Anthropic CCAR-P (Claude Certified Architect - Professional) Practice Test 5

Question 40

Claude Code reports passing tests locally, but CI fails on the same commit. The engineer has not compared the CI test command, working directory, or configuration with the local run. What should the engineer do first?

  1. Reproduce the CI test invocation with its working directory and effective configuration
  2. Increase the model context window while keeping the failing command unchanged
  3. Rewrite the tests to accept the output currently produced by CI
  4. Copy the developer’s package cache into CI before checking which command is run

Explanation

Correct Answer

A:

The discrepancy points to the execution environment rather than a need for a new model answer. Matching the actual CI invocation and configuration makes the failure reproducible and exposes assumptions hidden by the local run.

Incorrect Answers

B:

Context capacity affects what the assistant can consider, but does not correct an incorrect command or environment. The failure must first be reproduced under the conditions that produced it.

C:

Changing the expected output before diagnosing the difference risks turning a defect into accepted behavior. The test contract should be preserved while the environment and invocation are investigated.

D:

Dependency differences are one possibility, but copying a cache changes the environment without identifying the cause. Establishing the exact failing invocation is the more controlled first step.

Question 41

A billing assistant explains discounts, but the payable amount must exactly follow an approved arithmetic formula. Where should the amount be calculated?

  1. In the same prompt that writes explanations
  2. By choosing the most confident generated amount
  3. By averaging results from several models
  4. In a tested deterministic billing function

Explanation

Correct Answer

D:

A fixed formula belongs in executable business logic. Claude can explain the returned result without becoming the authority for arithmetic correctness.

Incorrect Answers

A:

Combining calculation with prose generation makes a deterministic obligation depend on model behavior, even when the explanation sounds internally consistent.

B:

Confidence is not a mathematical verification method. A plausible-looking amount can violate the formula while receiving a confident explanation.

C:

Agreement can reduce some random variation but does not establish compliance with the specified formula or eliminate a shared calculation error.

Question 42

A librarian assistant answers one question from a small, stable set of relevant passages. It needs no external action or follow-up investigation. Which baseline should be tested first?

  1. A manager coordinating five specialist agents
  2. A pipeline generating and ranking ten answers
  3. One model call augmented with the passages
  4. An autonomous loop that searches until satisfied

Explanation

Correct Answer

C:

The required evidence is already available and the task has a single output. Additional orchestration should earn its complexity through measured benefit.

Incorrect Answers

A:

Specialization adds coordination and synthesis costs without a demonstrated division of work. The stated task does not require several independent investigations.

B:

Candidate ranking may help a difficult generation problem, but there is no evidence that this simpler task needs the additional calls.

D:

Repeated searching is unnecessary when the relevant evidence is supplied. It can increase latency without improving this bounded question-answering task.

Question 43

A request router distinguishes account questions from technical faults. Some messages contain both. Silent misrouting is costly. What should its output contract support?

  1. The route most common in training examples
  2. A rule that sends every request to every available route
  3. Exactly one category for every message
  4. An explicit mixed or uncertain route

Explanation

Correct Answer

D:

A router should represent cases that do not fit one category. That enables controlled clarification or composite handling instead of inventing certainty.

Incorrect Answers

A:

A frequency-based default can favor the dominant category while systematically losing mixed cases. It does not expose uncertainty to the workflow.

B:

Universal fan-out avoids choosing but includes unrelated work and does not represent the actual ambiguity. A mixed or uncertain route should reflect the request rather than bypass classification entirely.

C:

Forcing a single label hides the ambiguity specified in the scenario. One part of a mixed request may then be omitted downstream.

Question 44

A coordinator receives code changes affecting an unpredictable number of independent modules. Which orchestration pattern fits assigning the actual affected modules to workers?

  1. A majority vote over complete repository reviews
  2. A coordinator that creates bounded module-specific tasks
  3. A fixed pipeline with one stage per known module
  4. A single worker repeated once per repository file

Explanation

Correct Answer

B:

The work breakdown becomes known after inspecting the change. Dynamic assignment can reflect that structure while limiting worker scope and concurrency.

Incorrect Answers

A:

Whole-repository voting duplicates broad work instead of partitioning it. It also makes attribution of module-specific findings harder for the coordinator.

C:

A predefined stage list becomes awkward when affected modules vary. It can perform unnecessary work or omit modules outside the anticipated list.

D:

Blind repetition ignores the dependency analysis that identifies affected modules. It adds cost without ensuring that the right change boundaries are reviewed.

Question 45

Specialist agents return free-form opinions that a coordinator often misattributes. Which interface change most directly improves reliable synthesis?

  1. Limit every specialist to one sentence
  2. Ask the coordinator to infer missing references
  3. Require all specialists to use identical conclusions
  4. Require finding, evidence reference, scope, and uncertainty fields

Explanation

Correct Answer

D:

Structured findings make the meaning and provenance of each contribution explicit. The coordinator can compare claims without guessing which source or scope they concern.

Incorrect Answers

A:

Brevity reduces text but may remove evidence and qualifiers. It does not establish a reliable relationship between a conclusion and its source.

B:

Inference can fabricate provenance when the worker omitted it. Source relationships should be supplied by the worker and checked during synthesis.

C:

Uniform conclusions would hide disagreement rather than resolve attribution. The interface needs consistent fields while allowing evidence-based differences in findings.

Question 46

A drafting task has a clear rubric and revisions usually improve when specific feedback is supplied. Which architecture is most suitable if extra latency is acceptable?

  1. Parallel drafts with no evaluation criteria
  2. A revision stage triggered by every completed draft regardless of rubric results
  3. A bounded draft-evaluate-revise loop
  4. One autonomous agent with no revision limit

Explanation

Correct Answer

C:

Explicit criteria let an evaluator produce actionable feedback. A revision limit and acceptance rule keep iterative improvement from becoming an uncontrolled process.

Incorrect Answers

A:

Several candidates offer diversity, but without a rubric the system lacks a grounded way to choose the one meeting the requirement.

B:

Mandatory revision can alter content that already meets the criteria. Feedback should identify an actual deficiency before another revision is attempted.

D:

Feedback can be useful, but unlimited revision creates unpredictable cost and delay. The task still requires an externally defined stopping policy.

Question 47

A team proposes an evaluator loop for creative designs but cannot define acceptance criteria or provide reliable preference judgments. What should be established first?

  1. A guarantee that two models will agree
  2. A usable evaluation signal and stopping policy
  3. A rule that the final draft always wins
  4. A larger number of automatic revision rounds

Explanation

Correct Answer

B:

An optimization loop needs a way to distinguish improvement from arbitrary change. Without it, repeated revisions may only add expense and instability.

Incorrect Answers

A:

Agreement between models is not equivalent to stakeholder acceptance. A shared preference or blind spot can produce consensus on unsuitable designs.

C:

Recency is not evidence of quality. The last draft could remove useful features or introduce defects that previous drafts did not contain.

D:

More rounds amplify the absence of a success criterion. They cannot establish whether later designs are more suitable than earlier ones.

Question 48

Three reviewing agents share the same prompt and source omission, and all approve an incorrect answer. What does this show about majority voting?

  1. The answer is correct because agreement is unanimous
  2. Only the number of voters needs increasing
  3. Three votes establish independent verification
  4. Consensus does not remove common-mode errors

Explanation

Correct Answer

D:

Voting benefits depend on the independence and quality of the evidence behind each judgment. Shared omissions can make every reviewer fail together.

Incorrect Answers

A:

Unanimity describes the votes, not their relationship to reality. A common false assumption can survive an entirely unanimous review.

B:

Adding more reviewers with the same blind spot may reinforce the error. The evaluation design must address the shared failure cause.

C:

Separate executions are not necessarily independent evidence. Identical missing information can produce the same wrong conclusion in every execution.

Question 49

Two agents independently update an order's status and overwrite each other's changes. Which architecture best prevents lost updates while retaining specialist analysis?

  1. Give one coordinator ownership of state transitions
  2. Let each agent maintain its own authoritative status
  3. Increase the agents' shared context window
  4. Append the latest model response to the order

Explanation

Correct Answer

A:

Specialists can propose changes while a single authoritative transition mechanism validates and commits them. This separates parallel analysis from conflicting writes.

Incorrect Answers

B:

Multiple independent authorities preserve the conflict. Downstream consumers still need a rule for deciding which incompatible status represents the order.

C:

More context may improve awareness but does not provide atomic update control. Concurrent writes still require a consistency mechanism outside the model.

D:

Arrival order does not establish a valid business transition. A slower response may have been computed from older state than a faster one.

Question 50

A user cancels a long research job while workers are still running. Which design best avoids continued unnecessary work and accidental publication?

  1. Hide the progress indicator and leave workers unchanged
  2. Delete the user's browser session immediately
  3. Propagate cancellation and gate final publication on job state
  4. Tell the coordinator to ignore future worker text

Explanation

Correct Answer

C:

Workers need a cancellation signal, and the publisher must check the authoritative job state. Cancellation is a lifecycle condition, not just a user-interface change.

Incorrect Answers

A:

Removing the indicator changes presentation only. Computation and downstream publication can continue unless execution components receive and honor cancellation.

B:

Session removal does not necessarily terminate server-side jobs. The backend needs an explicit cancellation contract for already accepted work.

D:

Ignoring output may avoid synthesis, but workers still consume resources. It also fails to constrain any independent publication or side-effect path.

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