Anthropic CCAR-P (Claude Certified Architect - Professional) Practice Test 6

Question 50

Two specialist agents disagree on a regulatory finding and the synthesizer silently chooses one answer. Which measurement is the most relevant release criterion?

  1. The verified result is that a replay after a worker crash resumes pending tasks without repeating confirmed external writes.
  2. The recorded test establishes that a workflow benchmark matches output quality with fewer handoffs and lower token usage.
  3. The representative case test finds that conflicting worker findings are surfaced with citations and a recorded coordinator decision.
  4. The negative-case evaluation confirms that simulated uncertainty terminates within budget and records an actionable escalation.

Explanation

Correct Answer

C:

Clear ownership allows parallel expertise while a reconciliation step prevents contradictory final recommendations; it tests the operating case in which a research project has independent legal, market and security workstreams that need reconciliation.

Incorrect Answers

A:

Durable checkpoints support retries while protecting write operations that already succeeded; this makes sense when a distributed agent run fails halfway through and supervisors cannot resume without duplicate actions, rather than the risk that two specialist agents disagree on a regulatory finding and the synthesizer silently chooses one answer.

B:

Agents add coordination cost when responsibilities are not truly independent or context-separated; this technique fits where an enterprise solution has two workers whose outputs are always used sequentially and never overlap, rather than the condition that two specialist agents disagree on a regulatory finding and the synthesizer silently chooses one answer.

D:

Bounded delegation prevents runaway coordination cost and makes operational failure predictable; the design helps when an orchestrator delegates tasks indefinitely when subordinate agents return uncertain results, but the present constraint is that two specialist agents disagree on a regulatory finding and the synthesizer silently chooses one answer.

Question 51

A concurrency test lets two assistants submit separate updates to the same accounting entry. Which production-like test finding is most convincing?

  1. The readiness assessment shows that the final recommendation links each contested claim to its producing worker and source.
  2. The independent test establishes that concurrent worker requests never produce more than one approved ledger mutation.
  3. The verified result is that simulated uncertainty terminates within budget and records an actionable escalation.
  4. The recorded test establishes that a replay after a worker crash resumes pending tasks without repeating confirmed external writes.

Explanation

Correct Answer

B:

Single-writer boundaries avoid race conditions and duplicated financial side effects between collaborating agents; the condition under test is that two proposed agents can both edit the same financial ledger during an execution.

Incorrect Answers

A:

Traceable handoffs let reviewers attribute findings and replay a particular specialist result; the other workload is one where a coordinator has no visibility into which specialist produced a disputed conclusion, not the risk where a concurrency test lets two assistants submit separate updates to the same accounting entry.

C:

Bounded delegation prevents runaway coordination cost and makes operational failure predictable; the alternative helps if an orchestrator delegates tasks indefinitely when subordinate agents return uncertain results, whereas this scenario involves a concurrency test lets two assistants submit separate updates to the same accounting entry.

D:

Durable checkpoints support retries while protecting write operations that already succeeded; it is justified if a distributed agent run fails halfway through and supervisors cannot resume without duplicate actions, but here a concurrency test lets two assistants submit separate updates to the same accounting entry.

Question 52

An uncertain subordinate repeatedly delegates to further workers with no escalation to the caller. Which finding should the release gate require?

  1. The staged rollout demonstrates that a workflow benchmark matches output quality with fewer handoffs and lower token usage.
  2. The audit sample demonstrates that simulated uncertainty terminates within budget and records an actionable escalation.
  3. The readiness assessment shows that red-team probes show no restricted evidence leaking through agent messages or delegated tools.
  4. The independent test establishes that a replay after a worker crash resumes pending tasks without repeating confirmed external writes.
  5. The verified result is that concurrent worker requests never produce more than one approved ledger mutation.

Explanation

Correct Answer

B:

Bounded delegation prevents runaway coordination cost and makes operational failure predictable; the observed protection matters when an orchestrator delegates tasks indefinitely when subordinate agents return uncertain results.

Incorrect Answers

A:

Agents add coordination cost when responsibilities are not truly independent or context-separated; it serves conditions where an enterprise solution has two workers whose outputs are always used sequentially and never overlap, but this decision concerns an uncertain subordinate repeatedly delegates to further workers with no escalation to the caller.

C:

Per-agent privilege boundaries matter even when orchestration shares a global business objective; this makes sense when a security worker may retrieve restricted findings that a marketing worker must never see, but the requested boundary concerns an uncertain subordinate repeatedly delegates to further workers with no escalation to the caller.

D:

Durable checkpoints support retries while protecting write operations that already succeeded; this technique fits where a distributed agent run fails halfway through and supervisors cannot resume without duplicate actions, not the present case where an uncertain subordinate repeatedly delegates to further workers with no escalation to the caller.

E:

Single-writer boundaries avoid race conditions and duplicated financial side effects between collaborating agents; the trigger would be that two proposed agents can both edit the same financial ledger during an execution, while this problem involves an uncertain subordinate repeatedly delegates to further workers with no escalation to the caller.

Question 53

A supervised multi-agent investigation produces a contested finding and then crashes while saving its work to an external system. Release assurance must cover both tracing the finding to a specific specialist and resuming execution without replaying completed writes. Which TWO observed results establish those distinct guarantees? Choose TWO.

  1. The pilot evaluation finds that a replay after a worker crash resumes pending tasks without repeating confirmed external writes.
  2. The comparison run confirms that red-team probes show no restricted evidence leaking through agent messages or delegated tools.
  3. The staged rollout demonstrates that the final recommendation links each contested claim to its producing worker and source.
  4. The audit sample demonstrates that concurrent worker requests never produce more than one approved ledger mutation.
  5. The readiness assessment shows that simulated uncertainty terminates within budget and records an actionable escalation.

Explanation

Correct Answers

A:

Durable checkpoints support retries while protecting write operations that already succeeded; the test is necessary where a distributed agent run fails halfway through and supervisors cannot resume without duplicate actions; another obligation arises when a coordinator has no visibility into which specialist produced a disputed conclusion.

C:

Traceable handoffs let reviewers attribute findings and replay a particular specialist result; the safeguard must work when a coordinator has no visibility into which specialist produced a disputed conclusion; another obligation arises when a distributed agent run fails halfway through and supervisors cannot resume without duplicate actions.

Incorrect Answers

B:

Per-agent privilege boundaries matter even when orchestration shares a global business objective; that is defensible if a security worker may retrieve restricted findings that a marketing worker must never see, while the immediate question involves a challenged technical recommendation cannot be linked to the worker and artifact that produced it; another obligation arises when a distributed agent run fails halfway through and supervisors cannot resume without duplicate actions.

D:

Single-writer boundaries avoid race conditions and duplicated financial side effects between collaborating agents; this applies when two proposed agents can both edit the same financial ledger during an execution, whereas the current issue is that a challenged technical recommendation cannot be linked to the worker and artifact that produced it; another obligation arises when a distributed agent run fails halfway through and supervisors cannot resume without duplicate actions.

E:

Bounded delegation prevents runaway coordination cost and makes operational failure predictable; the alternative helps if an orchestrator delegates tasks indefinitely when subordinate agents return uncertain results, whereas the release criterion addresses after a worker crash, supervisors cannot distinguish unfinished actions from changes already committed externally; another obligation arises when a coordinator has no visibility into which specialist produced a disputed conclusion.

Question 54

A review compares two constantly sequential agents with a single staged process and asks for cost-quality evidence. What independently verifiable observation is needed?

  1. The sign-off record demonstrates that a workflow benchmark matches output quality with fewer handoffs and lower token usage.
  2. The acceptance evidence shows that a replay after a worker crash resumes pending tasks without repeating confirmed external writes.
  3. The pilot evaluation finds that conflicting worker findings are surfaced with citations and a recorded coordinator decision.
  4. The comparison run confirms that the final recommendation links each contested claim to its producing worker and source.

Explanation

Correct Answer

A:

Agents add coordination cost when responsibilities are not truly independent or context-separated; the condition under test is that an enterprise solution has two workers whose outputs are always used sequentially and never overlap.

Incorrect Answers

B:

Durable checkpoints support retries while protecting write operations that already succeeded; it is relevant if a distributed agent run fails halfway through and supervisors cannot resume without duplicate actions, whereas this scenario involves a review compares two constantly sequential agents with a single staged process and asks for cost-quality evidence.

C:

Clear ownership allows parallel expertise while a reconciliation step prevents contradictory final recommendations; it serves conditions where a research project has independent legal, market and security workstreams that need reconciliation, but here a review compares two constantly sequential agents with a single staged process and asks for cost-quality evidence.

D:

Traceable handoffs let reviewers attribute findings and replay a particular specialist result; the other case is when a coordinator has no visibility into which specialist produced a disputed conclusion, instead of when a review compares two constantly sequential agents with a single staged process and asks for cost-quality evidence.

Question 55

A security summary passed between agents accidentally includes raw findings outside the recipient's clearance. What outcome would demonstrate this requirement in operation?

  1. A reproducible run shows that simulated uncertainty terminates within budget and records an actionable escalation.
  2. The signed test record confirms that red-team probes show no restricted evidence leaking through agent messages or delegated tools.
  3. The sign-off record demonstrates that a replay after a worker crash resumes pending tasks without repeating confirmed external writes.
  4. The acceptance evidence shows that a workflow benchmark matches output quality with fewer handoffs and lower token usage.

Explanation

Correct Answer

B:

Per-agent privilege boundaries matter even when orchestration shares a global business objective; a valid sign-off must address the risk that a security worker may retrieve restricted findings that a marketing worker must never see.

Incorrect Answers

A:

Bounded delegation prevents runaway coordination cost and makes operational failure predictable; this control is needed when an orchestrator delegates tasks indefinitely when subordinate agents return uncertain results, but the actual risk is that a security summary passed between agents accidentally includes raw findings outside the recipient's clearance.

C:

Durable checkpoints support retries while protecting write operations that already succeeded; it serves a setting where a distributed agent run fails halfway through and supervisors cannot resume without duplicate actions, not the present case where a security summary passed between agents accidentally includes raw findings outside the recipient's clearance.

D:

Agents add coordination cost when responsibilities are not truly independent or context-separated; that is defensible if an enterprise solution has two workers whose outputs are always used sequentially and never overlap, while this problem involves a security summary passed between agents accidentally includes raw findings outside the recipient's clearance.

Question 56

After a worker crash, supervisors cannot distinguish unfinished actions from changes already committed externally. Which observed test outcome would justify sign-off?

  1. The release report verifies that a replay after a worker crash resumes pending tasks without repeating confirmed external writes.
  2. The controlled replay verifies that a workflow benchmark matches output quality with fewer handoffs and lower token usage.
  3. A reproducible run shows that concurrent worker requests never produce more than one approved ledger mutation.
  4. The regression suite verifies that red-team probes show no restricted evidence leaking through agent messages or delegated tools.

Explanation

Correct Answer

A:

Durable checkpoints support retries while protecting write operations that already succeeded; it tests the operating case in which a distributed agent run fails halfway through and supervisors cannot resume without duplicate actions.

Incorrect Answers

B:

Agents add coordination cost when responsibilities are not truly independent or context-separated; the approach helps when an enterprise solution has two workers whose outputs are always used sequentially and never overlap, but the present constraint is that after a worker crash, supervisors cannot distinguish unfinished actions from changes already committed externally.

C:

Single-writer boundaries avoid race conditions and duplicated financial side effects between collaborating agents; that belongs to situations where two proposed agents can both edit the same financial ledger during an execution, whereas the current issue is that after a worker crash, supervisors cannot distinguish unfinished actions from changes already committed externally.

D:

Per-agent privilege boundaries matter even when orchestration shares a global business objective; it is relevant if a security worker may retrieve restricted findings that a marketing worker must never see, whereas the release criterion addresses after a worker crash, supervisors cannot distinguish unfinished actions from changes already committed externally.

Question 57

A contract review prompt misses exceptions because it tries to extract, assess and draft at once. Which proposed design action directly addresses the risk?

  1. As a scoped first step, maintain a structured plan and evidence ledger between bounded investigative steps.
  2. In the release design, retrieve evidence per incident component and verify cited passages before global synthesis.
  3. During architecture review, split extraction, rule checking and report drafting into independently validated stages.
  4. In the proposed implementation, use Claude for transcription and a deterministic arithmetic validator for reconciliation.

Explanation

Correct Answer

C:

Distinct intermediate outputs make omissions diagnosable and reduce conflicting instructions in one model call; validation needs to show that each stage has its own labeled evaluation, with missing clauses isolated before drafting.

Incorrect Answers

A:

Externalized state prevents a multi-step assistant from revisiting failed hypotheses as context grows; the design helps when a long troubleshooting task repeatedly loses which hypotheses were already disproved, whereas this scenario involves a contract review prompt misses exceptions because it tries to extract, assess and draft at once.

B:

Evidence-first decomposition makes claim attribution explicit instead of asking a final model to backfill sources; that fits cases where an incident report requires citations for every claim but one final prompt invents them, but here a contract review prompt misses exceptions because it tries to extract, assess and draft at once.

D:

Arithmetic should be checked by exact computation rather than relying on probabilistic text generation; that would be useful if an invoice assistant must verify totals exactly after interpreting handwritten line items, not the case in which a contract review prompt misses exceptions because it tries to extract, assess and draft at once.

Question 58

A document migration involves thousands of independent PDFs but processing is serial. Which option supplies the needed architectural boundary?

  1. For the operating team, decompose the request so deterministic policy decides eligibility and Claude drafts the explanation.
  2. For this deployment, retrieve evidence per incident component and verify cited passages before global synthesis.
  3. As a scoped first step, create task boundaries with separate budgets and parallelize checks without shared mutable state.
  4. In the release design, partition by document and run controlled parallel workers with deterministic result merging.
  5. Under the system plan, split extraction, rule checking and report drafting into independently validated stages.

Explanation

Correct Answer

D:

Independent document analysis can scale horizontally without changing the per-document correctness contract; reviewers would verify that batch throughput improves while duplicate documents and inconsistent merges remain absent.

Incorrect Answers

A:

Separating authority from language generation keeps business rules outside model discretion; it is justified if a user request combines a refund decision and a sentiment summary, but only the latter is model-suitable, but the requested boundary concerns a document migration involves thousands of independent PDFs but processing is serial.

B:

Evidence-first decomposition makes claim attribution explicit instead of asking a final model to backfill sources; its scope includes cases where an incident report requires citations for every claim but one final prompt invents them, not the present case where a document migration involves thousands of independent PDFs but processing is serial.

C:

Explicit dependency analysis avoids serialization when subtasks can be safely evaluated independently; it targets cases where a complex workflow spans independent specialist checks with different latency targets, while this problem involves a document migration involves thousands of independent PDFs but processing is serial.

E:

Distinct intermediate outputs make omissions diagnosable and reduce conflicting instructions in one model call; that belongs to situations where a contract review prompt misses exceptions because it tries to extract, assess and draft at once, rather than the risk that a document migration involves thousands of independent PDFs but processing is serial.

Question 59

An assistant reads handwritten invoices and prepares customer refund explanations. Its OCR step can misread arithmetic, and friendly language cannot decide whether a refund is authorized under company policy. Which TWO task boundaries must the architect enforce together? Choose TWO.

  1. As the design priority, maintain a structured plan and evidence ledger between bounded investigative steps.
  2. In the implementation sequence, decompose the request so deterministic policy decides eligibility and Claude drafts the explanation.
  3. For the operating team, use Claude for transcription and a deterministic arithmetic validator for reconciliation.
  4. For this deployment, create task boundaries with separate budgets and parallelize checks without shared mutable state.
  5. As a scoped first step, partition by document and run controlled parallel workers with deterministic result merging.

Explanation

Correct Answers

B:

Separating authority from language generation keeps business rules outside model discretion; the go-live check should establish that a disputed refund follows encoded policy while tone variation changes only the explanation; another obligation arises when an invoice assistant must verify totals exactly after interpreting handwritten line items.

C:

Arithmetic should be checked by exact computation rather than relying on probabilistic text generation; one measurable outcome is that injected sum errors are rejected by the calculation stage before invoices enter approval; another obligation arises when a user request combines a refund decision and a sentiment summary, but only the latter is model-suitable.

Incorrect Answers

A:

Externalized state prevents a multi-step assistant from revisiting failed hypotheses as context grows; this technique fits where a long troubleshooting task repeatedly loses which hypotheses were already disproved, but the present constraint is that a user request combines a refund decision and a sentiment summary, but only the latter is model-suitable; another obligation arises when an invoice assistant must verify totals exactly after interpreting handwritten line items.

D:

Explicit dependency analysis avoids serialization when subtasks can be safely evaluated independently; that fits cases where a complex workflow spans independent specialist checks with different latency targets, while the required safeguard concerns an invoice assistant must verify totals exactly after interpreting handwritten line items; another obligation arises when a user request combines a refund decision and a sentiment summary, but only the latter is model-suitable.

E:

Independent document analysis can scale horizontally without changing the per-document correctness contract; this control is needed when a document migration involves thousands of independent PDFs but processing is serial, not the risk where a user request combines a refund decision and a sentiment summary, but only the latter is model-suitable; another obligation arises when an invoice assistant must verify totals exactly after interpreting handwritten line items.

Question 60

A long troubleshooting task repeatedly loses which hypotheses were already disproved. What is the most appropriate initial architecture decision?

  1. For the current integration, retrieve evidence per incident component and verify cited passages before global synthesis.
  2. Under the system plan, maintain a structured plan and evidence ledger between bounded investigative steps.
  3. As the design priority, split extraction, rule checking and report drafting into independently validated stages.
  4. For the affected service, decompose the request so deterministic policy decides eligibility and Claude drafts the explanation.

Explanation

Correct Answer

B:

Externalized state prevents a multi-step assistant from revisiting failed hypotheses as context grows; validation needs to show that a replay shows unique investigated hypotheses and a final conclusion tied to observed evidence.

Incorrect Answers

A:

Evidence-first decomposition makes claim attribution explicit instead of asking a final model to backfill sources; this applies when an incident report requires citations for every claim but one final prompt invents them, but here a long troubleshooting task repeatedly loses which hypotheses were already disproved.

C:

Distinct intermediate outputs make omissions diagnosable and reduce conflicting instructions in one model call; it is justified if a contract review prompt misses exceptions because it tries to extract, assess and draft at once, not the case in which a long troubleshooting task repeatedly loses which hypotheses were already disproved.

D:

Separating authority from language generation keeps business rules outside model discretion; its scope includes cases where a user request combines a refund decision and a sentiment summary, but only the latter is model-suitable, whereas the problem presented is that a long troubleshooting task repeatedly loses which hypotheses were already disproved.

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