Question 450
A customer maintenance team receives application code but cannot run the model quality checks on its own. What independently verifiable observation is needed?
- The verified result is that engineers can find current and obsolete decisions with their reasons and review dates.
- The recorded test establishes that a fresh environment rebuilds the index and reproduces known evidence retrieval.
- The representative case test finds that a drill restores the preceding prompt-model-index combination without inconsistent state.
- The staged rollout demonstrates that the receiving team executes a regression test and interprets the results without vendor help.
Explanation
Correct Answer
D:Operations must be able to assess ongoing correctness independently of the original builders; the safeguard must work when a client receives final source code but no explanation of how to measure quality after deployment.
Incorrect Answers
A:Decision provenance reduces rediscovery when regulations, usage and models evolve; this makes sense when a solution has dozens of architectural assumptions recorded in chat messages only, rather than the risk that a customer maintenance team receives application code but cannot run the model quality checks on its own.
B:Retrieval behavior cannot be recreated from the generation prompt alone; this technique fits where a team documents prompts but not which corpus, embedding model and chunking policy produced answers, rather than the condition that a customer maintenance team receives application code but cannot run the model quality checks on its own.
C:Recovery readiness depends on known-good artifacts and practiced restoration steps; the trigger would be that a service outage requires rollback, but deployment notes contain no previous artifact versions, while the immediate question involves a customer maintenance team receives application code but cannot run the model quality checks on its own.
Question 451
A replay of an asynchronous workflow creates duplicate actions because no one recorded queue retry and idempotency semantics. What outcome would demonstrate this requirement in operation?
- The readiness assessment shows that on-call staff can identify permission owners and rotate credentials using the runbook.
- The independent test establishes that an operations drill identifies which queue can retry safely after a partial failure.
- The verified result is that a fresh environment rebuilds the index and reproduces known evidence retrieval.
- The recorded test establishes that engineers can find current and obsolete decisions with their reasons and review dates.
Explanation
Correct Answer
B:Multi-stage side effects need explicit operational contracts to prevent replay and duplication errors; the condition under test is that an orchestrated workflow uses multiple asynchronous queues whose retry semantics are undocumented.
Incorrect Answers
A:A supportable deployment requires responsibility and recovery documentation beyond diagrams; the other workload is one where the delivery team leaves after launch with no record of tool privileges or secret owners, not the risk where a replay of an asynchronous workflow creates duplicate actions because no one recorded queue retry and idempotency semantics.
C:Retrieval behavior cannot be recreated from the generation prompt alone; the alternative helps if a team documents prompts but not which corpus, embedding model and chunking policy produced answers, whereas this scenario involves a replay of an asynchronous workflow creates duplicate actions because no one recorded queue retry and idempotency semantics.
D:Decision provenance reduces rediscovery when regulations, usage and models evolve; it is justified if a solution has dozens of architectural assumptions recorded in chat messages only, but here a replay of an asynchronous workflow creates duplicate actions because no one recorded queue retry and idempotency semantics.
Question 452
A team plans to deploy directly after a working demo with no acceptance baseline. What change directly resolves this production-design issue?
- For the intended users, establish service ownership, on-call processes and post-launch budgets before handover.
- Within the delivery plan, introduce discovery, feasibility, pilot evaluation and staged release gates.
- At the system boundary, review demand forecasts, unit economics and capacity thresholds in regular lifecycle reviews.
- For the primary workstream, operate a change-management loop with pinned versions, canaries and rollback tests.
Explanation
Correct Answer
B:A demo establishes possibility, not production suitability under real data and governance constraints; a representative trial should prove that release records contain approved acceptance evidence for every transition.
Incorrect Answers
A:Operational readiness determines whether a technically sound pilot can sustain production service; it serves conditions where a pilot meets quality goals but fails due to unclear ongoing support responsibilities, but this decision concerns a team plans to deploy directly after a working demo with no acceptance baseline.
C:Scaling changes total cost and operational risk even when per-request behavior remains stable; this makes sense when a service costs twice its forecast after traffic grows tenfold, but the requested boundary concerns a team plans to deploy directly after a working demo with no acceptance baseline.
D:Model behavior belongs in the product lifecycle even when infrastructure configuration is stable; this technique fits where a model provider releases an update and production answers change unexpectedly, not the present case where a team plans to deploy directly after a working demo with no acceptance baseline.
Question 453
A model provider releases an update and production answers change unexpectedly. Which proposed design action directly addresses the risk?
- Under the reviewed approach, review demand forecasts, unit economics and capacity thresholds in regular lifecycle reviews.
- For this production path, operate a change-management loop with pinned versions, canaries and rollback tests.
- In the proposed implementation, establish service ownership, on-call processes and post-launch budgets before handover.
- Within the delivery plan, schedule ongoing retrieval freshness and quality checks linked to document ingestion.
- At the system boundary, define secure retirement covering indexes, logs, credentials and customer obligations.
Explanation
Correct Answer
B:Model behavior belongs in the product lifecycle even when infrastructure configuration is stable; the controlled test needs to find that canary evaluation detects regressions before the update reaches most users.
Incorrect Answers
A:Scaling changes total cost and operational risk even when per-request behavior remains stable; it serves a setting where a service costs twice its forecast after traffic grows tenfold, rather than the condition that a model provider releases an update and production answers change unexpectedly.
C:Operational readiness determines whether a technically sound pilot can sustain production service; the other workload is one where a pilot meets quality goals but fails due to unclear ongoing support responsibilities, but the present constraint is that a model provider releases an update and production answers change unexpectedly.
D:Continuous content change can invalidate a once-successful retrieval architecture; this applies when a knowledge corpus changes daily but the solution owner reviews quality only once per year, whereas the current issue is that a model provider releases an update and production answers change unexpectedly.
E:Lifecycle stewardship includes removing residual access and data, not only deleting a front-end application; the alternative helps if a system is decommissioned but old vectors, model logs and service tokens still exist, whereas the release criterion addresses a model provider releases an update and production answers change unexpectedly.
Question 454
A knowledge assistant runs on a corpus updated daily, while traffic is increasing faster than originally forecast and inference costs have surged. Which TWO ongoing lifecycle controls should the service owner institute to maintain freshness and predict operating cost? Choose TWO.
- In the initial solution, reassess permissions, abuse cases and safety evaluations whenever the tool surface changes.
- In the proposed design, introduce discovery, feasibility, pilot evaluation and staged release gates.
- Under the reviewed approach, operate a change-management loop with pinned versions, canaries and rollback tests.
- For this production path, schedule ongoing retrieval freshness and quality checks linked to document ingestion.
- For the intended users, review demand forecasts, unit economics and capacity thresholds in regular lifecycle reviews.
Explanation
Correct Answers
D:Continuous content change can invalidate a once-successful retrieval architecture; validation needs to show that automated tests detect stale documents within the declared update window; another obligation arises when a service costs twice its forecast after traffic grows tenfold.
E:Scaling changes total cost and operational risk even when per-request behavior remains stable; one operational check should confirm that capacity plans and cost dashboards trigger action before agreed budget ceilings are exceeded; another obligation arises when a knowledge corpus changes daily but the solution owner reviews quality only once per year.
Incorrect Answers
A:New external actions can invalidate earlier assumptions about what an agent is allowed to do; that belongs to situations where a product adds new tool capabilities without revisiting the original threat model, not this situation where a knowledge corpus changes daily but the solution owner reviews quality only once per year; another obligation arises when a service costs twice its forecast after traffic grows tenfold.
B:A demo establishes possibility, not production suitability under real data and governance constraints; it is relevant if a team plans to deploy directly after a working demo with no acceptance baseline, whereas this scenario involves a service costs twice its forecast after traffic grows tenfold; another obligation arises when a knowledge corpus changes daily but the solution owner reviews quality only once per year.
C:Model behavior belongs in the product lifecycle even when infrastructure configuration is stable; it serves conditions where a model provider releases an update and production answers change unexpectedly, but here a knowledge corpus changes daily but the solution owner reviews quality only once per year; another obligation arises when a service costs twice its forecast after traffic grows tenfold.
Question 455
A pilot meets quality goals but fails due to unclear ongoing support responsibilities. Which action addresses the root concern without unnecessary scope?
- At this decision point, establish service ownership, on-call processes and post-launch budgets before handover.
- As a candidate control, operate a change-management loop with pinned versions, canaries and rollback tests.
- In the initial solution, review demand forecasts, unit economics and capacity thresholds in regular lifecycle reviews.
- In the proposed design, reassess permissions, abuse cases and safety evaluations whenever the tool surface changes.
Explanation
Correct Answer
A:Operational readiness determines whether a technically sound pilot can sustain production service; a representative trial should prove that a readiness review names incident owners and funds evaluation and maintenance work.
Incorrect Answers
B:Model behavior belongs in the product lifecycle even when infrastructure configuration is stable; that would be useful if a model provider releases an update and production answers change unexpectedly, but the requested boundary concerns a pilot meets quality goals but fails due to unclear ongoing support responsibilities.
C:Scaling changes total cost and operational risk even when per-request behavior remains stable; it serves a setting where a service costs twice its forecast after traffic grows tenfold, not the present case where a pilot meets quality goals but fails due to unclear ongoing support responsibilities.
D:New external actions can invalidate earlier assumptions about what an agent is allowed to do; that is defensible if a product adds new tool capabilities without revisiting the original threat model, while this problem involves a pilot meets quality goals but fails due to unclear ongoing support responsibilities.
Question 456
A product adds new tool capabilities without revisiting the original threat model. What is the most appropriate initial architecture decision?
- During architecture review, introduce discovery, feasibility, pilot evaluation and staged release gates.
- In the proposed implementation, review demand forecasts, unit economics and capacity thresholds in regular lifecycle reviews.
- At this decision point, reassess permissions, abuse cases and safety evaluations whenever the tool surface changes.
- As a candidate control, define secure retirement covering indexes, logs, credentials and customer obligations.
Explanation
Correct Answer
C:New external actions can invalidate earlier assumptions about what an agent is allowed to do; the go-live check should establish that the release gate records refreshed abuse testing for each added capability.
Incorrect Answers
A:A demo establishes possibility, not production suitability under real data and governance constraints; it targets cases where a team plans to deploy directly after a working demo with no acceptance baseline, while the immediate question involves a product adds new tool capabilities without revisiting the original threat model.
B:Scaling changes total cost and operational risk even when per-request behavior remains stable; the approach helps when a service costs twice its forecast after traffic grows tenfold, but the present constraint is that a product adds new tool capabilities without revisiting the original threat model.
D:Lifecycle stewardship includes removing residual access and data, not only deleting a front-end application; it is relevant if a system is decommissioned but old vectors, model logs and service tokens still exist, whereas the release criterion addresses a product adds new tool capabilities without revisiting the original threat model.
Question 457
A service costs twice its forecast after traffic grows tenfold. Which architecture change should be approved first?
- As a scoped first step, introduce discovery, feasibility, pilot evaluation and staged release gates.
- In the release design, operate a change-management loop with pinned versions, canaries and rollback tests.
- During architecture review, review demand forecasts, unit economics and capacity thresholds in regular lifecycle reviews.
- In the proposed implementation, reassess permissions, abuse cases and safety evaluations whenever the tool surface changes.
Explanation
Correct Answer
C:Scaling changes total cost and operational risk even when per-request behavior remains stable; validation needs to show that capacity plans and cost dashboards trigger action before agreed budget ceilings are exceeded.
Incorrect Answers
A:A demo establishes possibility, not production suitability under real data and governance constraints; the design helps when a team plans to deploy directly after a working demo with no acceptance baseline, whereas this scenario involves a service costs twice its forecast after traffic grows tenfold.
B:Model behavior belongs in the product lifecycle even when infrastructure configuration is stable; that fits cases where a model provider releases an update and production answers change unexpectedly, but here a service costs twice its forecast after traffic grows tenfold.
D:New external actions can invalidate earlier assumptions about what an agent is allowed to do; that would be useful if a product adds new tool capabilities without revisiting the original threat model, not the case in which a service costs twice its forecast after traffic grows tenfold.
Question 458
A system is decommissioned but old vectors, model logs and service tokens still exist. What should be incorporated into the approved design?
- For the operating team, review demand forecasts, unit economics and capacity thresholds in regular lifecycle reviews.
- For this deployment, reassess permissions, abuse cases and safety evaluations whenever the tool surface changes.
- As a scoped first step, define secure retirement covering indexes, logs, credentials and customer obligations.
- In the release design, schedule ongoing retrieval freshness and quality checks linked to document ingestion.
- During architecture review, operate a change-management loop with pinned versions, canaries and rollback tests.
Explanation
Correct Answer
C:Lifecycle stewardship includes removing residual access and data, not only deleting a front-end application; the practical acceptance condition is that a retirement checklist confirms revoked tokens and policy-compliant disposal of stored artifacts.
Incorrect Answers
A:Scaling changes total cost and operational risk even when per-request behavior remains stable; it is justified if a service costs twice its forecast after traffic grows tenfold, but the requested boundary concerns a system is decommissioned but old vectors, model logs and service tokens still exist.
B:New external actions can invalidate earlier assumptions about what an agent is allowed to do; its scope includes cases where a product adds new tool capabilities without revisiting the original threat model, not the present case where a system is decommissioned but old vectors, model logs and service tokens still exist.
D:Continuous content change can invalidate a once-successful retrieval architecture; the approach helps when a knowledge corpus changes daily but the solution owner reviews quality only once per year, and not where a system is decommissioned but old vectors, model logs and service tokens still exist.
E:Model behavior belongs in the product lifecycle even when infrastructure configuration is stable; that belongs to situations where a model provider releases an update and production answers change unexpectedly, rather than the risk that a system is decommissioned but old vectors, model logs and service tokens still exist.
Question 459
An enthusiastic project team calls a demonstration 'launch complete' before any production acceptance evidence exists. What validation evidence would address this failure mode?
- The end-to-end check confirms that release records contain approved acceptance evidence for every transition.
- The observed outcome is that a retirement checklist confirms revoked tokens and policy-compliant disposal of stored artifacts.
- The regression suite verifies that canary evaluation detects regressions before the update reaches most users.
- A production-like test shows that the release gate records refreshed abuse testing for each added capability.
Explanation
Correct Answer
A:A demo establishes possibility, not production suitability under real data and governance constraints; the safeguard must work when a team plans to deploy directly after a working demo with no acceptance baseline.
Incorrect Answers
B:Lifecycle stewardship includes removing residual access and data, not only deleting a front-end application; the trigger would be that a system is decommissioned but old vectors, model logs and service tokens still exist, whereas the current issue is that an enthusiastic project team calls a demonstration 'launch complete' before any production acceptance evidence exists.
C:Model behavior belongs in the product lifecycle even when infrastructure configuration is stable; the design helps when a model provider releases an update and production answers change unexpectedly, whereas the release criterion addresses an enthusiastic project team calls a demonstration 'launch complete' before any production acceptance evidence exists.
D:New external actions can invalidate earlier assumptions about what an agent is allowed to do; that fits cases where a product adds new tool capabilities without revisiting the original threat model, while the required safeguard concerns an enthusiastic project team calls a demonstration 'launch complete' before any production acceptance evidence exists.
Question 460
A model change is approved in staging, but a canary later discovers changed classifications for real requests. What end-to-end observation would best settle the concern?
- The measured evidence confirms that a retirement checklist confirms revoked tokens and policy-compliant disposal of stored artifacts.
- The scenario test demonstrates that release records contain approved acceptance evidence for every transition.
- The end-to-end check confirms that the release gate records refreshed abuse testing for each added capability.
- The observed outcome is that automated tests detect stale documents within the declared update window.
- The representative case test finds that canary evaluation detects regressions before the update reaches most users.
Explanation
Correct Answer
E:Model behavior belongs in the product lifecycle even when infrastructure configuration is stable; the control should hold even if a model provider releases an update and production answers change unexpectedly.
Incorrect Answers
A:Lifecycle stewardship includes removing residual access and data, not only deleting a front-end application; this applies when a system is decommissioned but old vectors, model logs and service tokens still exist, but here a model change is approved in staging, but a canary later discovers changed classifications for real requests.
B:A demo establishes possibility, not production suitability under real data and governance constraints; the alternative helps if a team plans to deploy directly after a working demo with no acceptance baseline, instead of when a model change is approved in staging, but a canary later discovers changed classifications for real requests.
C:New external actions can invalidate earlier assumptions about what an agent is allowed to do; it is justified if a product adds new tool capabilities without revisiting the original threat model, not the case in which a model change is approved in staging, but a canary later discovers changed classifications for real requests.
D:Continuous content change can invalidate a once-successful retrieval architecture; its scope includes cases where a knowledge corpus changes daily but the solution owner reviews quality only once per year, whereas the problem presented is that a model change is approved in staging, but a canary later discovers changed classifications for real requests.