Anthropic CCAR-P (Claude Certified Architect - Professional) Practice Test 120

Question 1190

A public-sector casework incident exposes a specific weakness: optimizing only mean accuracy can impose concentrated harm. Which architectural correction addresses the cause rather than masking the symptom?

  1. Enforce Explainable evidence trail
  2. Enforce Fairness regression release gate
  3. Enforce Transparency on assistant limitations
  4. Enforce Subgroup error disparity evaluation

Explanation

Correct Answer

B:

The engineering fix changes the mechanism rather than the presentation: the control layer can block rollout unless declared subgroup safeguards remain satisfied. It requires agreed thresholds and repeatable subgroup tests; a regression test should examine subgroup gate pass rate on each release candidate in cases where optimizing only mean accuracy can impose concentrated harm.

Incorrect Answers

A:

Although the alternative reduces the risk that verbose chain-of-thought-style text is not a substitute for verifiable reasons because the control layer can record relevant policy criteria and source evidence with a clear decision summary; but this remedy operates on another boundary, not the case where optimizing only mean accuracy can impose concentrated harm.

C:

This addresses the failure where a hidden disclaimer elsewhere does not prevent false assumptions in workflow, because the control layer can label automation role, evidence basis and uncertainty at the decision point; yet it would not correct the observed weakness: optimizing only mean accuracy can impose concentrated harm.

D:

This addresses the failure where pooled averages can conceal unequal outcomes, because the control layer can test quality and harm rates separately for relevant populations; but this remedy operates on another boundary, not the case where optimizing only mean accuracy can impose concentrated harm.

Question 1191

The legal operations staging review finds that raw spend alone may rise legitimately with increased workload. Which design intervention is the most appropriate corrective action?

  1. Deploy Tool dependency error budgets
  2. Deploy Semantic failure event logging
  3. Deploy Prompt version rollback analysis
  4. Deploy Cost anomaly alert per outcome

Explanation

Correct Answer

D:

The reported failure calls for the control under which the evaluation process can alert on cost per accepted outcome and decompose input, output and retries. A prompt-only reminder or unrelated metric would not address the risk that raw spend alone may rise legitimately with increased workload. Evidence of resolution requires cost attribution and acceptance-event joining and an improvement in cost anomaly precision normalized by useful completions.

Incorrect Answers

A:

This addresses the failure where overall system uptime can mask a failing required tool, because the evaluation process can track dependency-specific error budgets and user-facing completion; whereas the legal operations defect follows the path in which raw spend alone may rise legitimately with increased workload.

B:

Where fluent responses sometimes omit mandatory customer warnings, the evaluation process can emit categorized validation failures, not only HTTP error codes; yet it would not correct the observed weakness: raw spend alone may rise legitimately with increased workload.

C:

When changing only a model name may leave the faulty prompt in production, the evaluation process can correlate failures to deployed configuration and restore a known-good bundle; but this remedy operates on another boundary, not the case where raw spend alone may rise legitimately with increased workload.

Question 1192

An independent assessment of a research administration Claude system reports that technical accuracy alone does not prove positive organizational value. Which change best closes the exposed control gap?

  1. Prioritize Load-shedding with graceful fallback
  2. Prioritize Capacity forecast from token demand
  3. Prioritize Availability design for downstream dependencies
  4. Prioritize Commercial stage-gate on measurable ROI

Explanation

Correct Answer

D:

This is the causal correction: the architecture can bind rollout gates to accepted outcomes, user adoption and total operating cost. With agreed baseline and financial measurement method in place, the team can test realized net benefit per cohort after deployment and show that the problematic case in which technical accuracy alone does not prove positive organizational value has been addressed.

Incorrect Answers

A:

Where traffic spikes exceed sustainable interactive capacity, the architecture can reject or queue nonessential work while preserving critical requests; yet it would not correct the observed weakness: technical accuracy alone does not prove positive organizational value.

B:

Where requests differ sharply in input length and tool loops, the architecture can forecast using token volume, concurrency and retry behavior rather than call count alone; but this remedy operates on another boundary, not the case where technical accuracy alone does not prove positive organizational value.

C:

Where Claude responses rely on a search index that sometimes fails, the architecture can specify degraded-mode behavior and dependency-aware reliability goals; whereas the research administration defect follows the path in which technical accuracy alone does not prove positive organizational value.

Question 1193

Following a regression in procurement services, the team establishes that raw fan-out results can dominate the remaining model context. Which proposed redesign most directly prevents recurrence?

  1. Enforce Scoped tool result summarization
  2. Enforce Two-stage source navigation
  3. Enforce Context compaction after tool fan-out
  4. Enforce Adaptive retrieval depth

Explanation

Correct Answer

C:

The problem lies in the absence of a mechanism where the integration can aggregate key evidence and preserve original trace references instead of replaying all raw logs. Reinstating it, backed by correlation IDs and an auditable summarization policy, limits the exposure that raw fan-out results can dominate the remaining model context; the engineering team should track provenance-preserving token reduction and decision correctness.

Incorrect Answers

A:

Where a service returns giant JSON objects when only three fields are needed, the integration can trim tool results to task-relevant validated fields before prompting; but this remedy operates on another boundary, not the case where raw fan-out results can dominate the remaining model context.

B:

This addresses the failure where supplying all document bodies creates noise and hides decisive lines, because the integration can locate relevant documents first and open targeted spans afterward; whereas the procurement services defect follows the path in which raw fan-out results can dominate the remaining model context.

D:

When full-depth retrieval on every question increases latency without equal value, the integration can expand retrieval breadth only when answer sufficiency checks fail; but this remedy operates on another boundary, not the case where raw fan-out results can dominate the remaining model context.

Question 1194

A cost-cutting proposal for software platform eliminates Post-incident regression capture. Which undesirable consequence would become most credible after that change?

  1. The risk that changing the system prompt cannot repair invalid connector payloads
  2. The risk that guessing from one screenshot hides true upstream causes
  3. The risk that a one-off patch without tests allows the same bug to return
  4. The risk that debugging in place keeps all users exposed to the broken state
  5. The risk that top-level request durations cannot identify the responsible dependency

Explanation

Correct Answer

C:

Removing the safeguard exposes a failure in which a one-off patch without tests allows the same bug to return. The architecture normally counters this because the engineering workflow can turn the root cause into a permanent automated test and review the rollout path. A compensating design would require incident-to-test workflow and protected release gates and evidence from recurrence rate of formerly resolved incidents.

Incorrect Answers

A:

If end users see failures only when one MCP connector returns malformed results, changing the system prompt cannot repair invalid connector payloads; however, the named safeguard addresses the distinct hazard that a one-off patch without tests allows the same bug to return.

B:

If intermittent agent bugs receive speculative prompt fixes without fixtures, guessing from one screenshot hides true upstream causes; whereas removing Post-incident regression capture primarily exposes the risk that a one-off patch without tests allows the same bug to return.

D:

When a newly enabled hook or agent config breaks developer workflows, debugging in place keeps all users exposed to the broken state; however, the named safeguard addresses the distinct hazard that a one-off patch without tests allows the same bug to return.

E:

When a slow multi-tool run obscures its longest step, top-level request durations cannot identify the responsible dependency; whereas removing Post-incident regression capture primarily exposes the risk that a one-off patch without tests allows the same bug to return.

Question 1195

An enterprise knowledge management audit challenges the need for Budget-aware Skill loading. Which risk would remain unmanaged if its function were not provided elsewhere?

  1. The risk that changing the cached prefix can invalidate reuse and trigger new writes
  2. The risk that shipping all instructions to every request increases interference and token cost
  3. The risk that always-on instructions spend context budget on irrelevant procedures
  4. The risk that copying long procedures into every chat creates drift and context bloat

Explanation

Correct Answer

C:

This is the direct residual hazard: always-on instructions spend context budget on irrelevant procedures. The omitted intervention matters because the model configuration can progressively load only the approved capability needed for the task. A reviewer should demand reliable discovery metadata and permission boundaries and measure loaded capability tokens per successful task before treating the risk as controlled.

Incorrect Answers

A:

If many requests reuse lengthy identical policy instructions, changing the cached prefix can invalidate reuse and trigger new writes; whereas removing Budget-aware Skill loading primarily exposes the risk that always-on instructions spend context budget on irrelevant procedures.

B:

If separate teams use one enormous prompt with conflicting task rules, shipping all instructions to every request increases interference and token cost; but this is not the immediate loss associated with omitting Budget-aware Skill loading: always-on instructions spend context budget on irrelevant procedures.

D:

The other exposure arises when teams repeat complex procedures with files: scripts across projects and copying long procedures into every chat creates drift and context bloat. Omitting the requested control instead leaves the more immediate risk that always-on instructions spend context budget on irrelevant procedures; whereas removing Budget-aware Skill loading primarily exposes the risk that always-on instructions spend context budget on irrelevant procedures.

Question 1196

A telecommunications support rollout must satisfy this design premise: the delivery team can maintain a compatibility review for API, model and governance revisions. Choose TWO linked items that form a workable architecture change and its essential preparation. Choose TWO.

  1. Establish service inventory and explicit change-watch owner
  2. Implement Production feedback-to-backlog loop
  3. Establish baseline evaluation suite and monitoring calendar
  4. Implement Vendor and version change watch
  5. Track containment time during simulated severe regressions

Explanation

Correct Answers

A:

This step creates the enabling basis of service inventory and explicit change-watch owner. That preparation supports the mechanism that the delivery team can maintain a compatibility review for API, model and governance revisions; merely choosing the control does not prove its configuration or authority is ready.

D:

This is the design control: the delivery team can maintain a compatibility review for API, model and governance revisions. It addresses the hazard that assuming a static vendor environment permits silent integration drift; the other selected element supplies the deployment foundation and, where required, the measurement of that control.

Incorrect Answers

B:

This design choice is aimed at cases where defect reports reach operations but not the teams designing prompts and retrieval; it is effective because the delivery team can turn classified incidents into prioritized, testable improvement work. The rollout described is instead governed by the case where model behavior and platform capabilities can change over time and requires a different primary intervention; however, the intended intervention requires that the delivery team can maintain a compatibility review for API, model and governance revisions.

C:

The alternative setup enables a different mechanism: the delivery team can run recurring regression tests with feedback-driven improvements; whereas this intervention requires service inventory and explicit change-watch owner as its operating precondition.

E:

This test would validate a different benefit: the delivery team can provide reversible releases and a tested stop mechanism for side effects; but that indicator does not verify impacted dependencies reviewed before upgrade decisions for the selected architecture.

Question 1197

A supply-chain planning change board is considering removing Fairness regression release gate to simplify a release. Which outcome is the primary residual risk if that protection is absent?

  1. The possibility that more data from the same biased process can reinforce misleading targets
  2. The possibility that verbose chain-of-thought-style text is not a substitute for verifiable reasons
  3. The possibility that pooled averages can conceal unequal outcomes
  4. The possibility that optimizing only mean accuracy can impose concentrated harm

Explanation

Correct Answer

D:

The relevant failure mode is that optimizing only mean accuracy can impose concentrated harm. The existing design can prevent it when the control layer can block rollout unless declared subgroup safeguards remain satisfied. The team needs agreed thresholds and repeatable subgroup tests and must monitor subgroup gate pass rate on each release candidate to support any claim of equivalent protection.

Incorrect Answers

A:

If historical training examples encode legacy operational disparities, more data from the same biased process can reinforce misleading targets; however, the named safeguard addresses the distinct hazard that optimizing only mean accuracy can impose concentrated harm.

B:

The competing hazard is that verbose chain-of-thought-style text is not a substitute for verifiable reasons in a workload where a customer requests reasons for a consequential recommendation; whereas removing Fairness regression release gate primarily exposes the risk that optimizing only mean accuracy can impose concentrated harm.

C:

The competing hazard is that pooled averages can conceal unequal outcomes in a workload where a service appears accurate overall but error rates differ across language groups; but this is not the immediate loss associated with omitting Fairness regression release gate: optimizing only mean accuracy can impose concentrated harm.

Question 1198

The architecture review for manufacturing quality has this operational target: the evaluation process can alert on cost per accepted outcome and decompose input, output and retries. Which THREE choices provide the solution, its enabling condition, and a defensible release measure? Choose THREE.

  1. Implement End-to-end request tracing
  2. Establish cost attribution and acceptance-event joining
  3. Establish task-specific output validators and privacy-safe log structure
  4. Track cost anomaly precision normalized by useful completions
  5. Implement Cost anomaly alert per outcome
  6. Track labeled quality degradation detection delay

Explanation

Correct Answers

B:

The prerequisite is cost attribution and acceptance-event joining. It provides the operating preparation for a system in which the evaluation process can alert on cost per accepted outcome and decompose input, output and retries; without this setup, the planned safeguard has not been demonstrated as deployable.

D:

Production validation should track cost anomaly precision normalized by useful completions. This measures the intended result of a control under which the evaluation process can alert on cost per accepted outcome and decompose input, output and retries, so a good-looking demonstration cannot hide the failure that raw spend alone may rise legitimately with increased workload.

E:

The control itself makes the following capability possible: the evaluation process can alert on cost per accepted outcome and decompose input, output and retries. That matters when raw spend alone may rise legitimately with increased workload; it does not replace the required enabling condition or outcome test.

Incorrect Answers

A:

This design choice is aimed at cases where complaints cite slowness but each component dashboard looks normal; it is effective because the evaluation process can instrument model, retrieval, orchestration and external tool spans together; however, the intended intervention requires that the evaluation process can alert on cost per accepted outcome and decompose input, output and retries.

C:

The alternative setup enables a different mechanism: the evaluation process can emit categorized validation failures, not only HTTP error codes; whereas this intervention requires cost attribution and acceptance-event joining as its operating precondition.

F:

This test would validate a different benefit: the evaluation process can monitor quality proxies and periodically score sampled expert-labeled outputs; but that indicator does not verify cost anomaly precision normalized by useful completions for the selected architecture.

Question 1199

A lean redesign in travel operations would retire Commercial stage-gate on measurable ROI while leaving the rest of the service unchanged. Which problem would remain without the retired capability?

  1. The possibility that requests-per-second estimates can understate compute demand
  2. The possibility that averages conceal the tail experienced by constrained workflows
  3. The possibility that technical accuracy alone does not prove positive organizational value
  4. The possibility that a model SLA cannot compensate for an unavailable required retrieval backend

Explanation

Correct Answer

C:

Removing the safeguard exposes a failure in which technical accuracy alone does not prove positive organizational value. The architecture normally counters this because the architecture can bind rollout gates to accepted outcomes, user adoption and total operating cost. A compensating design would require agreed baseline and financial measurement method and evidence from realized net benefit per cohort after deployment.

Incorrect Answers

A:

When requests differ sharply in input length and tool loops, requests-per-second estimates can understate compute demand; but this is not the immediate loss associated with omitting Commercial stage-gate on measurable ROI: technical accuracy alone does not prove positive organizational value.

B:

If average response time looks fine while urgent users face long delays, averages conceal the tail experienced by constrained workflows; however, the named safeguard addresses the distinct hazard that technical accuracy alone does not prove positive organizational value.

D:

When Claude responses rely on a search index that sometimes fails, a model SLA cannot compensate for an unavailable required retrieval backend; but this is not the immediate loss associated with omitting Commercial stage-gate on measurable ROI: technical accuracy alone does not prove positive organizational value.

Question 1200

An enterprise knowledge management architecture gate expects the following behavior: the integration can aggregate key evidence and preserve original trace references instead of replaying all raw logs. Which TWO actions together supply the design intervention and its operating dependency? Choose TWO.

  1. Implement Context compaction after tool fan-out
  2. Track answer accuracy versus tool-output tokens
  3. Establish searchable catalog with trusted metadata
  4. Establish correlation IDs and an auditable summarization policy
  5. Implement Two-stage source navigation

Explanation

Correct Answers

A:

Implementing the intervention changes the operational mechanism so that the integration can aggregate key evidence and preserve original trace references instead of replaying all raw logs. The failure to prevent is that raw fan-out results can dominate the remaining model context; configuration and validation remain separate release obligations.

D:

The prerequisite is correlation IDs and an auditable summarization policy. It provides the operating preparation for a system in which the integration can aggregate key evidence and preserve original trace references instead of replaying all raw logs; without this setup, the planned safeguard has not been demonstrated as deployable.

Incorrect Answers

B:

This test would validate a different benefit: the integration can trim tool results to task-relevant validated fields before prompting; but that indicator does not verify provenance-preserving token reduction and decision correctness for the selected architecture.

C:

The alternative setup enables a different mechanism: the integration can retrieve only candidate tool schemas after coarse capability discovery; whereas this intervention requires correlation IDs and an auditable summarization policy as its operating precondition.

E:

This design choice is aimed at cases where a repository has many documents but only a few sections matter per request; it is effective because the integration can locate relevant documents first and open targeted spans afterward. The rollout described is instead governed by the case where parallel tool results contain extensive redundant logs and requires a different primary intervention; however, the intended intervention requires that the integration can aggregate key evidence and preserve original trace references instead of replaying all raw logs.

Claude Certified Architect - Professional (CCAR-P) Practice Tests