Anthropic CCA-F (Claude Certified Architect - Foundations (CCAR-F)) Practice Test 40

Question 390

An MCP tool error contains only 'failed' and the support agent guesses a fix. What response would improve diagnosis?

  1. Include a safe error category and actionable next-step detail
  2. Tag the outcome as uncertain and request reconciliation before replay
  3. Produce a non-retryable validation error with the offending field identified
  4. Emit a tool result marked isError for the failed operation

Explanation

Correct Answer

A:

Structured failure context allows the caller to distinguish bad inputs, authentication problems and outages. A generic failure string invites hallucinated remediation.

Incorrect Answers

B:

A timeout does not prove nonexecution for mutating operations; its benefits for handling an ambiguous post-request timeout do not resolve the need for making tool failure diagnostic rather than opaque.

C:

Input-shape errors persist until arguments change; supporting distinguishing validation faults from transient errors cannot replace the required making tool failure diagnostic rather than opaque.

D:

MCP distinguishes a tool execution error from a normal successful result with an error flag; this contributes to signaling execution failure in an MCP result without meeting the need for making tool failure diagnostic rather than opaque.

Question 391

A long-running synthesis worker exceeds its context budget after receiving every raw research page. What should the parent change?

  1. Provide locale and other required business constraints in each relevant worker brief
  2. Provide the assignment with the precise file paths or retrieved evidence required for the task
  3. Pass condensed relevant findings and source handles rather than full unrelated transcripts
  4. Establish one shared input and output field contract before spawning workers

Explanation

Correct Answer

C:

Selecting task-relevant evidence controls token growth without erasing provenance. Raw parent histories burden specialist context and obscure the actual synthesis question.

Incorrect Answers

A:

Constraints do not reliably propagate by implication across context boundaries; a method suited to propagating shared constraints into tasks does not meet the need for controlling delegated context size.

B:

For passing task-specific context into a subagent, subagents have their own context and should be given task-critical inputs explicitly; however, this requirement concerns controlling delegated context size.

D:

Consistent schemas make aggregation mechanical and reduce misbound identifiers; this may support enforcing a shared worker data contract but cannot alone meet the requirement for controlling delegated context size.

Question 392

A forked coding session passes its test suite but the main branch has changed. What should happen before applying its patch?

  1. Track and compare policy versions before accepting the fork's recommendation
  2. Focus on saving checkpoint or session data in durable storage outside the worker process
  3. Reconcile the fork against the current codebase and rerun relevant checks
  4. Coordinate state-changing updates or use optimistic concurrency on the session

Explanation

Correct Answer

C:

A successful result on an older branch does not prove current compatibility. Merging requires version-aware verification after branch divergence.

Incorrect Answers

A:

A session branch can retain a different configuration snapshot; the deciding condition is validating branch changes before merge, whereas this action concerns detecting configuration drift across forks.

B:

External persistence allows another process to resume execution after a crash; the option would serve surviving worker restarts with durable sessions, not validating branch changes before merge as specified in the stem.

D:

Concurrent writes can produce lost updates and inconsistent tool state; that relates to preventing concurrent session-state conflicts, not the necessary validating branch changes before merge.

Question 393

A data extraction pipeline guesses a missing serial number but marks its result as high confidence. What correction is essential?

  1. Validate that every cited claim has an inspected supporting source
  2. Represent the value as missing with a documented reason and prohibit invented identifiers
  3. Treat cautiously the claim and request stronger corroboration before treating it as established
  4. Monitor source version and effective date for policy-dependent statements

Explanation

Correct Answer

B:

Confidence labels cannot make absent evidence real. A transparent missing-value record prevents downstream systems from treating fabrication as a validated fact.

Incorrect Answers

A:

When addressing checking citations against actually inspected evidence, a plausible citation string is not evidence of source support; the present task instead calls for documenting absence without manufacturing data.

C:

Weak or incomplete evidence warrants explicit uncertainty; the unsatisfied requirement is documenting absence without manufacturing data, not calibrating claims to evidence strength.

D:

Correct citations must identify the applicable revision, not just a title; in this scenario, maintaining revision provenance for policy evidence is secondary to documenting absence without manufacturing data.

Question 394

A validated extraction system should automate easy documents while keeping oversight for costly mistakes. What architecture balances the two?

  1. Call for an authenticated human approval before the mutation is dispatched
  2. Use automatic acceptance for verified low-risk cases and exception-based human review
  3. Call for evidence and calibration rather than trusting the self-reported confidence number
  4. Cross-check escalation outcomes with labeled ground truth and false-positive rates

Explanation

Correct Answer

B:

Hybrid workflows reserve human effort for uncertain or high-impact outcomes. Their acceptance boundaries should be tested against real error costs rather than assumed from fluency.

Incorrect Answers

A:

Approval must be checked at the controller boundary rather than inferred from model confidence; this can serve enforcing human authorization for high-value actions, but leaves the requirement for balancing automation with targeted human oversight unmet.

C:

A model's expressed confidence may not correspond to measured probability; this concerns avoiding uncalibrated confidence as proof, leaving the requirement for balancing automation with targeted human oversight outstanding.

D:

Thresholds should be calibrated empirically, not chosen solely from model self-descriptions; calibrating escalation thresholds against observed outcomes and balancing automation with targeted human oversight are distinct responsibilities under these constraints.

Question 395

A team needs a skill to review infrastructure plans but never apply them. Which definition helps keep the boundary clear?

  1. Specify isolated or forked context for the specialized skill
  2. Describe the skill as plan analysis and deny deployment or mutation tools
  3. Focus on packaging the release workflow as a project skill with clear instructions
  4. Narrow skill tool permissions to the necessary read-oriented operations

Explanation

Correct Answer

B:

Skill language and access control should agree. A review workflow should not possess execution capabilities that could turn suggestions into infrastructure changes.

Incorrect Answers

A:

Context isolation can keep reviewer instructions and intermediate analysis apart from the parent task; supporting isolating a specialized skill's context cannot replace the required separating review skills from deployment skills.

C:

A reusable skill can collect steps, supporting files and tool permissions under one invocable workflow; that serves making a repeatable release workflow reusable, whereas the case requires separating review skills from deployment skills.

D:

Because a skill's allowed tool surface should match its declared job, the measure supports defining least-privilege tools for a skill rather than the needed separating review skills from deployment skills.

Question 396

A data pipeline needs every generated database query to be inspected before execution. Where should the control reside?

  1. Ask a privileged diagnostics component inspect logs and return sanitized findings
  2. Surface separated tool sets or filtered capabilities according to worker identity
  3. Explain each tool's selection criteria and record type in descriptions
  4. Add a pre-execution authorization and query-validation layer at the tool dispatcher

Explanation

Correct Answer

D:

Model-generated arguments are proposals. Dispatch-time validation checks the request's scope and syntax before it reaches the privileged database service.

Incorrect Answers

A:

Dedicated diagnostics isolates sensitive infrastructure data while providing actionable summaries; that may serve segregating diagnostic access from regular agents, but it is no substitute for validating privileged tool calls before execution.

B:

Different trust levels deserve different callable surfaces; this may support partitioning sensitive and ordinary MCP capabilities but cannot alone meet the requirement for validating privileged tool calls before execution.

C:

Tool routing quality depends on distinct metadata; the stated problem requires validating privileged tool calls before execution rather than disambiguating adjacent retrieval tools.

Question 397

A prompt contains ten examples of nearly identical easy receipts but fails on handwritten totals. What change increases coverage?

  1. Provide examples where quoted user content is treated as data, not a control instruction
  2. Verify on held-out realistic tickets rather than judging only the shown demonstrations
  3. Select varied fictional entities and emphasize that examples illustrate structure, not actual values
  4. Select examples representing distinct layouts and difficult ambiguity cases

Explanation

Correct Answer

D:

Example diversity matters more than sheer count. Coverage of edge conditions helps the model generalize to realistic variations rather than memorize one template.

Incorrect Answers

A:

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 choosing representative examples rather than redundant ones.

B:

In-prompt examples are teaching material, not an independent evaluation set; that relates to checking generalization beyond few-shot examples, not the necessary choosing representative examples rather than redundant ones.

C:

Although reusing the same names across all demonstrations can invite copying, the choice addresses preventing sample values from leaking into outputs and not the requested choosing representative examples rather than redundant ones.

Question 398

A compliance reviewer wants evidence that a mutation actually occurred, including the resulting status. Which hook timing provides useful data?

  1. Send back a blocking control result instead of an informational log event
  2. Document ordered responsibilities and nonconflicting hook contracts
  3. Stop closed or suspend the protected action according to explicit policy
  4. Observe the tool completion event and record its returned outcome

Explanation

Correct Answer

D:

Post-execution telemetry can contain the result and status of a tool operation. A pre-execution check sees proposed arguments but not the actual business-system response.

Incorrect Answers

A:

Emitting telemetry alone does not enforce permission decisions; the unsatisfied requirement is auditing tool execution outcomes, not distinguishing logging from access denial.

B:

Clear handler scope and explicit failure precedence prevent accidental interactions; in this scenario, resolving conflicting lifecycle handlers is secondary to auditing tool execution outcomes.

C:

When considering handling a failed mandatory policy check, when the decision service is unavailable, a high-risk action should not silently proceed; the requirement for auditing tool execution outcomes remains unmet.

Question 399

A research system needs a single owner for specialist assignments and must merge only verified, evidence-tagged findings. Which TWO architecture choices achieve these aims? Choose TWO.

  1. Give a single coordinator ownership of bounded specialist assignments and their integration
  2. Accept only provenance-tagged findings while preserving uncertainty during synthesis
  3. Set a ceiling on delegation depth and concurrent subagent count at the orchestration layer
  4. Merge repeated records from planned search tasks and share a normalized evidence index

Explanation

Correct Answers

A:

A central orchestrator can define evidence requirements, route work by specialty and reconcile returned results. Independent peer-to-peer improvisation would not supply one decision point for consistency.

B:

Separating verified worker findings from guesses matters in this scenario: outputs from specialists are proposals until verified; A structured reconciliation pass can preserve claims, citations and confidence boundaries rather than laundering speculation into certainty.

Incorrect Answers

C:

Limits on nesting and worker breadth bound cost and latency; the action supports constraining recursive multi-agent delegation but falls short of centralized multi-agent coordination in this case.

D:

Centralizing a work plan and results catalogue prevents workers from independently repeating identical queries; this can serve avoiding redundant delegated searches, but leaves the requirement for centralized multi-agent coordination unmet.

Question 400

A customer lookup proposes an input value of the wrong JSON type even though the tool contract specifies a string. Where must input correctness be enforced?

  1. Validate generated arguments against the declared tool input schema before running business logic
  2. Deliver an explicit failed tool result with diagnostic information rather than fabricate data
  3. Couple bounded retries with an operation idempotency key and outcome lookup
  4. Hand off the refusal through the approved response and escalation policy

Explanation

Correct Answer

A:

Tool schemas describe the callable interface, but application-side validation still protects execution paths and prevents malformed or unsafe arguments from reaching external systems.

Incorrect Answers

B:

Because the tool-error signal should flow into the conversation so the agent can retry, escalate, or communicate limits, the measure supports surfacing client-tool execution failure rather than the needed checking tool input validity.

C:

Retries of side-effecting actions can duplicate work when acknowledgment is lost; its benefits for safe recovery after an ambiguous tool timeout do not resolve the need for checking tool input validity.

D:

A refusal is a distinct termination state; supporting handling refusal as a distinct state cannot replace the required checking tool input validity.

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