Anthropic CCA-F (Claude Certified Architect - Foundations (CCAR-F)) Practice Test 32

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.

Question 311

A repository wants all developers to share the same audited MCP server definition. Where should the project configuration live?

  1. Prefer supported authenticated transport and credential renewal rather than embedding a fixed secret
  2. Check a reviewed project .mcp.json definition into version control
  3. Restart or reload the affected server and verify its active configuration
  4. Focus on deriving or verify tenant authorization on the server for every tool request
  5. Initiate the MCP server as a local stdio process

Explanation

Correct Answer

B:

Project-scoped MCP configuration can be distributed with the repository. Team review and approval should still govern trust before a server is used.

Incorrect Answers

A:

Remote MCP services need a secure authentication lifecycle; this may support maintaining authenticated MCP transport but cannot alone meet the requirement for sharing an MCP server configuration with a project.

C:

Running sessions may retain existing transport state; a method suited to applying modified MCP server settings does not meet the need for sharing an MCP server configuration with a project.

D:

For enforcing tenant isolation at the MCP backend, user-supplied identifiers are untrusted inputs, not authorization; however, this requirement concerns sharing an MCP server configuration with a project.

E:

Stdio transport connects the client to a child process using local streams; that may serve identifying local stdio MCP transport, but it is no substitute for sharing an MCP server configuration with a project.

Question 312

A conversation becomes dominated by repeated assistant self-explanations instead of fresh evidence. What should be removed first?

  1. Retain claim-to-source mappings alongside the condensed findings
  2. Prune redundant narration while preserving user constraints and observed tool results
  3. Keep the verified case ID and current resolution status as durable fields
  4. Send the minimum relevant schema and task constraints to that specialist

Explanation

Correct Answer

B:

Model-authored commentary has lower evidentiary value than original requirements and verified observations. Context selection should privilege information needed for subsequent decisions.

Incorrect Answers

A:

Compression should preserve provenance links for verifiable conclusions; that relates to maintaining citations during context compression, not the necessary prioritizing authoritative context over redundant narration.

C:

Stable identifiers and status survive conversational paraphrasing; the deciding condition is prioritizing authoritative context over redundant narration, whereas this action concerns preserving stable case identity across turns.

D:

Context selection reduces noise and accidental disclosure; the option would serve minimizing context passed to specialist workers, not prioritizing authoritative context over redundant narration as specified in the stem.

Question 313

A customer-service prompt says 'be helpful' but agents disagree on when refunds are allowed. What prompt improvement is most effective?

  1. Specify explicit eligibility conditions and required evidence for each refund decision
  2. Document the defect categories and require a cited code location for every finding
  3. Mandate tool-confirmed completion before wording any definitive customer promise
  4. Specify severity levels by reproducible impact and required evidence

Explanation

Correct Answer

A:

Concrete criteria make a model's judgment testable against policy. General helpfulness language does not define which orders qualify or which records must be checked first.

Incorrect Answers

B:

When considering preventing unsupported PR review findings, evidence-based criteria distinguish observed defects from speculative criticism; the requirement for translating broad intent into checkable refund rules remains unmet.

C:

When addressing preventing premature promises of side effects, the prompt can distinguish proposed, attempted and confirmed actions; the present task instead calls for translating broad intent into checkable refund rules.

D:

Rubric-based severity promotes consistent triage; scoring review findings against a severity rubric and translating broad intent into checkable refund rules require different controls in this situation.

Question 314

A repository instruction refers to an old framework version after an upgrade. What maintenance step avoids misguided edits?

  1. Store additional directory-specific CLAUDE.md files near the relevant code
  2. Update version-sensitive project instructions alongside the framework migration
  3. Employ the user's personal CLAUDE.md for cross-project preferences
  4. Exclude secrets from instruction text and use secure environment-based credential provisioning

Explanation

Correct Answer

B:

Standing guidance can become stale as dependencies and conventions evolve. Keeping it in the code-review process prevents outdated patterns from propagating.

Incorrect Answers

A:

Scoped instruction files can provide rules for particular subtrees; localizing instructions to code directories and maintaining instruction accuracy after upgrades are distinct responsibilities under these constraints.

C:

User-scoped instructions express developer preferences beyond one repository; this can serve applying personal preferences across projects, but leaves the requirement for maintaining instruction accuracy after upgrades unmet.

D:

Instruction files are often shared or exposed as model context; that addresses protecting credentials from instruction-file exposure, whereas this case calls for maintaining instruction accuracy after upgrades.

Question 315

A coordinator wants two workers to compare hypotheses without either seeing the other's guess. What design enables this?

  1. Deliver condensed relevant findings and source handles rather than full unrelated transcripts
  2. Invoke the two workers in separate contexts and compare at the parent level
  3. Preserve approval for destructive decisions with the coordinator or an authorized human
  4. Create a reviewer with an isolated prompt and the source requirements only

Explanation

Correct Answer

B:

Isolated contexts create genuinely independent assessments. Shared scratchpads can spread one worker's early assumption before evidence has been checked.

Incorrect Answers

A:

Because selecting task-relevant evidence controls token growth without erasing provenance, the measure supports controlling delegated context size rather than the needed obtaining independent subagent hypotheses.

C:

Delegating writing does not imply delegating business authority; supporting retaining authority for irreversible actions cannot replace the required obtaining independent subagent hypotheses.

D:

A clean review context reduces anchoring on the first draft; this contributes to maintaining independence for critique without meeting the need for obtaining independent subagent hypotheses.

Question 316

A request involves retrieving records and applying a rule, but not generating creative text. What decomposition saves cost?

  1. Isolate read-only verification from the authorized state-changing step
  2. Use direct deterministic lookup and rule evaluation for the simple stages
  3. Route one bounded comparison to each worker and merge on a shared schema
  4. Organize closely related low-cost tasks into a meaningful worker assignment

Explanation

Correct Answer

B:

Not every step requires another model or subagent. Separating programmable work from inference reduces latency and makes the result more testable.

Incorrect Answers

A:

Distinct stages expose eligibility decisions and permit human approval before mutations; a method suited to separating validation from side-effecting tasks does not meet the need for separating deterministic steps from inference.

C:

Independent subtasks can run concurrently when they have comparable output contracts; that may serve decomposing parallel independent analyses, but it is no substitute for separating deterministic steps from inference.

D:

Excessive decomposition multiplies setup and context cost; this may support avoiding overfragmentation of delegated work but cannot alone meet the requirement for separating deterministic steps from inference.

Question 317

Reviewers disagree about a compliance finding, but the system records only a majority vote. What should the audit retain?

  1. Preserve individual rationales, evidence and the final adjudication basis
  2. Demand an authenticated human approval before the mutation is dispatched
  3. Link approval to immutable action parameters and require reapproval after material changes
  4. Reveal policy basis, supporting records, and the requested action before approval
  5. Demand evidence and calibration rather than trusting the self-reported confidence number

Explanation

Correct Answer

A:

Disagreement may expose ambiguous policies or missing facts. An audit trail should explain how the final determination was justified rather than hiding minority evidence.

Incorrect Answers

B:

Approval must be checked at the controller boundary rather than inferred from model confidence; the option would serve enforcing human authorization for high-value actions, not retaining review disagreement and adjudication provenance as specified in the stem.

C:

Approval of one transaction should not authorize a modified one; this addresses tying human approval to exact mutation details, while the question is about retaining review disagreement and adjudication provenance.

D:

Meaningful approval requires adequate context to judge eligibility and consequences; that relates to giving human reviewers sufficient decision evidence, not the necessary retaining review disagreement and adjudication provenance.

E:

Although a model's expressed confidence may not correspond to measured probability, the choice addresses avoiding uncalibrated confidence as proof and not the requested retaining review disagreement and adjudication provenance.

Question 318

Claude emits two independent client tool_use blocks in one assistant turn for an order and shipment lookup. What should the dispatcher support?

  1. Match and execute both tool calls, returning their respective results
  2. Report an explicit failed tool result with diagnostic information rather than fabricate data
  3. Deliver the result back into the model loop so Claude can decide the next action
  4. Apply a step budget and a defined fallback when the loop exceeds it

Explanation

Correct Answer

A:

A single model response can request multiple client tools; the host should process each authorized call and retain its unique tool-use identifier so results cannot be crossed.

Incorrect Answers

B:

The tool-error signal should flow into the conversation so the agent can retry, escalate, or communicate limits; surfacing client-tool execution failure and handling parallel client-tool calls require different controls in this situation.

C:

The application does not simply display raw tool output as the final assistant response; the unsatisfied requirement is handling parallel client-tool calls, not resuming model reasoning after a result.

D:

Independent limits on turns, time, and side effects prevent runaway agent cycles; in this scenario, bounding an autonomous agent loop is secondary to handling parallel client-tool calls.

Question 319

A code generator uses a dependency API from a different installed version. What should the repair workflow inspect?

  1. Select the smallest clear change that meets requirements and preserves interfaces
  2. Read the actual dependency version and local interface before rewriting calls
  3. Execute failure-path tests and add explicit handling where warranted
  4. Adopt failing integration output to refine the patch and rerun relevant tests

Explanation

Correct Answer

B:

Version drift can make a plausible call invalid. Grounding edits in installed package metadata and source avoids inventing unsupported parameters.

Incorrect Answers

A:

Simplicity is an engineering constraint: extra layers need a demonstrated benefit; this can serve reviewing unnecessary complexity in generated code, but leaves the requirement for aligning generated code to installed dependency APIs unmet.

C:

Reliable implementation must address realistic negative inputs and dependency errors; this concerns validating negative paths after implementation, leaving the requirement for aligning generated code to installed dependency APIs outstanding.

D:

Iterative development uses tool feedback as evidence; refining code based on integration-test feedback and aligning generated code to installed dependency APIs are distinct responsibilities under these constraints.

Question 320

An MCP tool succeeds from one laptop but fails in CI because an expected environment variable is absent. What should the engineer check?

  1. Harden the transport with TLS and authenticate clients before sensitive tool calls
  2. Test tool exposure, permission filters and server-side authorization
  3. Align server identifier and tool-name changes with client configuration
  4. Verify required environment variables are injected in the CI runtime

Explanation

Correct Answer

D:

An MCP configuration that expands environment references depends on that process's environment. Local developer shell settings do not automatically exist in pipeline runners.

Incorrect Answers

A:

Tool inputs and results may contain confidential operational data; supporting protecting an MCP network connection cannot replace the required supplying MCP configuration variables in CI.

B:

Connectivity alone does not grant every capability; this contributes to diagnosing a connected but unavailable MCP tool without meeting the need for supplying MCP configuration variables in CI.

C:

Clients refer to discovered capabilities using identifiers; that serves migrating MCP tool identifiers safely, whereas the case requires supplying MCP configuration variables in CI.

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