Question 680 A team wants to limit one consumer application to a token budget but currently sees only aggregate endpoint usage. Why does the current approach fail under these conditions? Cost visibility requires decomposing usage by request path and token consumption, not relying on a single aggregate invoice. Model serving cost is coupled to request rate and generated tokens; uncontrolled callers can exhaust capacity or spend. Agent spend and reliability include tool invocations; model-only tokens omit MCP service activity. Central gateway controls can attribute usage to callers and enforce request-level policies beyond the raw model API. ExplanationCorrect Answer D:Use approved Unity Gateway usage visibility and caller-scoped rate or budget policies where supported, with clear version/availability caveats. Other Answers A:That mechanism describes another problem: A team sees its monthly model bill grow, but metrics expose only total dollars without token counts or request attribution. It does not explain this case, where central gateway controls can attribute usage to callers and enforce request-level policies beyond the raw model API. B:Central gateway controls can attribute usage to callers and enforce request-level policies beyond the raw model API. C:Central gateway controls can attribute usage to callers and enforce request-level policies beyond the raw model API.
Question 681 A team wants to limit one consumer application to a token budget but currently sees only aggregate endpoint usage. Which observation would establish that the requirement is now met? Load-test above the configured quota and verify excess requests are throttled with clear telemetry. Execute a controlled MCP query and verify a corresponding attributed tool entry appears in permitted usage logs. Reconcile sampled request token usage and model pricing to recorded daily spend per application. Generate traffic from two identities and verify the selected policy throttles only the configured caller. ExplanationCorrect Answer D:Central gateway controls can attribute usage to callers and enforce request-level policies beyond the raw model API. Other Answers A:Generate traffic from two identities and verify the selected policy throttles only the configured caller. B:Generate traffic from two identities and verify the selected policy throttles only the configured caller. C:Generate traffic from two identities and verify the selected policy throttles only the configured caller.
Question 682 A team wants to limit one consumer application to a token budget but currently sees only aggregate endpoint usage. What lasting control best prevents recurrence of the identified issue? Define gateway usage dashboards and per-consumer budget thresholds under change control. Monitor both tool and model events with trace correlation and separate access policies. Maintain rate-limit policies and incident escalation thresholds per consumer identity. Configure usage tagging and cost anomaly alerts for serving endpoints and gateway policies. ExplanationCorrect Answer A:Central gateway controls can attribute usage to callers and enforce request-level policies beyond the raw model API. Other Answers B:That central gateway controls can attribute usage to callers and enforce request-level policies beyond the raw model API. C:Central gateway controls can attribute usage to callers and enforce request-level policies beyond the raw model API. D:This control addresses a different exposure: A team sees its monthly model bill grow, but metrics expose only total dollars without token counts or request attribution. It would not prevent this failure, because central gateway controls can attribute usage to callers and enforce request-level policies beyond the raw model API.
Question 683 A production manager asks whether an external MCP tool is overused, but only LLM model invocation counts are tracked. What is the most targeted corrective step for this observed failure? Apply rate limits and budget safeguards at the approved gateway/service boundary, with graceful error handling for excess traffic. Inspect gateway tool-usage telemetry for the MCP service and attribute calls, errors and latency to the tool and request identity. Enable approved inference telemetry capturing timestamps, latency, token counts, status and trace IDs while respecting privacy constraints. Use approved Unity Gateway usage visibility and caller-scoped rate or budget policies where supported, with clear version/availability caveats. ExplanationCorrect Answer B:The action targets the stated cause: Agent spend and reliability include tool invocations; model-only tokens omit MCP service activity. Check the repair by: execute a controlled MCP query and verify a corresponding attributed tool entry appears in permitted usage logs. Other Answers A:Agent spend and reliability include tool invocations; model-only tokens omit MCP service activity. C:Agent spend and reliability include tool invocations; model-only tokens omit MCP service activity. D:Agent spend and reliability include tool invocations; model-only tokens omit MCP service activity.
Question 684 A production manager asks whether an external MCP tool is overused, but only LLM model invocation counts are tracked. What is the decisive mechanism behind the observed result? Agent spend and reliability include tool invocations; model-only tokens omit MCP service activity. Request totals alone cannot explain tail latency linked to prompt length, retries or endpoint errors. Model serving cost is coupled to request rate and generated tokens; uncontrolled callers can exhaust capacity or spend. Central gateway controls can attribute usage to callers and enforce request-level policies beyond the raw model API. ExplanationCorrect Answer A:Inspect gateway tool-usage telemetry for the MCP service and attribute calls, errors and latency to the tool and request identity. Other Answers B:Agent spend and reliability include tool invocations; model-only tokens omit MCP service activity. C:Agent spend and reliability include tool invocations; model-only tokens omit MCP service activity. D:That mechanism describes another problem: A team wants to limit one consumer application to a token budget but currently sees only aggregate endpoint usage. It does not explain this case, where agent spend and reliability include tool invocations; model-only tokens omit MCP service activity.
Question 685 A production manager asks whether an external MCP tool is overused, but only LLM model invocation counts are tracked. Which verification is strongest for this particular failure mode? Generate traffic from two identities and verify the selected policy throttles only the configured caller. Load-test above the configured quota and verify excess requests are throttled with clear telemetry. Execute a controlled MCP query and verify a corresponding attributed tool entry appears in permitted usage logs. Correlate slow trace IDs with tokens, stage timings and endpoint status across the hourly spike. ExplanationCorrect Answer C:Agent spend and reliability include tool invocations; model-only tokens omit MCP service activity. Other Answers A:Execute a controlled MCP query and verify a corresponding attributed tool entry appears in permitted usage logs. B:Execute a controlled MCP query and verify a corresponding attributed tool entry appears in permitted usage logs. D:Execute a controlled MCP query and verify a corresponding attributed tool entry appears in permitted usage logs.
Question 686 A production manager asks whether an external MCP tool is overused, but only LLM model invocation counts are tracked. What change to engineering governance would reduce this risk over time? Monitor both tool and model events with trace correlation and separate access policies. Define gateway usage dashboards and per-consumer budget thresholds under change control. Maintain rate-limit policies and incident escalation thresholds per consumer identity. Set telemetry retention and completeness tests for supported endpoint observability. ExplanationCorrect Answer A:Agent spend and reliability include tool invocations; model-only tokens omit MCP service activity. Other Answers B:Agent spend and reliability include tool invocations; model-only tokens omit MCP service activity. C:This control addresses a different exposure: A public-facing assistant receives an abrupt bot-driven traffic surge that consumes the team's inference budget within minutes. It would not prevent this failure, because agent spend and reliability include tool invocations; model-only tokens omit MCP service activity. D:That agent spend and reliability include tool invocations; model-only tokens omit MCP service activity.
Question 687 An invoice-extraction agent passes a general fluency judge even when extracted line totals do not equal the declared invoice total. Which engineering action most directly addresses this specific requirement? Implement a deterministic custom scorer that compares parsed line-item sums with the stated total within the business tolerance. Add a task-specific numeric correctness scorer alongside the built-in safety scorer and inspect results per trace. Map the existing expected_answer column explicitly into the scorer invocation and validate its typed inputs before evaluation. Add a custom scorer checking every citation ID against the trace's authorized retrieval-set manifest and relevant supporting text. ExplanationCorrect Answer A:The action targets the stated cause: Generic language-quality judges are not substitutes for exact arithmetic invariants in structured outputs. Check the repair by: score one balanced invoice and one mismatched invoice and confirm only the mismatch fails with a clear reason. Other Answers B:Generic language-quality judges are not substitutes for exact arithmetic invariants in structured outputs. C:Generic language-quality judges are not substitutes for exact arithmetic invariants in structured outputs. D:Generic language-quality judges are not substitutes for exact arithmetic invariants in structured outputs.
Question 688 An invoice-extraction agent passes a general fluency judge even when extracted line totals do not equal the declared invoice total. What architectural constraint is most directly responsible? MLflow scorers measure distinct attributes; passing one risk dimension does not establish task correctness. Citation syntax validity is weaker than provenance integrity; a fabricated identifier can still look well formed. A scorer can misjudge valid traces when its required argument mapping omits available reference data even though the dataset itself is complete. Generic language-quality judges are not substitutes for exact arithmetic invariants in structured outputs. ExplanationCorrect Answer D:Implement a deterministic custom scorer that compares parsed line-item sums with the stated total within the business tolerance. Other Answers A:Generic language-quality judges are not substitutes for exact arithmetic invariants in structured outputs. B:That mechanism describes another problem: A citation judge passes responses with valid-looking IDs even when those IDs never appeared in the retrieved evidence. It does not explain this case, where generic language-quality judges are not substitutes for exact arithmetic invariants in structured outputs. C:Generic language-quality judges are not substitutes for exact arithmetic invariants in structured outputs.
Question 689 An invoice-extraction agent passes a general fluency judge even when extracted line totals do not equal the declared invoice total. What validation evidence would most specifically confirm the resolution? Run traces with known incorrect totals and confirm the numeric scorer flags them even when safety passes. Evaluate one fabricated and one genuinely grounded citation and inspect scorer feedback per trace. Score one balanced invoice and one mismatched invoice and confirm only the mismatch fails with a clear reason. Score one known correct and one known incorrect trace and verify the reference values appear in scorer inputs and recorded feedback. ExplanationCorrect Answer C:Generic language-quality judges are not substitutes for exact arithmetic invariants in structured outputs. Other Answers A:Score one balanced invoice and one mismatched invoice and confirm only the mismatch fails with a clear reason. B:Score one balanced invoice and one mismatched invoice and confirm only the mismatch fails with a clear reason. D:Score one balanced invoice and one mismatched invoice and confirm only the mismatch fails with a clear reason.
Question 690 An invoice-extraction agent passes a general fluency judge even when extracted line totals do not equal the declared invoice total. What prevention measure is most targeted to this underlying cause? Maintain a scorer portfolio aligned to business success, correctness and safety requirements. Maintain provenance-aware scorers and negative citation fixtures across retrieval releases. Keep versioned business-rule scorers and fixture-driven unit tests for numeric invariants. Enforce scorer signature-to-dataset column mapping tests and typed evaluation fixtures on each run. ExplanationCorrect Answer C:Generic language-quality judges are not substitutes for exact arithmetic invariants in structured outputs. Other Answers A:This control addresses a different exposure: An evaluation run produces a high Safety score but no measurement of whether the requested invoice total was calculated correctly. It would not prevent this failure, because generic language-quality judges are not substitutes for exact arithmetic invariants in structured outputs. B:That generic language-quality judges are not substitutes for exact arithmetic invariants in structured outputs. D:Generic language-quality judges are not substitutes for exact arithmetic invariants in structured outputs.
