Anthropic CCAR-P (Claude Certified Architect - Professional) Practice Test 10

Question 90

A structured labeling task uses unusual internal categories that the model confuses with common English meanings. Which design choice best matches the stated constraint?

  1. Separate development examples from held-out evaluation instances and prevent data leakage.
  2. Use balanced, varied demonstrations and check position sensitivity.
  3. Add a small, diverse set of labeled examples covering borderline distinctions.
  4. Keep a concise zero-shot instruction and verify outcomes on a held-out set.

Explanation

Correct Answer

C:

Few-shot patterns clarify proprietary decision boundaries better than repeating vague category names. Check few-shot category boundary calibration using independent cases and verified intermediate evidence, because demonstrations can otherwise hide brittle assumptions.

Incorrect Answers

A:

The control would be justified if uncontaminated few-shot evaluation were the limiting factor. In this scenario the decisive target is few-shot category boundary calibration; success requires an architecture that will add a small, diverse set of labeled examples covering borderline distinctions.

B:

The option belongs to the broader work on robust example ordering, but it cannot by itself deliver few-shot category boundary calibration. A decision on this case needs the team to add a small, diverse set of labeled examples covering borderline distinctions and verify that outcome.

D:

For a workflow whose primary weakness was efficient zero-shot operation, that proposal could be reasonable. Here the unmet obligation is few-shot category boundary calibration, and that requires the system to add a small, diverse set of labeled examples covering borderline distinctions.

Question 91

A straightforward sentiment task performs well without examples and the team must minimize prompt size. Which architecture adjustment is most appropriate?

  1. Use balanced, varied demonstrations and check position sensitivity.
  2. Separate development examples from held-out evaluation instances and prevent data leakage.
  3. Add a small, diverse set of labeled examples covering borderline distinctions.
  4. Keep a concise zero-shot instruction and verify outcomes on a held-out set.
  5. Remove unnecessary deliberation scaffolding and evaluate the concise prompt on the production workload.

Explanation

Correct Answer

D:

Additional examples consume context and cost without adding value when the rule is already clear. Check efficient zero-shot operation using independent cases and verified intermediate evidence, because demonstrations can otherwise hide brittle assumptions.

Incorrect Answers

A:

This addresses robust example ordering; however, the problem asks for efficient zero-shot operation. Without the safeguard to keep a concise zero-shot instruction and verify outcomes on a held-out set, the original failure mode is still possible.

B:

That choice would be responsive to uncontaminated few-shot evaluation, but the reported defect is different. It leaves efficient zero-shot operation unproved because there is no commitment to keep a concise zero-shot instruction and verify outcomes on a held-out set.

C:

There is a legitimate use case for few-shot category boundary calibration, but it does not supply the test or enforcement needed for efficient zero-shot operation. The design needs to keep a concise zero-shot instruction and verify outcomes on a held-out set.

E:

The alternative could reduce risk associated with minimal reasoning overhead for easy requests, yet the described behavior requires efficient zero-shot operation. That outcome depends on the capability to keep a concise zero-shot instruction and verify outcomes on a held-out set.

Question 92

A request requires reconciling multiple policy exceptions rather than extracting a single fact. Which approach should be recommended?

  1. Require each criterion to be checked against explicit evidence, then aggregate according to the written rule.
  2. Require concise decision justifications and cited evidence instead of hidden chain-of-thought content.
  3. Decompose reasoning into auditable checks with intermediate evidence before composing the conclusion.
  4. Broaden demonstrations to span plausible positives, negatives and ambiguous cases.

Explanation

Correct Answer

C:

Breaking a difficult decision into independently checkable operations reduces hidden assumption errors. Check verifiable multi-step reasoning using independent cases and verified intermediate evidence, because demonstrations can otherwise hide brittle assumptions.

Incorrect Answers

A:

This is a valid architectural concern when optimizing rubric-driven stepwise verification. Here it leaves verifiable multi-step reasoning unaddressed, whereas the intended change must decompose reasoning into auditable checks with intermediate evidence before composing the conclusion before the case can be considered resolved.

B:

This would make sense as an additional project activity for auditable justifications without private traces. The priority required here is verifiable multi-step reasoning, so the workflow still needs a control that will decompose reasoning into auditable checks with intermediate evidence before composing the conclusion.

D:

Its main effect concerns representative demonstration sampling. The case explicitly hinges on verifiable multi-step reasoning; to resolve it reliably, the implementation has to decompose reasoning into auditable checks with intermediate evidence before composing the conclusion, which this choice does not establish.

Question 93

A compliance few-shot evaluation is inflated because held-out cases closely resemble the prompt demonstrations. The same prompt uses only examples of the positive label, causing borderline negatives to be classified as positive. Which TWO improvements target these weaknesses? Choose TWO.

  1. Add more positive-label demonstrations selected from the same cases that will later be used to report test accuracy.
  2. Separate development examples from held-out evaluation instances and prevent data leakage.
  3. Add contrasting negative and boundary examples, then measure false positives on cases excluded from demonstration selection.
  4. Keep a concise zero-shot instruction and verify outcomes on a held-out set.
  5. Repeat borderline examples from the current reported test set until the few-shot prompt achieves the target pass rate.

Explanation

Correct Answers

B:

Development examples must not be reused as final evidence of generalization, even if their wording has been shuffled. A held-out set should be assembled independently, with documented split rules; contrasting negatives are a separate improvement to the training prompt itself.

C:

Balanced counterexamples make the decision boundary explicit instead of teaching the model that most cases are positive. A distinct evaluation slice checks that the corrective examples generalize rather than leak into grading.

Incorrect Answers

A:

More near-identical examples reinforce the skew toward positive predictions, and selecting them from the reported test population makes leakage worse. It neither creates an independent evaluation nor teaches a negative boundary.

D:

It emphasizes efficient zero-shot operation without closing the stated gap in uncontaminated few-shot evaluation and contrasting few-shot examples for borderline false positives. By contrast, the remedy to separate development examples from held-out evaluation instances and prevent data leakage acts on the condition that makes this scenario fail.

E:

Repeated tuning against the supposed holdout invalidates its independence and can overfit to the current benchmark. The goal is to establish a new separation and a genuinely balanced decision boundary.

Question 94

A few-shot classifier performs well on usual tickets but fails whenever the ordering of categories changes. Which change most directly satisfies this requirement?

  1. Add a small, diverse set of labeled examples covering borderline distinctions.
  2. Separate development examples from held-out evaluation instances and prevent data leakage.
  3. Keep a concise zero-shot instruction and verify outcomes on a held-out set.
  4. Use balanced, varied demonstrations and check position sensitivity.
  5. Broaden demonstrations to span plausible positives, negatives and ambiguous cases.
  6. Use a deterministic calculation tool and ask Claude to interpret validated results.

Explanation

Correct Answer

D:

Representative examples prevent the model from treating a specific sequence as the category rule. Check robust example ordering using independent cases and verified intermediate evidence, because demonstrations can otherwise hide brittle assumptions.

Incorrect Answers

A:

Consider its actual operating scope: few-shot category boundary calibration. The requested assurance is robust example ordering, which calls for an implementation that can use balanced, varied demonstrations and check position sensitivity, not merely for a related improvement.

B:

The alternative could reduce risk associated with uncontaminated few-shot evaluation, yet the described behavior requires robust example ordering. That outcome depends on the capability to use balanced, varied demonstrations and check position sensitivity.

C:

The proposed measure is aimed at efficient zero-shot operation, not robust example ordering. The architecture must instead use balanced, varied demonstrations and check position sensitivity; a separate change is still needed for robust example ordering.

E:

The control would be justified if representative demonstration sampling were the limiting factor. In this scenario the decisive target is robust example ordering; success requires an architecture that will use balanced, varied demonstrations and check position sensitivity.

F:

The option belongs to the broader work on external arithmetic verification, but it cannot by itself deliver robust example ordering. A decision on this case needs the team to use balanced, varied demonstrations and check position sensitivity and verify that outcome.

Question 95

A financial report asks for exact arithmetic across thousands of rows. Which solution best resolves this operational gap?

  1. Broaden demonstrations to span plausible positives, negatives and ambiguous cases.
  2. Use a deterministic calculation tool and ask Claude to interpret validated results.
  3. Use balanced, varied demonstrations and check position sensitivity.
  4. Require concise decision justifications and cited evidence instead of hidden chain-of-thought content.

Explanation

Correct Answer

B:

Language-model reasoning is not a substitute for exact aggregation when reproducibility matters. Check external arithmetic verification using independent cases and verified intermediate evidence, because demonstrations can otherwise hide brittle assumptions.

Incorrect Answers

A:

This addresses representative demonstration sampling; however, the problem asks for external arithmetic verification. Without the safeguard to use a deterministic calculation tool and ask Claude to interpret validated results, the original failure mode is still possible.

C:

The proposed action prioritizes robust example ordering. That can coexist with the right solution, but it cannot replace external arithmetic verification: the application must still use a deterministic calculation tool and ask Claude to interpret validated results.

D:

The design concern is real but different: auditable justifications without private traces. It does not impose the specific requirement for external arithmetic verification; a suitable solution must use a deterministic calculation tool and ask Claude to interpret validated results and demonstrate the result.

Question 96

A prompt asks the model to reveal private hidden reasoning as proof of accuracy. What design decision would best fit these facts?

  1. Remove unnecessary deliberation scaffolding and evaluate the concise prompt on the production workload.
  2. Decompose reasoning into auditable checks with intermediate evidence before composing the conclusion.
  3. Require concise decision justifications and cited evidence instead of hidden chain-of-thought content.
  4. Keep a concise zero-shot instruction and verify outcomes on a held-out set.

Explanation

Correct Answer

C:

Observable claims and verifiable calculations support auditability without depending on inaccessible internal reasoning.

Incorrect Answers

A:

It emphasizes minimal reasoning overhead for easy requests without closing the stated gap in auditable justifications without private traces. By contrast, the remedy to require concise decision justifications and cited evidence instead of hidden chain-of-thought content acts on the condition that makes this scenario fail.

B:

That is primarily a technique for verifiable multi-step reasoning. It lacks the explicit measure for auditable justifications without private traces that is necessary here, namely a system that will require concise decision justifications and cited evidence instead of hidden chain-of-thought content.

D:

This recommendation targets efficient zero-shot operation at a different point in the workflow. It leaves auditable justifications without private traces outside the design contract and does not require the team to require concise decision justifications and cited evidence instead of hidden chain-of-thought content.

Question 97

A small ticket-routing model is overly influenced by a single negative example in its template. Which intervention most directly addresses the failure?

  1. Broaden demonstrations to span plausible positives, negatives and ambiguous cases.
  2. Use a deterministic calculation tool and ask Claude to interpret validated results.
  3. Use balanced, varied demonstrations and check position sensitivity.
  4. Separate development examples from held-out evaluation instances and prevent data leakage.

Explanation

Correct Answer

A:

Example diversity reduces accidental overfitting to a narrow wording pattern. Check representative demonstration sampling using independent cases and verified intermediate evidence, because demonstrations can otherwise hide brittle assumptions.

Incorrect Answers

B:

The control would be justified if external arithmetic verification were the limiting factor. In this scenario the decisive target is representative demonstration sampling; success requires an architecture that will broaden demonstrations to span plausible positives, negatives and ambiguous cases.

C:

The option belongs to the broader work on robust example ordering, but it cannot by itself deliver representative demonstration sampling. A decision on this case needs the team to broaden demonstrations to span plausible positives, negatives and ambiguous cases and verify that outcome.

D:

This shifts attention toward uncontaminated few-shot evaluation when the relevant decision is about representative demonstration sampling. Unless the architecture can broaden demonstrations to span plausible positives, negatives and ambiguous cases, the failure described in the stem remains unresolved.

Question 98

A complicated compliance checklist gives inconsistent judgments when asked as one open-ended question. What should the solution architect prioritize?

  1. Decompose reasoning into auditable checks with intermediate evidence before composing the conclusion.
  2. Require each criterion to be checked against explicit evidence, then aggregate according to the written rule.
  3. Require concise decision justifications and cited evidence instead of hidden chain-of-thought content.
  4. Remove unnecessary deliberation scaffolding and evaluate the concise prompt on the production workload.
  5. Use a deterministic calculation tool and ask Claude to interpret validated results.

Explanation

Correct Answer

B:

Criterion-by-criterion verification narrows the space for unsupported intuition in composite decisions. Check rubric-driven stepwise verification using independent cases and verified intermediate evidence, because demonstrations can otherwise hide brittle assumptions.

Incorrect Answers

A:

The design concern is real but different: verifiable multi-step reasoning. It does not impose the specific requirement for rubric-driven stepwise verification; a suitable solution must require each criterion to be checked against explicit evidence, then aggregate according to the written rule and demonstrate the result.

C:

That choice would be responsive to auditable justifications without private traces, but the reported defect is different. It leaves rubric-driven stepwise verification unproved because there is no commitment to require each criterion to be checked against explicit evidence, then aggregate according to the written rule.

D:

There is a legitimate use case for minimal reasoning overhead for easy requests, but it does not supply the test or enforcement needed for rubric-driven stepwise verification. The design needs to require each criterion to be checked against explicit evidence, then aggregate according to the written rule.

E:

Consider its actual operating scope: external arithmetic verification. The requested assurance is rubric-driven stepwise verification, which calls for an implementation that can require each criterion to be checked against explicit evidence, then aggregate according to the written rule, not merely for a related improvement.

Question 99

A customer support answer should be short and direct, but a detailed reasoning prompt doubles latency without benefit. Which architecture adjustment is most appropriate?

  1. Require concise decision justifications and cited evidence instead of hidden chain-of-thought content.
  2. Require each criterion to be checked against explicit evidence, then aggregate according to the written rule.
  3. Keep a concise zero-shot instruction and verify outcomes on a held-out set.
  4. Remove unnecessary deliberation scaffolding and evaluate the concise prompt on the production workload.

Explanation

Correct Answer

D:

Prompt complexity should be justified by task difficulty and observed quality gains. Check minimal reasoning overhead for easy requests using independent cases and verified intermediate evidence, because demonstrations can otherwise hide brittle assumptions.

Incorrect Answers

A:

This recommendation targets auditable justifications without private traces at a different point in the workflow. It leaves minimal reasoning overhead for easy requests outside the design contract and does not require the team to remove unnecessary deliberation scaffolding and evaluate the concise prompt on the production workload.

B:

This addresses rubric-driven stepwise verification; however, the problem asks for minimal reasoning overhead for easy requests. Without the safeguard to remove unnecessary deliberation scaffolding and evaluate the concise prompt on the production workload, the original failure mode is still possible.

C:

This would make sense as an additional project activity for efficient zero-shot operation. The priority required here is minimal reasoning overhead for easy requests, so the workflow still needs a control that will remove unnecessary deliberation scaffolding and evaluate the concise prompt on the production workload.

Question 100

A long customer case exceeds the model's context budget after attaching years of chat history. Which next step is most defensible?

  1. Maintain a structured issue state and expire low-value transcripts under retention policy.
  2. Summarize resolved history into verified state and keep only relevant recent turns and evidence.
  3. Send role-specific task context and compact intermediate artifacts with trace references.
  4. Use section-aware indexing and fetch the relevant clauses plus nearby definitions on demand.

Explanation

Correct Answer

B:

A compact durable state preserves decision-critical facts without repeatedly transmitting stale conversation. Preserve the evidence required for relevant-context compression inside the real context and output budgets rather than relying on raw context volume.

Incorrect Answers

A:

This shifts attention toward structured session memory when the relevant decision is about relevant-context compression. Unless the architecture can summarize resolved history into verified state and keep only relevant recent turns and evidence, the failure described in the stem remains unresolved.

C:

Although minimal delegated context can be valuable, it is not the acceptance condition in the question. The missing provision is relevant-context compression, addressed by the control to summarize resolved history into verified state and keep only relevant recent turns and evidence.

D:

It is aimed at a different outcome, hierarchical retrieval for large documents. No evidence is given that it provides relevant-context compression; the needed mechanism is to summarize resolved history into verified state and keep only relevant recent turns and evidence at the appropriate boundary.

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