Offline evaluation estimates expected quality in controlled conditions; live monitoring detects drift and operational failures after deployment.
Question 377 A new agent has not shipped, but engineers enable production anomaly dashboards rather than checking whether it answers benchmark questions correctly. Which underlying technical principle explains why this occurs? Monitoring coverage, inference overhead and tail-risk sensitivity trade off; indiscriminate full-traffic judging may be costly. Offline evaluation estimates expected quality in controlled conditions; live monitoring detects drift and operational failures after deployment. Correct factual content does not make abusive presentation acceptable; quality evaluation needs separate safety dimensions. Static evaluation cannot detect all new input distributions or changing external knowledge after launch. ExplanationCorrect Answer B:Run offline evaluation on labeled or curated prompts before deployment, then establish production monitoring after release. Other Answers A:That mechanism describes another problem: A team proposes running expensive LLM judges on every production request, overwhelming budget while missing low-volume rare failure categories. It does not explain this case, where offline evaluation estimates expected quality in controlled conditions; live monitoring detects drift and operational failures after deployment. C:Offline evaluation estimates expected quality in controlled conditions; live monitoring detects drift and operational failures after deployment. D:Offline evaluation estimates expected quality in controlled conditions; live monitoring detects drift and operational failures after deployment.
Question 378 A new agent has not shipped, but engineers enable production anomaly dashboards rather than checking whether it answers benchmark questions correctly. After correction, what targeted test should demonstrate success? Verify sampling rates and compare flagged error categories against a separate labeled high-risk test set. Confirm evaluation scores on the frozen set before production traffic exists, then validate live-trace scoring once deployed. Score factuality and harmful-language behavior separately on adversarial tone prompts and confirm safe alternatives pass. Track scorer feedback by week and correlate rising unsupported claims with policy-version changes. ExplanationCorrect Answer B:Offline evaluation estimates expected quality in controlled conditions; live monitoring detects drift and operational failures after deployment. Other Answers A:Confirm evaluation scores on the frozen set before production traffic exists, then validate live-trace scoring once deployed. C:Confirm evaluation scores on the frozen set before production traffic exists, then validate live-trace scoring once deployed. D:Confirm evaluation scores on the frozen set before production traffic exists, then validate live-trace scoring once deployed.
Question 379 A new agent has not shipped, but engineers enable production anomaly dashboards rather than checking whether it answers benchmark questions correctly. Which release safeguard most directly addresses this failure class? Define separate pre-release quality gates and post-release monitoring thresholds. Establish sampling budgets plus escalation rules for incident-triggered high-risk audits. Schedule drift and freshness reviews from sampled live traces with approved retention boundaries. Maintain independent safety evaluation thresholds rather than collapsing every outcome into one overall correctness score. ExplanationCorrect Answer A:Offline evaluation estimates expected quality in controlled conditions; live monitoring detects drift and operational failures after deployment. Other Answers B:That offline evaluation estimates expected quality in controlled conditions; live monitoring detects drift and operational failures after deployment. C:Offline evaluation estimates expected quality in controlled conditions; live monitoring detects drift and operational failures after deployment. D:This control addresses a different exposure: A customer-service assistant gives a helpful policy summary but uses harassing language toward the requester in its final paragraph. It would not prevent this failure, because offline evaluation estimates expected quality in controlled conditions; live monitoring detects drift and operational failures after deployment.
Question 380 A deployed support assistant passes its original test set yet slowly degrades as billing policies change over several months. Which design change directly satisfies the stated constraints? Add production sampling and MLflow scorer monitoring for groundedness and policy freshness while retaining periodic offline regression tests. Register the model or agent artifact and configuration revision with MLflow and link deployed versions to their evaluation runs. Use configurable sampled production monitoring with risk-based selection and targeted offline challenges for known severe cases. Run offline evaluation on labeled or curated prompts before deployment, then establish production monitoring after release. ExplanationCorrect Answer A:The action targets the stated cause: Static evaluation cannot detect all new input distributions or changing external knowledge after launch. Check the repair by: track scorer feedback by week and correlate rising unsupported claims with policy-version changes. Other Answers B:Static evaluation cannot detect all new input distributions or changing external knowledge after launch. C:Static evaluation cannot detect all new input distributions or changing external knowledge after launch. D:Static evaluation cannot detect all new input distributions or changing external knowledge after launch.
