Anthropic CCA-F (Claude Certified Architect - Foundations (CCAR-F)) Practice Test 31

Question 300

A policy-sensitive contract extractor fails on rare jurisdictions and faces unclear exceptions requiring an authorized decision. Which TWO review actions are appropriate? Choose TWO.

  1. Refer undocumented jurisdictional exceptions to an authorized policy owner before commitment
  2. Sample unfamiliar, high-risk jurisdictions for deeper human review rather than auto-acceptance
  3. Connect approval to immutable action parameters and require reapproval after material changes
  4. Auto-approve unfamiliar jurisdictions when they appear infrequently, since a small cohort cannot affect overall accuracy much

Explanation

Correct Answers

A:

Routing uncertain policy exceptions to authorized review matters in this scenario: policy ambiguity is not resolved by fluent reasoning; A controlled exception channel lets a human decide without making an unauthorized promise.

B:

Aggregate accuracy can hide weak performance on small important cohorts. Risk-aware stratified checks expose failures that a uniform random sample might miss.

Incorrect Answers

C:

Approval of one transaction should not authorize a modified one; its benefits for tying human approval to exact mutation details do not resolve the need for targeting review to difficult high-impact document cohorts.

D:

Low-volume jurisdictions can carry disproportionate regulatory risk. Novel or unclear exceptions should be escalated rather than accepted on global averages.

Question 301

A generated solution needs both code quality and factual documentation review. What task design supports genuine independence?

  1. Formalize measurable deliverables and pass-fail conditions for each subtask
  2. Assign separate evaluators with distinct rubrics and merge after both finish
  3. Prefer direct deterministic lookup and rule evaluation for the simple stages
  4. Batch closely related low-cost tasks into a meaningful worker assignment
  5. Delegate one bounded comparison to each worker and merge on a shared schema

Explanation

Correct Answer

B:

Different quality dimensions need different tests. An author self-check is less independent than separate bounded evaluation of code behavior and external claims.

Incorrect Answers

A:

A task boundary is useful when the parent can decide whether its output satisfies requirements; the stated problem requires decomposing evaluation into distinct quality checks rather than specifying task-level acceptance criteria.

C:

For separating deterministic steps from inference, not every step requires another model or subagent; however, this requirement concerns decomposing evaluation into distinct quality checks.

D:

Excessive decomposition multiplies setup and context cost; that may serve avoiding overfragmentation of delegated work, but it is no substitute for decomposing evaluation into distinct quality checks.

E:

Independent subtasks can run concurrently when they have comparable output contracts; this may support decomposing parallel independent analyses but cannot alone meet the requirement for decomposing evaluation into distinct quality checks.

Question 302

A team repeats the same release checklist through a long prompt on every deployment. What reusable extension fits?

  1. Decouple dependency installation behind explicit approval and documented prerequisites
  2. Persist supporting resources within the skill and load them when needed
  3. Write a release skill's trigger description without creating its actual executable workflow instructions
  4. Package the release workflow as a project skill with clear instructions

Explanation

Correct Answer

D:

A reusable skill can collect steps, supporting files and tool permissions under one invocable workflow. This reduces repeated manual prompting and centralizes review.

Incorrect Answers

A:

Although skills are reusable instructions, not a blanket permission grant for arbitrary software installation, the choice addresses requiring approval for skill-driven dependency changes and not the requested making a repeatable release workflow reusable.

B:

Skills can bundle reference material and executable helpers for a focused task; the deciding condition is making a repeatable release workflow reusable, whereas this action concerns packaging on-demand skill resources.

C:

A description assists skill discovery, but an invocable reusable checklist still needs an implemented skill containing its steps and supporting materials.

Question 303

The test suite is flaky and sporadically fails on unchanged code. What evidence should guide acceptance?

  1. Prioritize analyzing the root cause before attempting another identical patch
  2. Preserve changes bounded to the requested behavior and a small verified diff
  3. Separate reproducible regressions from known nondeterministic failures
  4. Cross-check behavioral acceptance cases rather than treating lint success as completion

Explanation

Correct Answer

C:

A single intermittent result is weak evidence. Repeated controlled runs and baseline comparison help avoid attributing unrelated instability to a patch.

Incorrect Answers

A:

When considering escaping a repetitive unsuccessful repair loop, repeated unmodified retries do not add evidence; the requirement for interpreting flaky test signals correctly remains unmet.

B:

When addressing limiting a patch to the requested behavior, narrow modifications are easier to review, test and revert; the present task instead calls for interpreting flaky test signals correctly.

D:

Linting checks style and some static issues; it does not prove runtime correctness; the unsatisfied requirement is interpreting flaky test signals correctly, not distinguishing static hygiene from functional success.

Question 304

A batch of automated PR reviews must be traceable to the exact commit examined. What metadata should each run record?

  1. Attach the commit SHA, workflow run ID and validated result artifact
  2. Perform Claude Code non-interactively with claude -p in CI
  3. Verify the review payload and apply a documented failure policy
  4. Demand machine-readable JSON output and validate its schema

Explanation

Correct Answer

A:

Immutable revision identifiers distinguish reviews of changed code and support audit trails. Without them, old findings can be applied to newer commits incorrectly.

Incorrect Answers

B:

The print-style command supports scripting without a chat session; this can serve invoking Claude Code in a noninteractive pipeline, but leaves the requirement for binding CI review evidence to a code revision unmet.

C:

CI must distinguish tooling errors from confirmed code defects; that addresses handling invalid AI-review output in a pipeline, whereas this case calls for binding CI review evidence to a code revision.

D:

Structured output is easier for CI to parse and gate deterministically; this concerns parsing review output safely in CI, leaving the requirement for binding CI review evidence to a code revision outstanding.

Question 305

A tool must update an existing YAML key but preserve comments and formatting elsewhere. Which file operation minimizes collateral change?

  1. Employ Read rather than invoking Bash to print the file
  2. Apply a scoped Edit with an exact matching target
  3. Employ Glob for pathname discovery before reading content
  4. Use Write to regenerate and overwrite the complete YAML file after reading the original

Explanation

Correct Answer

B:

A targeted replacement can preserve neighboring configuration content. A whole-file Write may unnecessarily discard formatting or edits made by others.

Incorrect Answers

A:

Reading file content through the dedicated tool avoids an unrestricted shell surface; its benefits for choosing direct file inspection over shell execution do not resolve the need for making a minimal configuration edit.

C:

Path globbing matches file names without requiring content search; this contributes to searching by path rather than file contents without meeting the need for making a minimal configuration edit.

D:

A whole-file rewrite can unnecessarily alter comments, ordering and whitespace. A matching-target Edit constrains the mutation to the requested key while leaving adjacent content intact.

Question 306

A refund worker says 'done' despite an authorization error, and the dashboard counts failed attempts as successes. Which TWO corrections make completion reporting reliable? Choose TWO.

  1. Separate confirmed refunds, failed attempts and still-pending work in dashboard metrics
  2. Retry the authorization-denied refund until the workflow reports success
  3. Treat any returned tool payload, including a failed status, as successful completion
  4. Count every branch as completed once a worker has returned a message, regardless of tool errors
  5. Check the authoritative authorization and execution result before accepting a worker's done claim

Explanation

Correct Answers

A:

This choice addresses distinguishing task completion from task success: progress counters must reflect task outcomes rather than merely finished attempts; clear status reporting prevents operators from mistaking errors for completed deliverables.

E:

Subagent prose should not override an explicit failed execution record. Completion gates must use observed results before telling customers a financial mutation succeeded.

Incorrect Answers

B:

An authorization denial is not evidence of transient failure. Repeated execution can be unsafe, and it cannot justify showing the branch as completed.

C:

Receiving an error response does not mean the refund was authorized or executed. Completion status must come from the actual tool outcome.

D:

Returning text is not the same as accomplishing an authorized action. A dashboard must separate genuine completion from attempted, failed and pending work based on tool state.

Question 307

A few-shot prompt performs well on training examples but poorly on newly written support tickets. What evaluation change is needed?

  1. Test on held-out realistic tickets rather than judging only the shown demonstrations
  2. Add examples where quoted user content is treated as data, not a control instruction
  3. Offer a few labeled input-output examples illustrating the distinction
  4. Implement an example that correctly returns an unknown or escalation category

Explanation

Correct Answer

A:

In-prompt examples are teaching material, not an independent evaluation set. Held-out cases expose overfitting and unforeseen phrasing.

Incorrect Answers

B:

Adversarial examples can illustrate the boundary between source text and system requirements; this addresses teaching separation of quoted data and instructions, while the question is about checking generalization beyond few-shot examples.

C:

Representative examples demonstrate how to resolve ambiguous document fields; that relates to teaching field disambiguation by example, not the necessary checking generalization beyond few-shot examples.

D:

Although boundary examples teach that abstention is acceptable, the choice addresses calibrating abstention with a boundary example and not the requested checking generalization beyond few-shot examples.

Question 308

An MCP tool returns a successful transport response containing an error message as plain data. What field should indicate tool-level failure?

  1. Expose a retryable temporary-failure classification with backoff guidance
  2. Provide a structured invalid-arguments failure with an expected-shape hint
  3. Return a tool result marked isError for the failed operation
  4. Sanitize secrets in public tool errors and retain details in protected logs

Explanation

Correct Answer

C:

MCP distinguishes a tool execution error from a normal successful result with an error flag. Agents can then reason about recovery without mistaking the message for business data.

Incorrect Answers

A:

Transient capacity problems may clear without changing the business request; identifying safely retryable service overload and signaling execution failure in an MCP result require different controls in this situation.

B:

Errors about parsing and schema constraints should identify how the caller can repair the input; the unsatisfied requirement is signaling execution failure in an MCP result, not explaining invalid tool arguments.

D:

When considering sanitizing error information returned to models, agent-visible diagnostics should be actionable without exposing credentials; the requirement for signaling execution failure in an MCP result remains unmet.

Question 309

A model generates a valid object containing customer_id, but the account belongs to another tenant. Which validation belongs outside the output schema?

  1. Label required properties and disallow additional properties in the output schema
  2. Adopt a finite enumeration for the allowed status values
  3. Use semantic evidence checks in addition to schema compliance
  4. Verify tenant authorization and business references against the system of record

Explanation

Correct Answer

D:

JSON types and patterns cannot establish whether an identifier is authorized for the caller. Business-level permission checks must follow structural validation.

Incorrect Answers

A:

Required fields and closed object shapes prevent silently missing or invented keys; that addresses forbidding missing and unexpected JSON fields, whereas this case calls for validating authorization beyond JSON shape.

B:

Enumerated values constrain the output vocabulary and simplify downstream branching; this concerns restricting status output to known categories, leaving the requirement for validating authorization beyond JSON shape outstanding.

C:

A schema constrains representation, not factual truth; distinguishing schema conformance from factual accuracy and validating authorization beyond JSON shape are distinct responsibilities under these constraints.

Question 310

A long-running incident response has multiple branches and one branch fails. Which coordination contract aids partial recovery?

  1. Track each branch's completion state and retry only the failed dependency
  2. Collapse repeated entries in planned search tasks and share a normalized evidence index
  3. Escalate cross-specialty requests through an orchestrator with an explicit task contract
  4. Give analysis-only capabilities to the reviewer and editing tools to a separate implementer
  5. Bound delegation depth and concurrent subagent count at the orchestration layer

Explanation

Correct Answer

A:

Durable per-worker status lets the orchestrator resume a specific failed task. Restarting all successful branches wastes work and may duplicate side effects.

Incorrect Answers

B:

Centralizing a work plan and results catalogue prevents workers from independently repeating identical queries; that serves avoiding redundant delegated searches, whereas the case requires recovering a failed worker without rerunning completed branches.

C:

Because hub-and-spoke mediation controls ownership, context, and failure propagation, the measure supports minimizing unpredictable peer-to-peer dependencies rather than the needed recovering a failed worker without rerunning completed branches.

D:

Tool partitioning enforces role boundaries; its benefits for separating creator and reviewer privileges do not resolve the need for recovering a failed worker without rerunning completed branches.

E:

Limits on nesting and worker breadth bound cost and latency; supporting constraining recursive multi-agent delegation cannot replace the required recovering a failed worker without rerunning completed branches.

Claude Certified Architect - Foundations (CCAR-F) (CCA-F) Practice Tests