Question 500
Site reliability engineers set operational targets for a Foundry application. The team is required to measure the entire agent request including retrieval, tools, and model generation rather than only model time. Which option is the best fit?
- Throughput counter
- End-to-end response duration
- P95 latency percentile
- Time to first token
Explanation
Correct Answer
B:End-to-end duration reflects the user's observed experience. A fast model can still be part of a slow multi-tool workflow.
Incorrect Answers
A:Throughput represents delivered work over time; high throughput alone does not mean responses meet relevance or safety criteria; the outstanding need is to measure the entire agent request including retrieval, tools, and model generation rather than only model time.
C:P95 captures response time near the slow end of the distribution; it helps detect intermittent timeouts hidden by averages; the needed capability is to measure the entire agent request including retrieval, tools, and model generation rather than only model time.
D:Time to first token captures perceived responsiveness for streaming calls; it differs from total completion time for the full answer; the requirement, by contrast, is to measure the entire agent request including retrieval, tools, and model generation rather than only model time.
Question 501
Site reliability engineers set operational targets for a Foundry application. An operator is tasked to track slow-tail response behavior when the median remains healthy. Which mechanism provides that capability?
- P95 latency percentile
- Error budget tracking
- Throughput counter
- End-to-end response duration
Explanation
Correct Answer
A:P95 captures response time near the slow end of the distribution. It helps detect intermittent timeouts hidden by averages.
Incorrect Answers
B:An error budget connects reliability measurements to release and operational decisions; it does not measure whether a generated answer is grounded; the team would still need to track slow-tail response behavior when the median remains healthy.
C:Throughput represents delivered work over time; high throughput alone does not mean responses meet relevance or safety criteria; the specified outcome instead requires the team to track slow-tail response behavior when the median remains healthy.
D:End-to-end duration reflects the user's observed experience; a fast model can still be part of a slow multi-tool workflow; the implementation under review needs to track slow-tail response behavior when the median remains healthy.
Question 502
Site reliability engineers set operational targets for a Foundry application. The immediate design goal is to monitor completed application requests per unit time to understand saturation and scaling needs. Which approach addresses that goal?
- Dependency latency breakdown
- End-to-end response duration
- Throughput counter
- P95 latency percentile
Explanation
Correct Answer
C:Throughput represents delivered work over time. High throughput alone does not mean responses meet relevance or safety criteria.
Incorrect Answers
A:Dependency-level timing identifies the bottleneck responsible for slow interactions; adding model capacity will not help a search service that dominates delay; the implementation under review needs to monitor completed application requests per unit time to understand saturation and scaling needs.
B:End-to-end duration reflects the user's observed experience; a fast model can still be part of a slow multi-tool workflow; the requested task remains to monitor completed application requests per unit time to understand saturation and scaling needs.
D:P95 captures response time near the slow end of the distribution; it helps detect intermittent timeouts hidden by averages; the actual design criterion is to monitor completed application requests per unit time to understand saturation and scaling needs.
Question 503
Site reliability engineers set operational targets for a Foundry application. The team is required to quantify how many failed or excessively slow requests an agreed reliability objective tolerates. Which option is the best fit?
- Error budget tracking
- End-to-end response duration
- Time to first token
- P95 latency percentile
Explanation
Correct Answer
A:An error budget connects reliability measurements to release and operational decisions. It does not measure whether a generated answer is grounded.
Incorrect Answers
B:End-to-end duration reflects the user's observed experience; a fast model can still be part of a slow multi-tool workflow; the outstanding need is to quantify how many failed or excessively slow requests an agreed reliability objective tolerates.
C:Time to first token captures perceived responsiveness for streaming calls; it differs from total completion time for the full answer; this approach leaves the team unable to quantify how many failed or excessively slow requests an agreed reliability objective tolerates.
D:P95 captures response time near the slow end of the distribution; it helps detect intermittent timeouts hidden by averages; the needed capability is to quantify how many failed or excessively slow requests an agreed reliability objective tolerates.
Question 504
Site reliability engineers set operational targets for a Foundry application. An operator is tasked to separate search, tool, and model call time when investigating a rising end-to-end p95. Which mechanism provides that capability?
- End-to-end response duration
- Throughput counter
- P95 latency percentile
- Dependency latency breakdown
- Error budget tracking
Explanation
Correct Answer
D:Dependency-level timing identifies the bottleneck responsible for slow interactions. Adding model capacity will not help a search service that dominates delay.
Incorrect Answers
A:End-to-end duration reflects the user's observed experience; a fast model can still be part of a slow multi-tool workflow; the needed capability is to separate search, tool, and model call time when investigating a rising end-to-end p95.
B:Throughput represents delivered work over time; high throughput alone does not mean responses meet relevance or safety criteria; the requirement, by contrast, is to separate search, tool, and model call time when investigating a rising end-to-end p95.
C:P95 captures response time near the slow end of the distribution; it helps detect intermittent timeouts hidden by averages; the team would still need to separate search, tool, and model call time when investigating a rising end-to-end p95.
E:An error budget connects reliability measurements to release and operational decisions; it does not measure whether a generated answer is grounded; the implementation under review needs to separate search, tool, and model call time when investigating a rising end-to-end p95.
Question 505
The finance and reliability teams need actionable information about generative AI operating expense. The immediate design goal is to track how many prompt and retrieved-context tokens are consumed per call before model generation. Which approach addresses that goal?
- Input token count
- Cache-hit metric
- Cost per completed task
- Model-specific cost attribution
Explanation
Correct Answer
A:Input token metrics reveal prompt bloat and retrieval payload costs. Request counts alone cannot explain a sudden increase in context size.
Incorrect Answers
B:Cache-hit rates show whether caching reduces repeated work; the metric must be interpreted with correctness, staleness, and privacy constraints; the implementation under review needs to track how many prompt and retrieved-context tokens are consumed per call before model generation.
C:Task-normalized cost captures retries and multi-step workflows; a cheap individual call may still be costly when an agent invokes it many times; the requested task remains to track how many prompt and retrieved-context tokens are consumed per call before model generation.
D:Attribution enables targeted optimization of high-cost applications; one pooled cost total cannot show which release caused a spike; this alternative does not achieve the requirement to track how many prompt and retrieved-context tokens are consumed per call before model generation.
Question 506
The finance and reliability teams need actionable information about generative AI operating expense. The team is required to measure generated-token volume so teams can assess response-length cost and throughput effects. Which option is the best fit?
- Cache-hit metric
- Token budget alert
- Model-specific cost attribution
- Output token count
Explanation
Correct Answer
D:Output token metrics reveal how much text the model produces. Long generations can dominate latency and consumption even with short inputs.
Incorrect Answers
A:Cache-hit rates show whether caching reduces repeated work; the metric must be interpreted with correctness, staleness, and privacy constraints; this alternative does not achieve the requirement to measure generated-token volume so teams can assess response-length cost and throughput effects.
B:A budget alert detects unexpected growth in prompt context or generation length; it is a cost signal rather than a factual-quality grade; the actual design criterion is to measure generated-token volume so teams can assess response-length cost and throughput effects.
C:Attribution enables targeted optimization of high-cost applications; one pooled cost total cannot show which release caused a spike; the outstanding need is to measure generated-token volume so teams can assess response-length cost and throughput effects.
Question 507
The finance and reliability teams need actionable information about generative AI operating expense. An operator is tasked to compare end-to-end model and tool expense per successful business outcome rather than cost per raw API call. Which mechanism provides that capability?
- Cost per completed task
- Output token count
- Token budget alert
- Input token count
Explanation
Correct Answer
A:Task-normalized cost captures retries and multi-step workflows. A cheap individual call may still be costly when an agent invokes it many times.
Incorrect Answers
B:Output token metrics reveal how much text the model produces; long generations can dominate latency and consumption even with short inputs; the needed capability is to compare end-to-end model and tool expense per successful business outcome rather than cost per raw API call.
C:A budget alert detects unexpected growth in prompt context or generation length; it is a cost signal rather than a factual-quality grade; the requirement, by contrast, is to compare end-to-end model and tool expense per successful business outcome rather than cost per raw API call.
D:Input token metrics reveal prompt bloat and retrieval payload costs; request counts alone cannot explain a sudden increase in context size; the team would still need to compare end-to-end model and tool expense per successful business outcome rather than cost per raw API call.
Question 508
The finance and reliability teams need actionable information about generative AI operating expense. The immediate design goal is to separate spend by agent, deployment, and workload so owners can act on expensive routes. Which approach addresses that goal?
- Token budget alert
- Cost per completed task
- Model-specific cost attribution
- Input token count
Explanation
Correct Answer
C:Attribution enables targeted optimization of high-cost applications. One pooled cost total cannot show which release caused a spike.
Incorrect Answers
A:A budget alert detects unexpected growth in prompt context or generation length; it is a cost signal rather than a factual-quality grade; the team would still need to separate spend by agent, deployment, and workload so owners can act on expensive routes.
B:Task-normalized cost captures retries and multi-step workflows; a cheap individual call may still be costly when an agent invokes it many times; the specified outcome instead requires the team to separate spend by agent, deployment, and workload so owners can act on expensive routes.
D:Input token metrics reveal prompt bloat and retrieval payload costs; request counts alone cannot explain a sudden increase in context size; the requested task remains to separate spend by agent, deployment, and workload so owners can act on expensive routes.
Question 509
The finance and reliability teams need actionable information about generative AI operating expense. The team is required to trigger investigation when average input or output tokens per request exceed the planned envelope. Which option is the best fit?
- Model-specific cost attribution
- Token budget alert
- Cost per completed task
- Cache-hit metric
Explanation
Correct Answer
B:A budget alert detects unexpected growth in prompt context or generation length. It is a cost signal rather than a factual-quality grade.
Incorrect Answers
A:Attribution enables targeted optimization of high-cost applications; one pooled cost total cannot show which release caused a spike; the requested task remains to trigger investigation when average input or output tokens per request exceed the planned envelope.
C:Task-normalized cost captures retries and multi-step workflows; a cheap individual call may still be costly when an agent invokes it many times; the actual design criterion is to trigger investigation when average input or output tokens per request exceed the planned envelope.
D:Cache-hit rates show whether caching reduces repeated work; the metric must be interpreted with correctness, staleness, and privacy constraints; the outstanding need is to trigger investigation when average input or output tokens per request exceed the planned envelope.
Question 510
The finance and reliability teams need actionable information about generative AI operating expense. An operator is tasked to estimate how often safe response or retrieval caches avoid redundant paid model operations. Which mechanism provides that capability?
- Output token count
- Cache-hit metric
- Cost per completed task
- Token budget alert
Explanation
Correct Answer
B:Cache-hit rates show whether caching reduces repeated work. The metric must be interpreted with correctness, staleness, and privacy constraints.
Incorrect Answers
A:Output token metrics reveal how much text the model produces; long generations can dominate latency and consumption even with short inputs; the outstanding need is to estimate how often safe response or retrieval caches avoid redundant paid model operations.
C:Task-normalized cost captures retries and multi-step workflows; a cheap individual call may still be costly when an agent invokes it many times; the needed capability is to estimate how often safe response or retrieval caches avoid redundant paid model operations.
D:A budget alert detects unexpected growth in prompt context or generation length; it is a cost signal rather than a factual-quality grade; the requirement, by contrast, is to estimate how often safe response or retrieval caches avoid redundant paid model operations.