Anthropic CCAR-P (Claude Certified Architect - Professional) Practice Test 35

Question 340

A vendor announces that an API behavior used by the application will change in three months. What lifecycle response is most appropriate?

  1. Wait until users report failures to determine which dependencies matter
  2. Assess affected dependencies, validate a migration, and schedule rollout before the change
  3. Freeze the local code and assume the remote behavior will remain unchanged
  4. Immediately migrate every integration without testing compatibility

Explanation

Correct Answer

B:

A known external change creates a planning window. The team should identify impact and test the adaptation rather than wait for production breakage.

Incorrect Answers

A:

The dependency can be investigated before the deadline. Waiting discards the opportunity to reduce avoidable disruption.

C:

A local freeze does not prevent a vendor-side change. The application still depends on the announced behavior and needs an impact assessment.

D:

Urgency does not eliminate migration risk. The available window should be used for validation and an orderly rollout.

Question 341

An internal assistant is being replaced, but some teams still depend on its scheduled reports and stored links. What should the retirement plan include?

  1. Identify dependents, communicate a transition, migrate needed artifacts, and define shutdown criteria
  2. Keep both services indefinitely without assigning maintenance ownership
  3. Delete historical artifacts first to encourage faster adoption of the replacement
  4. Disable the old interface immediately because the replacement is available

Explanation

Correct Answer

A:

Retirement affects workflows and references beyond the visible interface. A transition plan prevents abrupt loss while establishing when the old service can safely stop.

Incorrect Answers

B:

An indefinite overlap preserves cost and risk without a decision path. The plan should establish responsibilities and a clear retirement condition.

C:

Removing needed records can disrupt operations and violate retention needs. Migration and lifecycle requirements should guide artifact handling.

D:

Replacement availability does not prove that all dependencies have migrated. Scheduled reports and stored references may still be in active use.

Question 342

The roadmap schedules autonomous case resolution before reliable case lookup and write authorization are implemented. What sequencing principle should guide the revision?

  1. Prioritize solely by estimated coding hours per feature
  2. Build the most impressive demo first and add prerequisites after broad release
  3. Deliver prerequisite capabilities and controls before dependent autonomy
  4. Start all capabilities together and assume the prerequisites finish first

Explanation

Correct Answer

C:

The higher-authority workflow relies on reliable data access and enforced permissions. Sequencing should reflect those dependencies rather than the visibility of the final feature.

Incorrect Answers

A:

Effort estimates matter, but short tasks are not necessarily prerequisites. The roadmap must consider dependency and risk as well as cost.

B:

A demo can hide missing operational controls. Broad release should not precede capabilities required for the workflow to function safely.

D:

Concurrent work can be useful, but readiness cannot rest on an unverified ordering assumption. The dependent capability needs its prerequisites demonstrated before it is enabled.

Question 343

A proposed assistant depends on policy documents, but nobody owns their approval status and many versions contradict each other. What should be addressed before treating retrieval quality as the main blocker?

  1. Ask users to choose whichever returned policy is most convenient
  2. Increase the embedding dimension to separate contradictory policies
  3. Choose the most frequently retrieved version as the authoritative policy
  4. Establish authoritative ownership, version status, and a usable source corpus
  5. Expand the model's context until all versions fit in one request

Explanation

Correct Answer

D:

Retrieval cannot reliably establish policy truth when authority is undefined. The source collection needs ownership and version clarity before search tuning can resolve the central problem.

Incorrect Answers

A:

Convenience is not a valid authority rule. The application needs an agreed source of policy truth for its intended use.

B:

Vector representation cannot determine which version has organizational authority. The missing decision concerns the source corpus and its governance.

C:

Retrieval frequency can reflect usage or ranking bias, not approval status. Authority must come from the responsible source owner and version process.

E:

Including every version may expose the model to more contradiction. It does not establish which policy is approved and current.

Question 344

A pilot meets accuracy and latency targets, but staff avoid it because using it requires duplicate entry into an existing case system. What should the next iteration prioritize?

  1. Treat low usage as evidence that the accuracy evaluation was necessarily wrong
  2. Remove or reduce the workflow duplication and measure task completion in context
  3. Increase promotional messages without changing the duplicate-entry requirement
  4. Replace the model with a larger version before studying the workflow

Explanation

Correct Answer

B:

Adoption is blocked by the surrounding process rather than the measured model quality. Integrating the workflow can address the extra work that staff currently face.

Incorrect Answers

A:

A system can be accurate and still inconvenient to use. The adoption problem requires examining workflow fit rather than assuming a measurement error.

C:

Promotion may raise awareness but does not remove the additional work. The product needs to address the reason staff avoid it.

D:

The stated barrier is duplicate entry, so model capacity is not the first supported explanation. A replacement may leave the same friction unchanged.

Question 345

Staff either accept every assistant answer or refuse to use it at all. What training approach best supports appropriate reliance?

  1. Teach only the fastest way to submit a prompt and copy the answer
  2. Practice representative successful and failing cases, verification steps, and escalation decisions
  3. Train staff only on the provider's general capability benchmark results
  4. Present only flawless demonstration outputs to build confidence

Explanation

Correct Answer

B:

Users need experience recognizing when the assistant is useful and when additional checking is required. Concrete cases support calibrated reliance better than blanket confidence or rejection.

Incorrect Answers

A:

Interface speed does not teach users how to assess reliability. The observed problem concerns judgment about verification and escalation.

C:

Benchmarks do not show how to verify this workflow's outputs or recognize its failure modes. Users need task-specific practice in appropriate reliance.

D:

Selective examples conceal limitations and can encourage overreliance. Training should include relevant failure cases and appropriate responses.

Question 346

A sponsor wants a slide stating that the assistant will eliminate all manual review. Current evidence supports automation for a bounded subset of low-risk cases. What should the architect present?

  1. The provider's strongest benchmark as proof that manual review is unnecessary
  2. The supported scope, measured evidence, remaining review needs, and expansion criteria
  3. No performance claim at all until every possible case has been tested
  4. The requested absolute claim, with limitations kept in private notes

Explanation

Correct Answer

B:

Communication should reflect what has been demonstrated and what remains uncertain. A bounded claim gives stakeholders a usable basis for planning and investment.

Incorrect Answers

A:

A general benchmark does not establish this workflow's review requirements. The deployment decision should follow relevant application evidence.

C:

Useful evidence can be communicated with clear scope and uncertainty. Exhaustive certainty is not required to describe the demonstrated subset.

D:

Private caveats do not correct the public commitment. Stakeholders may plan around an unsupported promise of complete automation.

Question 347

A service has exhausted its agreed error budget. The operating policy pauses discretionary releases until reliability recovers, while allowing urgent reliability fixes. What should the release owner do with a cosmetic feature update?

  1. Release it because cosmetic changes never affect service reliability
  2. Defer it under the agreed policy while prioritizing reliability recovery
  3. Approve it by excluding recent failures from the budget calculation
  4. Pause all changes, including the urgent fix needed to restore reliability

Explanation

Correct Answer

B:

The policy explicitly distinguishes discretionary changes from recovery work. Applying it consistently preserves the agreed response to exhausted reliability allowance.

Incorrect Answers

A:

Even a small change can affect deployment or behavior. The policy does not exempt cosmetic features from the discretionary-release pause.

C:

Changing measurement to permit the release would evade the agreed policy. Any legitimate correction to the calculation needs evidence rather than a desired outcome.

D:

The policy explicitly allows urgent reliability fixes. A blanket freeze would disregard that exception and could delay the required recovery.

Question 348

A small pilot averages ten requests per minute, but production will involve a synchronized start-of-day workflow across many teams. What forecast should drive readiness planning?

  1. The pilot's average request rate multiplied only by calendar days
  2. The highest possible user count assuming every user sends one request per month
  3. Expected peak concurrency and task mix, with explicit assumptions and headroom
  4. The provider's published token limit without modeling synchronized arrivals

Explanation

Correct Answer

C:

The synchronized workflow can create demand far above the pilot average. Planning should reflect when work arrives and what resources the tasks consume.

Incorrect Answers

A:

Calendar duration does not capture simultaneous demand. The production pattern changes concurrency and can expose bottlenecks absent from the pilot.

B:

A monthly count obscures the stated start-of-day burst. Readiness needs a demand model that reflects the actual timing of work.

D:

A token limit constrains a request or quota but does not specify the application's arrival pattern and service demand. Peak workload and task mix are needed to assess capacity.

Question 349

A project milestone says 'agent finished,' but engineering, security, and operations interpret it differently. What revision would make progress review more reliable?

  1. Use the number of commits as the completion criterion
  2. Define observable completion criteria and required evidence for each relevant owner
  3. Split the phrase into several similarly vague milestones
  4. Replace the milestone with a percentage complete estimated by the lead developer

Explanation

Correct Answer

B:

A milestone should identify what is complete and how that is demonstrated. Shared criteria reduce ambiguous status reports and late disagreements.

Incorrect Answers

A:

Commit volume measures activity rather than readiness. It does not establish that engineering, security, or operational requirements have been met.

C:

More labels do not create objective criteria. Each milestone still needs observable outcomes that the relevant owners can assess.

D:

A percentage can conceal different definitions of completion. The underlying acceptance and evidence requirements remain unresolved.

Question 350

A pilot succeeds with expert volunteers who know every system limitation. The intended rollout includes occasional users with little domain knowledge. What evidence is missing before extrapolating the result?

  1. More repetitions of the same tasks by the original expert volunteers
  2. Results from representative users performing the intended production tasks
  3. A larger synthetic benchmark scored without user interaction
  4. Higher throughput measurements using the expert volunteers' saved prompts
  5. A longer satisfaction survey sent only to the expert volunteers

Explanation

Correct Answer

B:

Expert volunteers may compensate for weaknesses that occasional users cannot recognize. The rollout decision needs evidence about the actual user population and workflow.

Incorrect Answers

A:

Repetition can improve estimates for that cohort but does not address the population gap. The planned users differ in relevant knowledge and behavior.

C:

A benchmark may test model quality while omitting how inexperienced users interpret and act on results. The identified uncertainty concerns the human workflow.

D:

Throughput evidence addresses capacity rather than user representativeness. It cannot show whether occasional users can complete tasks correctly.

E:

Detailed opinions from the same cohort do not establish usability for occasional users. The evaluation needs the missing population represented.

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