Anthropic CCAR-P (Claude Certified Architect - Professional) Practice Test 8

Question 70

A litigation analysis pipeline handles long, contradictory evidence sets and errors require expensive attorney review. What design decision would best fit these facts?

  1. Run regression evaluations on contracts, tool calls and exception rates before cutover.
  2. Gate upgrades on measurable gains and compatibility risk, not release recency.
  3. Verify approved model availability and the exact platform data path before choosing a provider endpoint.
  4. Test a stronger reasoning model on a curated difficult-case suite before choosing the production tier.
  5. Benchmark a lower-latency Claude model on extraction accuracy before assigning a higher-capability model by default.

Explanation

Correct Answer

D:

Complex synthesis can justify increased inference cost when it substantially reduces legal-review rework. Confirm quality for complex reasoning with representative task cases and deployment constraints; rankings do not prove that quality for complex reasoning works under this deployment.

Incorrect Answers

A:

Consider its actual operating scope: safe migration between model options. The requested assurance is quality for complex reasoning, which calls for an implementation that can test a stronger reasoning model on a curated difficult-case suite before choosing the production tier, not merely for a related improvement.

B:

The alternative could reduce risk associated with evidence-based upgrade decisions, yet the described behavior requires quality for complex reasoning. That outcome depends on the capability to test a stronger reasoning model on a curated difficult-case suite before choosing the production tier.

C:

The proposed measure is aimed at residency-constrained model availability, not quality for complex reasoning. The architecture must instead test a stronger reasoning model on a curated difficult-case suite before choosing the production tier; a separate change is still needed for quality for complex reasoning.

E:

The control would be justified if fit-for-purpose model choice were the limiting factor. In this scenario the decisive target is quality for complex reasoning; success requires an architecture that will test a stronger reasoning model on a curated difficult-case suite before choosing the production tier.

Question 71

A support bot contains easy FAQ questions and occasional complex policy disputes. What should the solution architect prioritize?

  1. Run regression evaluations on contracts, tool calls and exception rates before cutover.
  2. Route by measured task difficulty and escalate ambiguous cases to a more capable model or human.
  3. Verify approved model availability and the exact platform data path before choosing a provider endpoint.
  4. Benchmark a lower-latency Claude model on extraction accuracy before assigning a higher-capability model by default.

Explanation

Correct Answer

B:

Tiered model selection spends extra capacity only where first-pass uncertainty is meaningful. Confirm adaptive routing between model capabilities with representative task cases and deployment constraints; rankings do not prove that adaptive routing between model capabilities works under this deployment.

Incorrect Answers

A:

Its main effect concerns safe migration between model options. The case explicitly hinges on adaptive routing between model capabilities; to resolve it reliably, the implementation has to route by measured task difficulty and escalate ambiguous cases to a more capable model or human, which this choice does not establish.

C:

The proposed action prioritizes residency-constrained model availability. That can coexist with the right solution, but it cannot replace adaptive routing between model capabilities: the application must still route by measured task difficulty and escalate ambiguous cases to a more capable model or human.

D:

The design concern is real but different: fit-for-purpose model choice. It does not impose the specific requirement for adaptive routing between model capabilities; a suitable solution must route by measured task difficulty and escalate ambiguous cases to a more capable model or human and demonstrate the result.

Question 72

A batch document pipeline has no interactive users and thousands of reports can finish overnight. Which design choice best matches the stated constraint?

  1. Optimize for total job economics and throughput rather than choosing solely on streaming latency.
  2. Compare the effective cost per accepted outcome, including retries and manual handling.
  3. Test a stronger reasoning model on a curated difficult-case suite before choosing the production tier.
  4. Verify approved model availability and the exact platform data path before choosing a provider endpoint.

Explanation

Correct Answer

A:

Asynchronous workloads may exploit batching and concurrency controls while tolerating slower individual completions.

Incorrect Answers

B:

That is primarily a technique for cost per successful task. It lacks the explicit measure for throughput economics for noninteractive jobs that is necessary here, namely a system that will optimize for total job economics and throughput rather than choosing solely on streaming latency.

C:

Improving quality for complex reasoning does not imply throughput economics for noninteractive jobs. The selected design must optimize for total job economics and throughput rather than choosing solely on streaming latency, whereas this choice omits that direct protection or verification step.

D:

This recommendation targets residency-constrained model availability at a different point in the workflow. It leaves throughput economics for noninteractive jobs outside the design contract and does not require the team to optimize for total job economics and throughput rather than choosing solely on streaming latency.

Question 73

A vendor demo shows one model winning on generic math tasks, but the project is specialized medical coding. Which architecture adjustment is most appropriate?

  1. Benchmark language-specific intent, refusal and terminology accuracy before committing.
  2. Run regression evaluations on contracts, tool calls and exception rates before cutover.
  3. Evaluate shortlisted models on de-identified domain cases and safety measures.
  4. Verify approved model availability and the exact platform data path before choosing a provider endpoint.

Explanation

Correct Answer

C:

General leaderboards do not establish performance on the application's actual distribution and error costs. Confirm workload-specific model evidence with representative task cases and deployment constraints; rankings do not prove that workload-specific model evidence works under this deployment.

Incorrect Answers

A:

The relevant trade-off is not primarily multilingual production capability. A successful fix needs workload-specific model evidence; otherwise workload-specific model evidence remains unverified even after this change.

B:

The control would be justified if safe migration between model options were the limiting factor. In this scenario the decisive target is workload-specific model evidence; success requires an architecture that will evaluate shortlisted models on de-identified domain cases and safety measures.

D:

This shifts attention toward residency-constrained model availability when the relevant decision is about workload-specific model evidence. Unless the architecture can evaluate shortlisted models on de-identified domain cases and safety measures, the failure described in the stem remains unresolved.

Question 74

A model migration promises lower token cost but changes structured outputs and tool-selection behavior. Which next step is most defensible?

  1. Run regression evaluations on contracts, tool calls and exception rates before cutover.
  2. Test a stronger reasoning model on a curated difficult-case suite before choosing the production tier.
  3. Evaluate shortlisted models on de-identified domain cases and safety measures.
  4. Gate upgrades on measurable gains and compatibility risk, not release recency.

Explanation

Correct Answer

A:

Version or tier changes require application-level compatibility evidence beyond a price comparison. Confirm safe migration between model options with representative task cases and deployment constraints; rankings do not prove that safe migration between model options works under this deployment.

Incorrect Answers

B:

This addresses quality for complex reasoning; however, the problem asks for safe migration between model options. Without the safeguard to run regression evaluations on contracts, tool calls and exception rates before cutover, the original failure mode is still possible.

C:

That choice would be responsive to workload-specific model evidence, but the reported defect is different. It leaves safe migration between model options unproved because there is no commitment to run regression evaluations on contracts, tool calls and exception rates before cutover.

D:

There is a legitimate use case for evidence-based upgrade decisions, but it does not supply the test or enforcement needed for safe migration between model options. The design needs to run regression evaluations on contracts, tool calls and exception rates before cutover.

Question 75

A model performs well offline but at peak traffic its tail latency violates the customer service objective. Which approach should be recommended?

  1. Benchmark a lower-latency Claude model on extraction accuracy before assigning a higher-capability model by default.
  2. Benchmark language-specific intent, refusal and terminology accuracy before committing.
  3. Measure p95/p99 end-to-end latency under realistic load before deciding on its deployment.
  4. Gate upgrades on measurable gains and compatibility risk, not release recency.

Explanation

Correct Answer

C:

Production suitability depends on queueing and integration overhead as well as single-request model speed. Confirm tail-latency based model selection with representative task cases and deployment constraints; rankings do not prove that tail-latency based model selection works under this deployment.

Incorrect Answers

A:

This recommendation targets fit-for-purpose model choice at a different point in the workflow. It leaves tail-latency based model selection outside the design contract and does not require the team to measure p95/p99 end-to-end latency under realistic load before deciding on its deployment.

B:

This is a valid architectural concern when optimizing multilingual production capability. Here it leaves tail-latency based model selection unaddressed, whereas the intended change must measure p95/p99 end-to-end latency under realistic load before deciding on its deployment before the case can be considered resolved.

D:

By itself, this improves evidence-based upgrade decisions, but the case is not asking the architect to optimize that dimension. The requirement is tail-latency based model selection, supported by an approach to measure p95/p99 end-to-end latency under realistic load before deciding on its deployment.

Question 76

A research team wants the newest model immediately, but its current deployment meets all agreed quality and cost thresholds. What is the best production design response?

  1. Measure p95/p99 end-to-end latency under realistic load before deciding on its deployment.
  2. Gate upgrades on measurable gains and compatibility risk, not release recency.
  3. Test a stronger reasoning model on a curated difficult-case suite before choosing the production tier.
  4. Run regression evaluations on contracts, tool calls and exception rates before cutover.

Explanation

Correct Answer

B:

Model freshness alone does not outweigh regression, operational effort or user impact. Confirm evidence-based upgrade decisions with representative task cases and deployment constraints; rankings do not prove that evidence-based upgrade decisions works under this deployment.

Incorrect Answers

A:

This shifts attention toward tail-latency based model selection when the relevant decision is about evidence-based upgrade decisions. Unless the architecture can gate upgrades on measurable gains and compatibility risk, not release recency, the failure described in the stem remains unresolved.

C:

This addresses quality for complex reasoning; however, the problem asks for evidence-based upgrade decisions. Without the safeguard to gate upgrades on measurable gains and compatibility risk, not release recency, the original failure mode is still possible.

D:

It is aimed at a different outcome, safe migration between model options. No evidence is given that it provides evidence-based upgrade decisions; the needed mechanism is to gate upgrades on measurable gains and compatibility risk, not release recency at the appropriate boundary.

Question 77

A sensitive workload has jurisdictional constraints and not every deployment option meets required residency. Which change most directly satisfies this requirement?

  1. Evaluate shortlisted models on de-identified domain cases and safety measures.
  2. Benchmark language-specific intent, refusal and terminology accuracy before committing.
  3. Test a stronger reasoning model on a curated difficult-case suite before choosing the production tier.
  4. Verify approved model availability and the exact platform data path before choosing a provider endpoint.
  5. Benchmark a lower-latency Claude model on extraction accuracy before assigning a higher-capability model by default.

Explanation

Correct Answer

D:

Capability is only one axis; where prompts and outputs are processed can be a hard architectural constraint. Confirm residency-constrained model availability with representative task cases and deployment constraints; rankings do not prove that residency-constrained model availability works under this deployment.

Incorrect Answers

A:

There is a legitimate use case for workload-specific model evidence, but it does not supply the test or enforcement needed for residency-constrained model availability. The design needs to verify approved model availability and the exact platform data path before choosing a provider endpoint.

B:

Consider its actual operating scope: multilingual production capability. The requested assurance is residency-constrained model availability, which calls for an implementation that can verify approved model availability and the exact platform data path before choosing a provider endpoint, not merely for a related improvement.

C:

The alternative could reduce risk associated with quality for complex reasoning, yet the described behavior requires residency-constrained model availability. That outcome depends on the capability to verify approved model availability and the exact platform data path before choosing a provider endpoint.

E:

The relevant trade-off is not primarily fit-for-purpose model choice. A successful fix needs residency-constrained model availability; otherwise residency-constrained model availability remains unverified even after this change.

Question 78

A multilingual call center's chosen model is accurate in English but inconsistent in Urdu and Arabic. What design decision would best fit these facts?

  1. Evaluate shortlisted models on de-identified domain cases and safety measures.
  2. Benchmark language-specific intent, refusal and terminology accuracy before committing.
  3. Verify approved model availability and the exact platform data path before choosing a provider endpoint.
  4. Benchmark a lower-latency Claude model on extraction accuracy before assigning a higher-capability model by default.

Explanation

Correct Answer

B:

Model selection should cover required languages and scripts; aggregate English metrics hide locale gaps. Confirm multilingual production capability with representative task cases and deployment constraints; rankings do not prove that multilingual production capability works under this deployment.

Incorrect Answers

A:

By itself, this improves workload-specific model evidence, but the case is not asking the architect to optimize that dimension. The requirement is multilingual production capability, supported by an approach to benchmark language-specific intent, refusal and terminology accuracy before committing.

C:

The option might improve residency-constrained model availability; that remains insufficient for multilingual production capability. The case is resolved by multilingual production capability, which depends on a design that will benchmark language-specific intent, refusal and terminology accuracy before committing.

D:

The proposed action prioritizes fit-for-purpose model choice. That can coexist with the right solution, but it cannot replace multilingual production capability: the application must still benchmark language-specific intent, refusal and terminology accuracy before committing.

Question 79

A migration proposal uses model token pricing alone and ignores a tripling of human escalations. Which intervention most directly addresses the failure?

  1. Evaluate shortlisted models on de-identified domain cases and safety measures.
  2. Gate upgrades on measurable gains and compatibility risk, not release recency.
  3. Compare the effective cost per accepted outcome, including retries and manual handling.
  4. Run regression evaluations on contracts, tool calls and exception rates before cutover.

Explanation

Correct Answer

C:

Business economics depend on verified completed tasks, not raw API spend in isolation. Confirm cost per successful task with representative task cases and deployment constraints; rankings do not prove that cost per successful task works under this deployment.

Incorrect Answers

A:

It is aimed at a different outcome, workload-specific model evidence. No evidence is given that it provides cost per successful task; the needed mechanism is to compare the effective cost per accepted outcome, including retries and manual handling at the appropriate boundary.

B:

It emphasizes evidence-based upgrade decisions without closing the stated gap in cost per successful task. By contrast, the remedy to compare the effective cost per accepted outcome, including retries and manual handling acts on the condition that makes this scenario fail.

D:

Improving safe migration between model options does not imply cost per successful task. The selected design must compare the effective cost per accepted outcome, including retries and manual handling, whereas this choice omits that direct protection or verification step.

Question 80

An enterprise assistant must maintain the same output format across several departments despite inconsistent employee phrasing. What should the solution architect prioritize?

  1. Specify prohibited inference boundaries and evaluate representative edge cases with hard downstream checks.
  2. Delimit external document text as data, not instructions, and keep tool permissions separate.
  3. Put stable role, policy boundaries and output requirements in a controlled system instruction and versioned template.
  4. Extract shared policy into a versioned prompt component managed by release control.

Explanation

Correct Answer

C:

Higher-level instructions provide reusable behavior; app-side validation still checks the actual response contract.

Incorrect Answers

A:

The proposed measure is aimed at sensitive-inference guardrails, not centralized stable instruction design. The architecture must instead put stable role, policy boundaries and output requirements in a controlled system instruction and versioned template; a separate change is still needed for centralized stable instruction design.

B:

The relevant trade-off is not primarily trust boundaries in retrieved documents. A successful fix needs centralized stable instruction design; otherwise centralized stable instruction design remains unverified even after this change.

D:

The option belongs to the broader work on versioned reusable prompt policy, but it cannot by itself deliver centralized stable instruction design. A decision on this case needs the team to put stable role, policy boundaries and output requirements in a controlled system instruction and versioned template and verify that outcome.

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