Anthropic CCAR-P (Claude Certified Architect - Professional) Practice Test 39

Question 380

A data-flow diagram lists services but fails to assign controllers and processors to sensitive data categories. Which observed test outcome would justify sign-off?

  1. A reproducible run shows that deletion validation confirms all in-scope copies follow the approved retention and removal process.
  2. The signed test record confirms that a signed control assessment confirms access, audit and contractual requirements for the use case.
  3. The sign-off record demonstrates that a responsibility matrix assigns accountable parties and tested controls for every requirement.
  4. The release check confirms that the control inventory maps sensitive fields to authorized processors and retention periods.

Explanation

Correct Answer

D:

Compliance architecture needs accountable owners and a traceable basis for each processing activity; this checks the situation in which an AI solution processes customer records without showing which team controls each data category.

Incorrect Answers

A:

Derived indexes and logs can preserve personal data after the source record is deleted; this applies when a deletion request removes user records but leaves embeddings and debug traces behind, but here a data-flow diagram lists services but fails to assign controllers and processors to sensitive data categories.

B:

Encryption helps but does not independently establish HIPAA obligations or vendor contractual eligibility; the alternative helps if a healthcare application handles protected health information and assumes 'encrypted' means HIPAA-ready, instead of when a data-flow diagram lists services but fails to assign controllers and processors to sensitive data categories.

C:

Managed hosting does not eliminate obligations for how an application accesses and uses customer data; it is justified if a team assumes model hosting in its cloud account transfers all regulatory duties to the platform, not the case in which a data-flow diagram lists services but fails to assign controllers and processors to sensitive data categories.

Question 381

A buyer assumes the vendor's certification removes the application's own need for policies and evidence. Which sign-off finding best checks the actual boundary?

  1. The release report verifies that the control inventory maps sensitive fields to authorized processors and retention periods.
  2. The controlled replay verifies that procurement evidence maps the selected deployment to the required authorized boundary.
  3. A reproducible run shows that end-to-end flow diagrams show approved locations for every personal-data copy.
  4. The signed test record confirms that a responsibility matrix assigns accountable parties and tested controls for every requirement.

Explanation

Correct Answer

D:

Managed hosting does not eliminate obligations for how an application accesses and uses customer data; the check directly targets the condition that a team assumes model hosting in its cloud account transfers all regulatory duties to the platform.

Incorrect Answers

A:

Compliance architecture needs accountable owners and a traceable basis for each processing activity; the other case is when an AI solution processes customer records without showing which team controls each data category, not the present case where a buyer assumes the vendor's certification removes the application's own need for policies and evidence.

B:

A provider's unrelated certification cannot be extended to every product or region by assumption; this makes sense when a federal workflow needs FedRAMP-authorized services at a defined impact level, while this problem involves a buyer assumes the vendor's certification removes the application's own need for policies and evidence.

C:

Data-location obligations cover the complete flow, not just the primary database; this technique fits where a client requires data to stay in its approved jurisdiction during processing and logging, and not where a buyer assumes the vendor's certification removes the application's own need for policies and evidence.

Question 382

A hiring screener has high overall agreement with reviewers but rejects one demographic subgroup more often. Which option supplies the needed architectural boundary?

  1. As a scoped first step, review training and evaluation labels for historical inequities before optimizing agreement.
  2. In the release design, provide transparent decision notices and a documented review or contest pathway.
  3. During architecture review, evaluate decision error rates by relevant groups with legal and privacy safeguards.
  4. In the proposed implementation, create representative language and dialect evaluation sets reviewed by local experts.

Explanation

Correct Answer

C:

Aggregate performance can conceal disparate impact and different false-negative rates; the solution is testable when stratified analysis reports performance gaps and confidence intervals to governance reviewers.

Incorrect Answers

A:

Replicating past patterns can reinforce rather than remediate unfair outcomes; that is defensible if a policy assistant treats historical decisions as infallible labels although earlier practices may have been biased, whereas the current issue is that a hiring screener has high overall agreement with reviewers but rejects one demographic subgroup more often.

B:

Meaningful transparency includes recourse when automated assistance affects important outcomes; the other workload is one where an automated recommendation influences access to a service without showing users how to appeal, whereas the release criterion addresses a hiring screener has high overall agreement with reviewers but rejects one demographic subgroup more often.

D:

Different linguistic groups may experience disproportionate factual or safety failures; the alternative helps if a multilingual service is tested only on standard English despite serving several communities, not the risk where a hiring screener has high overall agreement with reviewers but rejects one demographic subgroup more often.

Question 383

An automated recommendation influences access to a service without showing users how to appeal. Which action addresses the root concern without unnecessary scope?

  1. For the operating team, explain uncertainty, known limitations, human accountability and evidence provenance.
  2. For this deployment, avoid unnecessary sensitive-attribute inference and use governed, consent-appropriate fairness analysis.
  3. As a scoped first step, audit grader agreement across styles and calibrate against diverse human judgments.
  4. In the release design, create representative language and dialect evaluation sets reviewed by local experts.
  5. During architecture review, provide transparent decision notices and a documented review or contest pathway.

Explanation

Correct Answer

E:

Meaningful transparency includes recourse when automated assistance affects important outcomes; a representative trial should prove that affected users can request review and receive an accountable human decision.

Incorrect Answers

A:

Model outputs reflect data and design choices and should not be represented as inherently neutral; it is relevant if a public agency describes AI recommendations as guaranteed objective, instead of when an automated recommendation influences access to a service without showing users how to appeal.

B:

Guessing protected status can introduce new errors and privacy risk without fixing system bias; it serves conditions where a team asks Claude to infer protected traits from names to 'improve fairness', not the case in which an automated recommendation influences access to a service without showing users how to appeal.

C:

Measurement bias can create apparent performance gaps or mask actual ones; the other case is when a model-based evaluation grader consistently prefers a particular writing style used by one user group, whereas the problem presented is that an automated recommendation influences access to a service without showing users how to appeal.

D:

Different linguistic groups may experience disproportionate factual or safety failures; this makes sense when a multilingual service is tested only on standard English despite serving several communities, but this decision concerns an automated recommendation influences access to a service without showing users how to appeal.

Question 384

A multilingual service is tested only on standard English despite serving several communities. What is the most appropriate initial architecture decision?

  1. As the design priority, avoid unnecessary sensitive-attribute inference and use governed, consent-appropriate fairness analysis.
  2. In the implementation sequence, create representative language and dialect evaluation sets reviewed by local experts.
  3. For the operating team, review training and evaluation labels for historical inequities before optimizing agreement.
  4. For this deployment, provide transparent decision notices and a documented review or contest pathway.

Explanation

Correct Answer

B:

Different linguistic groups may experience disproportionate factual or safety failures; reviewers would verify that each supported community has calibrated acceptance metrics rather than one pooled score.

Incorrect Answers

A:

Guessing protected status can introduce new errors and privacy risk without fixing system bias; that would be useful if a team asks Claude to infer protected traits from names to 'improve fairness', while this problem involves a multilingual service is tested only on standard English despite serving several communities.

C:

Replicating past patterns can reinforce rather than remediate unfair outcomes; that is defensible if a policy assistant treats historical decisions as infallible labels although earlier practices may have been biased, rather than the risk that a multilingual service is tested only on standard English despite serving several communities.

D:

Meaningful transparency includes recourse when automated assistance affects important outcomes; the other workload is one where an automated recommendation influences access to a service without showing users how to appeal, rather than the condition that a multilingual service is tested only on standard English despite serving several communities.

Question 385

A team asks Claude to infer protected traits from names to 'improve fairness'. Which architecture change should be approved first?

  1. For the current integration, create representative language and dialect evaluation sets reviewed by local experts.
  2. Under the system plan, evaluate decision error rates by relevant groups with legal and privacy safeguards.
  3. As the design priority, audit grader agreement across styles and calibrate against diverse human judgments.
  4. In the implementation sequence, avoid unnecessary sensitive-attribute inference and use governed, consent-appropriate fairness analysis.

Explanation

Correct Answer

D:

Guessing protected status can introduce new errors and privacy risk without fixing system bias; a meaningful test result is that fairness evaluation relies on approved data practices and measures outcomes rather than name guesses.

Incorrect Answers

A:

Different linguistic groups may experience disproportionate factual or safety failures; the approach helps when a multilingual service is tested only on standard English despite serving several communities, whereas the release criterion addresses a team asks Claude to infer protected traits from names to 'improve fairness'.

B:

Aggregate performance can conceal disparate impact and different false-negative rates; that belongs to situations where a hiring screener has high overall agreement with reviewers but rejects one demographic subgroup more often, while the required safeguard concerns a team asks Claude to infer protected traits from names to 'improve fairness'.

C:

Measurement bias can create apparent performance gaps or mask actual ones; it is relevant if a model-based evaluation grader consistently prefers a particular writing style used by one user group, not the risk where a team asks Claude to infer protected traits from names to 'improve fairness'.

Question 386

A model-based evaluation grader consistently prefers a particular writing style used by one user group. What should be incorporated into the approved design?

  1. Within the rollout design, review training and evaluation labels for historical inequities before optimizing agreement.
  2. For the next architecture revision, audit grader agreement across styles and calibrate against diverse human judgments.
  3. At the system boundary, create representative language and dialect evaluation sets reviewed by local experts.
  4. Under the system plan, provide transparent decision notices and a documented review or contest pathway.

Explanation

Correct Answer

B:

Measurement bias can create apparent performance gaps or mask actual ones; the intended outcome is that blinded expert comparisons expose style-driven grading differences.

Incorrect Answers

A:

Replicating past patterns can reinforce rather than remediate unfair outcomes; that fits cases where a policy assistant treats historical decisions as infallible labels although earlier practices may have been biased, not the case in which a model-based evaluation grader consistently prefers a particular writing style used by one user group.

C:

Different linguistic groups may experience disproportionate factual or safety failures; that would be useful if a multilingual service is tested only on standard English despite serving several communities, but this decision concerns a model-based evaluation grader consistently prefers a particular writing style used by one user group.

D:

Meaningful transparency includes recourse when automated assistance affects important outcomes; it serves a setting where an automated recommendation influences access to a service without showing users how to appeal, but the actual risk is that a model-based evaluation grader consistently prefers a particular writing style used by one user group.

Question 387

A public agency describes AI recommendations as guaranteed objective. Which approach should the design team choose?

  1. For the affected service, avoid unnecessary sensitive-attribute inference and use governed, consent-appropriate fairness analysis.
  2. In the project scope, explain uncertainty, known limitations, human accountability and evidence provenance.
  3. Within the rollout design, create representative language and dialect evaluation sets reviewed by local experts.
  4. Within the delivery plan, audit grader agreement across styles and calibrate against diverse human judgments.

Explanation

Correct Answer

B:

Model outputs reflect data and design choices and should not be represented as inherently neutral; its effect can be measured when user materials identify model involvement, limitations and a named review owner.

Incorrect Answers

A:

Guessing protected status can introduce new errors and privacy risk without fixing system bias; its scope includes cases where a team asks Claude to infer protected traits from names to 'improve fairness', and not where a public agency describes AI recommendations as guaranteed objective.

C:

Different linguistic groups may experience disproportionate factual or safety failures; the approach helps when a multilingual service is tested only on standard English despite serving several communities, rather than the condition that a public agency describes AI recommendations as guaranteed objective.

D:

Measurement bias can create apparent performance gaps or mask actual ones; that belongs to situations where a model-based evaluation grader consistently prefers a particular writing style used by one user group, while the immediate question involves a public agency describes AI recommendations as guaranteed objective.

Question 388

A policy assistant treats historical decisions as infallible labels although earlier practices may have been biased. Where should the solution architect focus the implementation?

  1. For the primary workstream, avoid unnecessary sensitive-attribute inference and use governed, consent-appropriate fairness analysis.
  2. For this workflow, create representative language and dialect evaluation sets reviewed by local experts.
  3. For the affected service, review training and evaluation labels for historical inequities before optimizing agreement.
  4. For this production path, explain uncertainty, known limitations, human accountability and evidence provenance.
  5. Within the rollout design, evaluate decision error rates by relevant groups with legal and privacy safeguards.

Explanation

Correct Answer

C:

Replicating past patterns can reinforce rather than remediate unfair outcomes; a meaningful test result is that an adjudicated audit set separates valid policy criteria from questionable precedent.

Incorrect Answers

A:

Guessing protected status can introduce new errors and privacy risk without fixing system bias; the trigger would be that a team asks Claude to infer protected traits from names to 'improve fairness', while the required safeguard concerns a policy assistant treats historical decisions as infallible labels although earlier practices may have been biased.

B:

Different linguistic groups may experience disproportionate factual or safety failures; the design helps when a multilingual service is tested only on standard English despite serving several communities, not the risk where a policy assistant treats historical decisions as infallible labels although earlier practices may have been biased.

D:

Model outputs reflect data and design choices and should not be represented as inherently neutral; this control is needed when a public agency describes AI recommendations as guaranteed objective, whereas this scenario involves a policy assistant treats historical decisions as infallible labels although earlier practices may have been biased.

E:

Aggregate performance can conceal disparate impact and different false-negative rates; that would be useful if a hiring screener has high overall agreement with reviewers but rejects one demographic subgroup more often, but here a policy assistant treats historical decisions as infallible labels although earlier practices may have been biased.

Question 389

A candidate-ranking pilot's average score looks good until reviewers analyze error rates by population segment. What result discriminates a working safeguard from a demo?

  1. The audit sample demonstrates that stratified analysis reports performance gaps and confidence intervals to governance reviewers.
  2. The readiness assessment shows that blinded expert comparisons expose style-driven grading differences.
  3. The independent test establishes that affected users can request review and receive an accountable human decision.
  4. The verified result is that an adjudicated audit set separates valid policy criteria from questionable precedent.

Explanation

Correct Answer

A:

Aggregate performance can conceal disparate impact and different false-negative rates; this checks the situation in which a hiring screener has high overall agreement with reviewers but rejects one demographic subgroup more often.

Incorrect Answers

B:

Measurement bias can create apparent performance gaps or mask actual ones; it is justified if a model-based evaluation grader consistently prefers a particular writing style used by one user group, but this decision concerns a candidate-ranking pilot's average score looks good until reviewers analyze error rates by population segment.

C:

Meaningful transparency includes recourse when automated assistance affects important outcomes; its scope includes cases where an automated recommendation influences access to a service without showing users how to appeal, but the actual risk is that a candidate-ranking pilot's average score looks good until reviewers analyze error rates by population segment.

D:

Replicating past patterns can reinforce rather than remediate unfair outcomes; it targets cases where a policy assistant treats historical decisions as infallible labels although earlier practices may have been biased, but the requested boundary concerns a candidate-ranking pilot's average score looks good until reviewers analyze error rates by population segment.

Question 390

A citizen disputes a rejected application but the assistant provides no accountable review route. What outcome must the reviewer see before proceeding?

  1. The comparison run confirms that stratified analysis reports performance gaps and confidence intervals to governance reviewers.
  2. The staged rollout demonstrates that blinded expert comparisons expose style-driven grading differences.
  3. The audit sample demonstrates that affected users can request review and receive an accountable human decision.
  4. The readiness assessment shows that each supported community has calibrated acceptance metrics rather than one pooled score.
  5. The independent test establishes that user materials identify model involvement, limitations and a named review owner.

Explanation

Correct Answer

C:

Meaningful transparency includes recourse when automated assistance affects important outcomes; it tests the operating case in which an automated recommendation influences access to a service without showing users how to appeal.

Incorrect Answers

A:

Aggregate performance can conceal disparate impact and different false-negative rates; this makes sense when a hiring screener has high overall agreement with reviewers but rejects one demographic subgroup more often, rather than the risk that a citizen disputes a rejected application but the assistant provides no accountable review route.

B:

Measurement bias can create apparent performance gaps or mask actual ones; this technique fits where a model-based evaluation grader consistently prefers a particular writing style used by one user group, rather than the condition that a citizen disputes a rejected application but the assistant provides no accountable review route.

D:

Different linguistic groups may experience disproportionate factual or safety failures; the design helps when a multilingual service is tested only on standard English despite serving several communities, but the present constraint is that a citizen disputes a rejected application but the assistant provides no accountable review route.

E:

Model outputs reflect data and design choices and should not be represented as inherently neutral; that fits cases where a public agency describes AI recommendations as guaranteed objective, whereas the current issue is that a citizen disputes a rejected application but the assistant provides no accountable review route.

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