Anthropic CCAR-P (Claude Certified Architect - Professional) Practice Test 22

Question 210

During solution discovery in procurement services, long-tail failure classes rarely appear in a random convenience sample. Which design response addresses the stated constraint most directly?

  1. Deploy Held-out evaluation split
  2. Deploy Stratified production-like test set
  3. Deploy Automated contract test suite
  4. Deploy Temporal holdout for changing policies

Explanation

Correct Answer

B:

The intervention acts on the system behavior directly: the evaluation process can stratify by task type, risk, language and source complexity. Before rollout, verify privacy-cleared examples and a sampling design. The relevant success test is subgroup coverage and performance against known strata; otherwise a high average score can mask catastrophic minority failures.

Incorrect Answers

A:

When prompt engineers tune repeatedly on the same example corpus, the evaluation process can reserve untouched cases for final comparison after iteration; that does not provide the capability actually needed: the evaluation process can stratify by task type, risk, language and source complexity.

C:

When downstream software expects exact structured fields and allowed enum values, the evaluation process can validate outputs mechanically on every release candidate; whereas the procurement services case must guard against the outcome that a high average score can mask catastrophic minority failures.

D:

When policy retrieval quality looks good using documents from the test period, the evaluation process can test on later source changes to measure freshness and drift response; that does not provide the capability actually needed: the evaluation process can stratify by task type, risk, language and source complexity.

Question 211

The travel operations architect proposes omitting Data readiness assessment from the production design. What is the most direct failure mode that decision would expose?

  1. An exposure in which open-ended satisfaction language cannot determine release readiness
  2. An exposure in which redesigning a guessed process can shift work rather than remove it
  3. An exposure in which a strong model cannot compensate for absent or unauthorized source data
  4. An exposure in which political priority alone may select a low-return high-risk deployment

Explanation

Correct Answer

C:

The loss of the control permits the situation where a strong model cannot compensate for absent or unauthorized source data. Its governing mechanism is that the delivery team can validate quality, permissions, update cadence and ownership before promising RAG results; without an equivalent mechanism, a favorable demonstration cannot establish safety or quality.

Incorrect Answers

A:

If stakeholders ask for a system that is accurate enough without specifying thresholds, open-ended satisfaction language cannot determine release readiness; whereas removing Data readiness assessment primarily exposes the risk that a strong model cannot compensate for absent or unauthorized source data.

B:

The other exposure arises when automation is proposed without mapping present handoffs: failure paths and redesigning a guessed process can shift work rather than remove it. Omitting the requested control instead leaves the more immediate risk that a strong model cannot compensate for absent or unauthorized source data; but this is not the immediate loss associated with omitting Data readiness assessment: a strong model cannot compensate for absent or unauthorized source data.

D:

If teams propose many attractive pilots without common selection criteria, political priority alone may select a low-return high-risk deployment; whereas removing Data readiness assessment primarily exposes the risk that a strong model cannot compensate for absent or unauthorized source data.

Question 212

The public-sector casework program needs a control under which the control layer can treat retrieved text as evidence and prohibit it from changing tool policy. Select TWO components of a credible implementation plan: the design decision and what must exist first. Choose TWO.

  1. Establish layered controls and assigned incident owners
  2. Implement High-risk human approval gate
  3. Implement Untrusted retrieved-content isolation
  4. Track sensitive data leakage rate in response audits
  5. Establish parsing boundary and policy enforcement outside the model

Explanation

Correct Answers

C:

The control itself makes the following capability possible: the control layer can treat retrieved text as evidence and prohibit it from changing tool policy. That matters when instruction blending permits external authors to steer agent behavior; it does not replace the required enabling condition or outcome test.

E:

This step creates the enabling basis of parsing boundary and policy enforcement outside the model. That preparation supports the mechanism that the control layer can treat retrieved text as evidence and prohibit it from changing tool policy; merely choosing the control does not prove its configuration or authority is ready.

Incorrect Answers

A:

The alternative setup enables a different mechanism: the control layer can combine prevention, detection, containment and audited recovery; whereas this intervention requires parsing boundary and policy enforcement outside the model as its operating precondition.

B:

This design choice is aimed at cases where proposed external transfers could produce irreversible customer harm; it is effective because the control layer can stop before execution and require authorized human confirmation; however, the intended intervention requires that the control layer can treat retrieved text as evidence and prohibit it from changing tool policy.

D:

This test would validate a different benefit: the control layer can enforce recipient-specific redaction and validated output fields; but that indicator does not verify policy override rate in hostile-document evaluations for the selected architecture.

Question 213

An enterprise IT service owner is reviewing Per-class model policy after a canary rollout. The critical regression would occur if a single default model policy may violate different governance requirements. What observable acceptance signal belongs in the dashboard?

  1. Track relevant-fact recall by document position and input length
  2. Track routed quality at the target cost and p95 latency
  3. Track audited synthesis accuracy on difficult cases
  4. Track safe fallback completion rate under simulated outages
  5. Track policy-compliant invocation rate by workload class

Explanation

Correct Answer

E:

The correct signal is policy-compliant invocation rate by workload class; it must be computed for representative workloads and compared with a baseline. It checks whether the model configuration can set explicit allowed model configurations by data class and purpose while surfacing the exposure that a single default model policy may violate different governance requirements.

Incorrect Answers

A:

When legal analysis requests require unusually long source packets, the model configuration can test effective context utilization and recall before choosing a model configuration; however, that proxy would not demonstrate policy-compliant invocation rate by workload class for the selected intervention.

B:

The alternative indicator checks whether a fast adequate tier handles easy cases and escalates only uncertain ones when most messages are short and low-complexity while a minority need deep analysis; but the acceptance gate must instead inspect policy-compliant invocation rate by workload class to detect the targeted failure.

C:

The proposed measure would suit a setting where multi-document arbitration requires several interdependent inferences, since a stronger model can justify tradeoffs across heterogeneous evidence; by contrast, the enterprise IT release needs evidence from policy-compliant invocation rate by workload class.

D:

The proposed measure would suit a setting where primary inference service can time out during a critical support window, since the model configuration can route to an approved fallback model with compatible output contracts; however, that proxy would not demonstrate policy-compliant invocation rate by workload class for the selected intervention.

Question 214

For a healthcare operations rollout, the proposed design uses Server-side tool authorization, relying on the claim that the executor evaluates policy before invoking protected downstream APIs. What enabling control must be in place to make that design enforceable?

  1. Prepare propagated identity and resource-level access policy
  2. Prepare supported authorization flow and secure token storage
  3. Prepare trusted auth context and policy engine
  4. Prepare authoritative tenant IDs and indexing ACL metadata

Explanation

Correct Answer

C:

What makes the control operable is trusted auth context and policy engine. It supplies the precondition for the mechanism whereby the executor evaluates policy before invoking protected downstream APIs. The residual failure to guard against is that prompt rules alone cannot serve as a security boundary, measured through blocked unauthorized operations in adversarial tool-call tests.

Incorrect Answers

A:

For the separate challenge where a shared integration must reflect each employee's actual record rights, the integration can enforce each call using the requesting user's authorized scope; however, the named intervention needs trusted auth context and policy engine before its protection can be enforced.

B:

For the separate challenge where employees connect personal enterprise resources through an integration, the integration can obtain user consent for minimum scopes and handle revocation; but the production dependency for Server-side tool authorization is trusted auth context and policy engine.

D:

For the separate challenge where one vector index holds embeddings for several client organizations, the integration can apply tenant and document ACL filters before results reach the model; however, the named intervention needs trusted auth context and policy engine before its protection can be enforced.

Question 215

The solution design for field-service coordination assumes that schema checks and normalization reject malformed inputs before they enter the model context. Before treating Validated input contract as production-ready, which dependency needs an owner and verification?

  1. Arrange an agreed output schema and invalid-output handler
  2. Arrange stable request identifiers and a deduplication store
  3. Arrange an explicit schema and ownership of input validation
  4. Arrange agreed field-necessity rules and masking tests
  5. Arrange stage IDs, persistence design and replay policy

Explanation

Correct Answer

C:

Readiness depends on an explicit schema and ownership of input validation, not just a design approval. That condition supports the mechanism in which schema checks and normalization reject malformed inputs before they enter the model context. Without it, unvalidated inputs propagate ambiguity into generated actions; monitoring invalid-input rate and successful recovery by input category helps verify the deployed result.

Incorrect Answers

A:

For the separate challenge where downstream software requires machine-readable dispositions with fixed enums, the architecture can validate generated payloads against a contract and route failures to repair or review; but the production dependency for Validated input contract is an explicit schema and ownership of input validation.

B:

For the separate challenge where a retrying agent sometimes creates the same support task twice, idempotency keys and deduplication protect side-effecting writes across retries; for this release, the team must first verify an explicit schema and ownership of input validation.

D:

Where raw customer messages contain identifiers irrelevant to classification, preprocessing redacts unnecessary fields while preserving task-bearing content; but the production dependency for Validated input contract is an explicit schema and ownership of input validation.

E:

Where multi-step reviews lose progress during retries and restarts, durable state records completed stages and enables safe continuation; for this release, the team must first verify an explicit schema and ownership of input validation.

Question 216

A program architect at telecommunications support has endorsed Stratified production-like test set. The proposed technical effect is that the evaluation process can stratify by task type, risk, language and source complexity. Which implementation condition is necessary to realize that effect?

  1. Arrange dated corpora and known change events
  2. Arrange leakage-resistant dataset versioning
  3. Arrange privacy-cleared examples and a sampling design
  4. Arrange threat model and approved red-team payloads

Explanation

Correct Answer

C:

The required preparation is privacy-cleared examples and a sampling design. It makes the following architecture claim enforceable: the evaluation process can stratify by task type, risk, language and source complexity. If that setup is skipped, a high average score can mask catastrophic minority failures, so the release should examine subgroup coverage and performance against known strata.

Incorrect Answers

A:

Where policy retrieval quality looks good using documents from the test period, the evaluation process can test on later source changes to measure freshness and drift response; for this release, the team must first verify privacy-cleared examples and a sampling design.

B:

For the separate challenge where prompt engineers tune repeatedly on the same example corpus, the evaluation process can reserve untouched cases for final comparison after iteration; however, the named intervention needs privacy-cleared examples and a sampling design before its protection can be enforced.

D:

This prerequisite supports a method used when a tool agent ingests external documents that can contain hostile instructions, because the evaluation process can add attacker-controlled content and escalation attempts to evaluation; for this release, the team must first verify privacy-cleared examples and a sampling design.

Question 217

The defining problem for a supply-chain planning pilot is that stakeholders ask for a system that is accurate enough without specifying thresholds. Which proposed approach addresses it most precisely?

  1. Introduce Current-state service blueprint
  2. Introduce Stakeholder and decision-rights mapping
  3. Introduce Acceptance criteria negotiation
  4. Introduce Use-case prioritization workshop

Explanation

Correct Answer

C:

The architectural distinction is that the delivery team can turn goals into testable task-specific, risk-sensitive success criteria. The decision should be implemented with baseline metrics and defined evaluation sample and evaluated against approved acceptance conditions with measurable targets; otherwise, open-ended satisfaction language cannot determine release readiness.

Incorrect Answers

A:

For a workload where automation is proposed without mapping present handoffs and failure paths, the delivery team can chart the existing input-to-outcome process and exception volumes; the stated requirement remains unmet: the delivery team can turn goals into testable task-specific, risk-sensitive success criteria.

B:

When product, security and legal teams disagree on who can approve release, the delivery team can define accountable decision owners and consultation rights early; whereas the supply-chain planning case must guard against the outcome that open-ended satisfaction language cannot determine release readiness.

D:

If teams propose many attractive pilots without common selection criteria, the delivery team can jointly rank candidates by value, feasibility, harm and data readiness; the stated requirement remains unmet: the delivery team can turn goals into testable task-specific, risk-sensitive success criteria.

Question 218

A manufacturing quality team is selecting the next improvement for its AI service; failures may bypass one input or model safety layer. What design choice has the strongest direct justification?

  1. Adopt Defense-in-depth incident response
  2. Adopt High-risk human approval gate
  3. Adopt Tool-execution policy gate
  4. Adopt Output privacy filter
  5. Adopt Rate and spend limits by tenant

Explanation

Correct Answer

A:

Under the stated constraints, the control layer can combine prevention, detection, containment and audited recovery. That directly addresses the operational failure without relying on a generic prompt reminder. Production depends on layered controls and assigned incident owners; the result to monitor is attack containment time and residual failure severity.

Incorrect Answers

B:

If proposed external transfers could produce irreversible customer harm, the control layer can stop before execution and require authorized human confirmation; the alternative leaves the central requirement unproven: the control layer can combine prevention, detection, containment and audited recovery.

C:

If assistants propose privileged write operations in customer systems, the control layer can require server-side authorization and business-policy checks before each side effect; it does not ensure the required behavior: the control layer can combine prevention, detection, containment and audited recovery.

D:

When generated summaries may echo sensitive identifiers unnecessary for recipients, the control layer can enforce recipient-specific redaction and validated output fields; whereas the manufacturing quality case must guard against the outcome that a single classifier cannot handle every emergent attack pattern.

E:

If one customer can monopolize shared model capacity through abusive retries, the control layer can enforce bounded usage budgets and anomaly controls per tenant; the alternative leaves the central requirement unproven: the control layer can combine prevention, detection, containment and audited recovery.

Question 219

After installing Server-side tool authorization in an energy operations workflow, the team must validate whether it has reduced the exposure that prompt rules alone cannot serve as a security boundary. Which measurement is most relevant?

  1. Track blocked unauthorized operations in adversarial tool-call tests
  2. Track forged and replayed webhook rejection rate
  3. Track revoked-token access denial and scope mismatch rate
  4. Track cross-tenant retrieval violations in isolation tests

Explanation

Correct Answer

A:

The correct signal is blocked unauthorized operations in adversarial tool-call tests; it must be computed for representative workloads and compared with a baseline. It checks whether the executor evaluates policy before invoking protected downstream APIs while surfacing the exposure that prompt rules alone cannot serve as a security boundary.

Incorrect Answers

B:

The alternative indicator checks whether the integration can verify message authenticity, timestamp and replay protection before accepting commands when an upstream application posts actions to a model-driven workflow; but the acceptance gate must instead inspect blocked unauthorized operations in adversarial tool-call tests to detect the targeted failure.

C:

The proposed measure would suit a setting where employees connect personal enterprise resources through an integration, since the integration can obtain user consent for minimum scopes and handle revocation; by contrast, the energy operations release needs evidence from blocked unauthorized operations in adversarial tool-call tests.

D:

Where one vector index holds embeddings for several client organizations, a relevant check would establish whether the integration can apply tenant and document ACL filters before results reach the model; however, that proxy would not demonstrate blocked unauthorized operations in adversarial tool-call tests for the selected intervention.

Question 220

Following a regression in human-resources services, the team establishes that a single default model policy may violate different governance requirements. Which proposed redesign most directly prevents recurrence?

  1. Prioritize Per-class model policy
  2. Prioritize Reasoning-capable model for complex synthesis
  3. Prioritize Empirical model bake-off
  4. Prioritize Context-capacity fit before model selection
  5. Prioritize Small-model first-pass routing
  6. Prioritize Fallback model under failure

Explanation

Correct Answer

A:

The remediation acts at the relevant design boundary: the model configuration can set explicit allowed model configurations by data class and purpose. That counteracts the observed hazard that a single default model policy may violate different governance requirements. Rollout should include approved data classification and deployment policy, with regression evidence from policy-compliant invocation rate by workload class.

Incorrect Answers

B:

Where multi-document arbitration requires several interdependent inferences, a stronger model can justify tradeoffs across heterogeneous evidence; but this remedy operates on another boundary, not the case where a single default model policy may violate different governance requirements.

C:

When generic leaderboard scores may not predict domain performance, the model configuration can evaluate candidate models on identical task data and operating constraints; whereas the human-resources services defect follows the path in which a single default model policy may violate different governance requirements.

D:

Although the alternative reduces the risk that nominal context limits alone do not prove reliable use of every passage because the model configuration can test effective context utilization and recall before choosing a model configuration; yet it would not correct the observed weakness: a single default model policy may violate different governance requirements.

E:

This addresses the failure where always using the strongest model increases cost and wait time, because a fast adequate tier handles easy cases and escalates only uncertain ones; but this remedy operates on another boundary, not the case where a single default model policy may violate different governance requirements.

F:

The alternative reduces the risk that failover to an untested model can change schema and decision quality because the model configuration can route to an approved fallback model with compatible output contracts; however, the human-resources services defect follows the path in which a single default model policy may violate different governance requirements.

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