Anthropic CCAR-P (Claude Certified Architect - Professional) Practice Test 38

Question 370

The team is unsure whether a scanned archive can yield usable text. It has one week for a technical spike. Which deliverable best supports the next investment decision?

  1. A polished production interface around a single clean scanned page
  2. A complete long-term operating model without testing extraction quality
  3. Measured extraction quality on representative scans, failure categories, and a recommendation
  4. A count of OCR packages installed during the week

Explanation

Correct Answer

C:

A feasibility spike should resolve the targeted technical uncertainty with bounded effort. Representative results and failure analysis explain whether and how to proceed.

Incorrect Answers

A:

The interface does not establish feasibility across the archive. A clean example can conceal the document conditions that determine the investment decision.

B:

Operating plans may be needed later, but they do not answer the spike's central uncertainty. The immediate deliverable should establish technical feasibility.

D:

Installed tools measure activity rather than usable extraction. The decision needs observed quality and remaining failure modes.

Question 371

The assistant produces technically correct explanations that customer-facing staff find too specialized to use. How should acceptance be refined?

  1. Require longer explanations so every technical detail is included
  2. Let engineers alone define the vocabulary because they know the system best
  3. Remove factual correctness from acceptance and judge only friendliness
  4. Test explanations with intended users for comprehension and task usefulness alongside correctness

Explanation

Correct Answer

D:

Correctness is essential but does not ensure that the audience can use the explanation. Acceptance should include evidence about comprehension in the intended setting.

Incorrect Answers

A:

Additional detail can increase cognitive burden without improving comprehension. The appropriate content and language should be tested with the audience.

B:

Technical expertise helps ensure accuracy but does not represent every user's comprehension. Intended users should inform the communication criteria.

C:

Audience fit does not replace accuracy. The explanation must remain correct while becoming understandable and useful to its intended users.

Question 372

During discovery, a stakeholder suggests that the assistant could eventually approve purchases. The notes later appear in a signed scope as an approved capability. What document practice would prevent this confusion?

  1. Treat every stakeholder suggestion as approved unless challenged within an hour
  2. Clearly distinguish ideas, assumptions, decisions, and authorized commitments
  3. Record only final decisions and delete all discovery notes immediately
  4. Include more detailed implementation estimates next to each suggestion

Explanation

Correct Answer

B:

Discovery produces possibilities as well as agreed requirements. Explicit status labels prevent an exploratory suggestion from silently becoming a delivery or authority commitment.

Incorrect Answers

A:

Silence under an arbitrary deadline does not establish authorization. Consequential scope should follow the agreed decision process.

C:

Discarding context is unnecessary and can harm traceability. The issue is distinguishing status, not eliminating the record of exploration.

D:

An estimate can make an idea look more concrete without changing its approval status. The document must explicitly distinguish proposal from commitment.

Question 373

A fragile ingestion script causes frequent missed updates, while a code-style cleanup would make the repository more consistent. How should the team compare these tasks?

  1. Use reliability impact, recurrence, risk, and effort rather than grouping all debt as equivalent
  2. Defer both indefinitely because neither adds a new visible feature
  3. Choose the code-style cleanup because its completion is easier to measure
  4. Allocate equal time because both tasks are labeled technical debt

Explanation

Correct Answer

A:

Technical debt has different consequences. The ingestion failures affect source currency and service quality, which should be weighed against the cleanup's benefits and cost.

Incorrect Answers

B:

Maintenance can protect reliability and delivery capacity. Ignoring it solely because it is not a feature can increase user-facing failures.

C:

Ease of measurement does not establish higher value. The recurring ingestion failures have an operational impact that must be assessed.

D:

A shared label does not imply equal impact or urgency. The tasks should be evaluated by their effects on the service and future work.

Question 374

A team is preparing a go-live review for a new assistant. Which three areas should have explicit readiness evidence? Choose THREE.

  1. Acceptance results for the intended users, tasks, and supported boundaries
  2. Operational ownership, monitoring, response, and recovery arrangements
  3. A polished pilot demo used as the sole quality sign-off
  4. A plan to assign service responsibility after the first production incident
  5. Required security and data-handling controls for the deployment
  6. A passing vendor benchmark used instead of deployment-specific control evidence

Explanation

Correct Answers

A:

Go-live should be grounded in demonstrated fitness for the actual scope. Evidence from unrelated tasks does not establish that the intended workflow is ready.

B:

A service needs people and processes to sustain it after release. Technical feature completion alone does not establish operational readiness.

E:

The production environment must meet its applicable control requirements. These are part of readiness rather than optional follow-up work.

Incorrect Answers

C:

A selected demonstration does not establish acceptance across intended tasks and boundaries. Readiness needs representative results and documented treatment of failures.

D:

An operating service needs accountable ownership before an incident occurs. Deferring assignment leaves response and escalation responsibilities unresolved at launch.

F:

A vendor result evaluates its stated setup, not the custom deployment's safeguards. Required security and data-handling controls need evidence for the actual application.

Question 375

The team has time for two feedback activities before refining a case assistant. Which two would most directly reveal workflow problems? Choose TWO.

  1. Tally requested features without linking them to observed task failures
  2. Review structured failure reports with the people responsible for the affected workflow
  3. Collect a satisfaction rating without observing task outcomes or failure details
  4. Ask the project team to predict user complaints without observing users
  5. Observe representative users completing real task types with the assistant

Explanation

Correct Answers

B:

Failure reports provide concrete cases, while workflow owners can explain their operational consequences. Together they support targeted and relevant improvements.

E:

Observation can reveal workarounds, confusion, and handoff failures that users may not describe in surveys. It connects feedback to actual task behavior.

Incorrect Answers

A:

Feature votes can indicate interest but do not show where the current workflow breaks. Observation and structured failure review connect feedback to actual task problems.

C:

Satisfaction can inform product decisions but does not directly locate workflow problems. The limited investigation should connect feedback to concrete task behavior.

D:

Internal predictions can generate hypotheses but do not replace user evidence. The chosen activities should test assumptions against actual experience.

Question 376

A repository's CLAUDE.md says not to access production credentials, but the development process can still read those credentials from its environment. What change establishes the intended security boundary?

  1. Move the credential into an untracked file within the same readable workspace
  2. Repeat the restriction in more sections of the repository instructions
  3. Rely on a pre-commit secret scan while leaving runtime access unchanged
  4. Remove production credentials from the development process and enforce appropriate access restrictions
  5. Ask the coding assistant to confirm compliance before each edit

Explanation

Correct Answer

D:

Repository instructions guide behavior but do not remove operating-system access. The environment should make production credentials unavailable to this development workflow.

Incorrect Answers

A:

Keeping a file out of version control does not prevent local access. The development environment would still be able to read the secret.

B:

Repetition may reinforce guidance but does not change process permissions. The credential remains accessible to code running in the environment.

C:

A repository scan can detect some committed secrets but does not remove the process's ability to read production credentials. The runtime boundary still needs enforcement.

E:

A confirmation is not an access-control boundary. The development process still possesses the credential regardless of the assistant's statement.

Question 377

A monorepo contains a Python service and a TypeScript application with different testing commands. A single ambiguous instruction repeatedly sends the assistant to the wrong test runner. What guidance structure would help?

  1. Require both complete test suites after every single-file inspection
  2. Keep shared guidance general and place component-specific instructions near their relevant code
  3. Delete all test guidance and rely on the assistant's default conventions
  4. Put both commands in one sentence without identifying their component

Explanation

Correct Answer

B:

Scoped instructions make the applicable commands clearer for each part of the repository. This reduces ambiguity while preserving common conventions at the shared level.

Incorrect Answers

A:

Running everything may hide the guidance problem while adding avoidable cost. Component-specific instructions provide a clearer and more targeted workflow.

C:

Default conventions may not match this repository. Removing known information makes correct command selection less dependable.

D:

The assistant would still need to infer which command applies. The failure stems from ambiguous scope, not simply the absence of command names.

Question 378

A team repeatedly asks its coding assistant to perform the same release-note procedure with a defined checklist and supporting scripts. What packaging best supports consistent reuse?

  1. A stored final release note reused unchanged for future versions
  2. A reviewed reusable skill containing the procedure and references to its supporting scripts
  3. A fresh free-form prompt reconstructed from memory for every release
  4. A large comment copied into every source file in the repository

Explanation

Correct Answer

B:

A reusable procedural package makes the workflow discoverable and maintainable. Review and versioning keep repeated execution aligned with the team's intended process.

Incorrect Answers

A:

A previous output does not encode the procedure for new changes. Reusing it can omit current work and preserve stale content.

C:

Reconstruction introduces omissions and inconsistent interpretation. The procedure is stable enough to capture once and maintain deliberately.

D:

Duplicating the procedure across files creates maintenance drift and unnecessary context. A single maintained package better represents the shared workflow.

Question 379

An assistant writes code using a method that looks plausible but is absent from the installed package version. What should the developer do before extending that code?

  1. Suppress the exception and return success to the caller
  2. Add a compatibility wrapper that calls the same unverified method
  3. Inspect the installed API or its version-matched documentation and correct the call
  4. Add retries around the nonexistent method call
  5. Upgrade every package without first checking which interface is required

Explanation

Correct Answer

C:

Generated API names can be plausible without existing in the relevant version. The implementation should be grounded in the actual dependency contract.

Incorrect Answers

A:

That would conceal the missing operation rather than implement the task. The application needs the intended behavior through a valid API.

B:

A wrapper does not make an absent API available. The developer must first establish a supported operation in the installed dependency version.

D:

Retries cannot create a method missing from the package. The failure is an invalid interface assumption rather than a transient runtime condition.

E:

A broad upgrade introduces unrelated changes and may not provide the imagined method. The first step is to establish the relevant supported contract.

Question 380

A generated test compares an expiry timestamp with the real current clock and fails intermittently near midnight. What design makes the boundary test reliable?

  1. Remove the expiry assertion and keep only the object-creation check
  2. Increase the allowed runtime so the test has longer to complete
  3. Run the test only during office hours in one timezone
  4. Inject a controllable clock and test explicit times around the expiry boundary

Explanation

Correct Answer

D:

The test should determine its time input rather than depend on when it runs. Explicit before, at, and after cases exercise the boundary reproducibly.

Incorrect Answers

A:

That avoids the failure by abandoning the behavior under test. The expiry rule still needs coverage with deterministic time inputs.

B:

A longer runtime does not fix an unstable time assumption and can cross additional boundaries. The test needs controlled inputs.

C:

Scheduling restricts when the defect appears without validating the boundary reliably. A controlled clock permits the same meaningful test at any run time.

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