Anthropic CCA-F (Claude Certified Architect - Foundations (CCAR-F)) Practice Test 7

Question 60

A product team decomposes complex Claude tasks into subtasks. A customer email requires one classification label and no external data. What is the most appropriate design choice?

  1. Use an agent loop with step budgets and decisions conditioned on observations
  2. Implement a deterministic pipeline around Claude calls rather than a free-form planner
  3. Call Claude once with a precise classification instruction and output contract
  4. Extract, validate, calculate, then explain in that order
  5. Collect sources concurrently and synthesize once all results pass validation

Explanation

Correct Answer

C:

A single well-scoped task avoids unnecessary agent coordination when no independent dependencies or tools are needed.

Incorrect Answers

A:

Adaptive tool choice is useful for uncertain exploration, provided the host constrains possible actions and termination; the alternative problem occurs when the number of investigative steps cannot be known until initial evidence arrives. The present scenario instead requires the system to use the simplest sufficient execution pattern.

B:

This choice is relevant when a workflow repeats the same fixed five stages for every case; known stable stages are easier to test and audit as code; open-ended planning adds variance without meeting an unmet requirement. Here, however, a customer email requires one classification label and no external data.

D:

It would be appropriate when an analysis must validate extracted figures before calculating a ratio; the arithmetic is only meaningful after source values pass type and completeness checks; dependencies determine stage order. Here, a customer email requires one classification label and no external data, so a different mechanism is required.

E:

Independent lookups can run together, but writing should wait for the join to prevent omissions; it protects against the case where a report must combine three unrelated data sources before drafting, rather than the failure described here: a customer email requires one classification label and no external data.

Question 61

A compliance research coordinator receives output from multiple agents. A fact-checking worker reports a source but not its publication date. Which response is most reliable?

  1. Track artifact versions and rerun only synthesis nodes that depended on the corrected result
  2. Have analysis output the estimate, assumptions, and limits separately
  3. Define canonical units and convert values at the aggregation boundary
  4. Require source URL or ID, publication date, and claim in the worker handoff

Explanation

Correct Answer

D:

A dated provenance record lets the coordinator detect obsolete policies and avoid presenting historical facts as current.

Incorrect Answers

A:

This choice is relevant when a fact-checking worker corrects a statistic after downstream synthesis already used an older result; dependency-aware versioning locates conclusions that rely on outdated evidence, leaving independent completed analyses intact while affected synthesis is rebuilt. Here, however, a fact-checking worker reports a source but not its publication date.

B:

The writing agent can only qualify a number responsibly when its underlying assumptions survive the handoff; it protects against the case where an analyst calculates an estimate but does not state its uncertainty, rather than the failure described here: a fact-checking worker reports a source but not its publication date.

C:

Without unit normalization, a coordinator can compare numerically similar values that represent different quantities; the alternative problem occurs when two evidence workers use different measurement units. The present scenario instead requires the system to preserve time qualifiers for cited evidence.

Question 62

An operations team monitors an agent-based customer workflow in production. An agent makes repeated tool calls but the trace does not show why each was chosen. What should the team do to resolve this particular gap?

  1. Propagate structured error status through handoffs to the final decision layer
  2. Verify the expected task-graph completion states before marking the request successful
  3. Instrument fan-out, join, and worker durations separately
  4. Correlate model turns, tool requests, task IDs, and tool outcomes in structured traces

Explanation

Correct Answer

D:

End-to-end correlation reveals the sequence leading to a failure without relying on a final narrative that may omit intermediate decisions.

Incorrect Answers

A:

Explicit error states prevent a later worker from smoothing a failed dependency into an apparently successful report; the alternative problem occurs when a tool sometimes returns an error that a summarizer conceals. The present scenario instead requires the system to capture the causal chain of agent decisions.

B:

A polished model response is not proof that required delegated checks completed; orchestration state supplies that guarantee; it protects against the case where a final reply is fluent although a required specialist never ran, rather than the failure described here: an agent makes repeated tool calls but the trace does not show why each was chosen.

C:

It would be appropriate when the team wants to compare latency of serial and parallel research designs; per-stage timing shows whether concurrency removes a real bottleneck or merely moves latency to a saturated shared service. Here, an agent makes repeated tool calls but the trace does not show why each was chosen, so a different mechanism is required.

Question 63

A support agent returns structured case-resolution decisions. A refund decision needs a rationale plus a reviewed policy reference. Which control is most directly relevant?

  1. Run policy and permission checks independently of JSON shape
  2. Define typed decision, rationale, and policy-reference fields
  3. Validate nonempty identifiers and documented format requirements
  4. Include conditional reason fields tied to the decision status

Explanation

Correct Answer

B:

A consistent structured record supports auditability and guards against free-form messages that omit why the decision was made.

Incorrect Answers

A:

This choice is relevant when a decision object passes schema validation but requests an unauthorized refund; structured output can be syntactically perfect while semantically unsafe; the execution boundary must reject unauthorized actions. Here, however, a refund decision needs a rationale plus a reviewed policy reference.

C:

It would be appropriate when the response schema accepts any account ID string, including empty values; field-level constraints catch malformed records before downstream calls, although existence and authorization still require service checks. Here, a refund decision needs a rationale plus a reviewed policy reference, so a different mechanism is required.

D:

A yes/no value alone is often insufficient for safe escalation or customer communication when the answer is no; it protects against the case where the model returns a Boolean eligibility flag but omits the reason for ineligibility, rather than the failure described here: a refund decision needs a rationale plus a reviewed policy reference.

Question 64

A support application selects tool-use policies for sensitive operations. A customer asks for current order status, and the service has a read-only order API. How should the workflow be adjusted?

  1. Update the tool registry and descriptions so only the current lookup is eligible
  2. Set tool_choice to none on the request that should not dispatch a tool
  3. Keep the refund capability unavailable or execution-blocked for this state
  4. Expose the scoped order-status read tool to the support agent

Explanation

Correct Answer

D:

A narrow read capability can answer the question while preserving the boundary against unauthorized refunds or edits.

Incorrect Answers

A:

This choice is relevant when an agent repeatedly chooses an obsolete lookup despite a new supported endpoint; retiring ambiguous or obsolete tools attacks the selection problem at its source instead of adding reminders to avoid the old one. Here, however, a customer asks for current order status, and the service has a read-only order API.

B:

It would be appropriate when an audit-only response must not invoke any available tools while explaining a policy decision; the none mode prevents tool calls on that request even when tool definitions are present; later actions still need application-side authorization and controls. Here, a customer asks for current order status, and the service has a read-only order API, so a different mechanism is required.

C:

Tool choice is not the same as permission; the financial action must be gated on verified identity and policy; it protects against the case where a tool can issue refunds but the current conversation has not passed identity verification, rather than the failure described here: a customer asks for current order status, and the service has a read-only order API.

Question 65

A CI pipeline runs Claude Code non-interactively for pull-request review. The build runner has no interactive terminal for answering Claude prompts. Which implementation decision is the best fit?

  1. Set job timeout, token or iteration budgets, and explicit failure behavior
  2. Request a defined JSON output format and validate its schema in the pipeline
  3. Use an independent check or human approval before merging
  4. Run Claude Code in noninteractive print mode with a bounded task prompt
  5. Run in a restricted sandbox with minimal tokens and no secret-bearing privileged steps

Explanation

Correct Answer

D:

A headless execution path avoids UI expectations and lets CI capture the completed response as an artifact. Here it serves the explicit need to use headless invocation suitable for automation without adding a separate unverified inference.

Incorrect Answers

A:

This choice is relevant when a CI task might continue indefinitely under repeated tool calls; CI should fail visibly when budgets are exhausted rather than hang or silently return partial success. Here, however, the build runner has no interactive terminal for answering Claude prompts.

B:

Structured output supports deterministic gating and avoids brittle parsing of free-form review prose; it protects against the case where a downstream CI job must parse the review result reliably, rather than the failure described here: the build runner has no interactive terminal for answering Claude prompts.

C:

A model reviewing its own output is correlated with its mistakes; a distinct gate reduces unchecked self-approval; the alternative problem occurs when the team wants Claude's review to recommend fixes but not approve its own changes. The present scenario instead requires the system to use headless invocation suitable for automation.

E:

It would be appropriate when a pull request from an untrusted fork contains instructions to exfiltrate secrets; pull-request content is untrusted input; prompt instructions cannot substitute for runner isolation or secret scoping. Here, the build runner has no interactive terminal for answering Claude prompts, so a different mechanism is required.

Question 66

A document agent selects tools based on the phase of work. The team is exploring file types before deciding how to parse them. What should the architect specify in the design?

  1. Allow targeted page reads rather than forcing a full-document re-extraction
  2. Make file-type detection available without enabling write or submission tools
  3. Update application state so submission tools reject the cancelled operation
  4. Use a dedicated schema checker on the extracted object

Explanation

Correct Answer

B:

Discovery should determine the appropriate downstream parser without changing the source document or external records.

Incorrect Answers

A:

The narrowest retrieval operation reduces cost and avoids perturbing already verified fields from unrelated pages; the alternative problem occurs when a correction requires retrieving one missing page from a previously parsed document. The present scenario instead requires the system to use read-only metadata inspection first.

C:

This choice is relevant when a user cancels a submission before the commit stage; revocation must be enforced by the host; removing a tool description from a prompt alone cannot cancel queued transactions. Here, however, the team is exploring file types before deciding how to parse them.

D:

Validation checks structure and semantics of existing data; another extraction pass does not directly prove that the object satisfies the contract; it protects against the case where a PDF has already been parsed and needs only schema validation, rather than the failure described here: the team is exploring file types before deciding how to parse them.

Question 67

A multi-agent study summarizes academic papers and compares their claims. Search and synthesis agents disagree over which paper version they cited. Which change most directly addresses this condition?

  1. Create a conflict record containing both definitions and their source contexts
  2. Attach a document identifier and version to evidence handed to the writer
  3. Make methods extraction a prerequisite for the evaluation stage
  4. Cache safe read-only metadata results keyed by source and version

Explanation

Correct Answer

B:

Versioned citations prevent claims from drifting when preprints or revised papers differ in important findings. Here it serves the explicit need to link every claim to a particular source revision without adding a separate unverified inference.

Incorrect Answers

A:

Numerical reconciliation is not meaningful if the measurements refer to different constructs; definitions must be resolved first; the alternative problem occurs when a review worker finds two incompatible definitions of the same metric. The present scenario instead requires the system to link every claim to a particular source revision.

C:

A conclusion without methods context can overstate reliability; dependency ordering preserves critical evaluation evidence; it protects against the case where a paper has a methods section that must be checked before its conclusion is summarized, rather than the failure described here: search and synthesis agents disagree over which paper version they cited.

D:

This choice is relevant when the research workload includes ten identical metadata lookups; reusing verified immutable metadata reduces repeated calls without treating dynamic or personalized search results as permanently correct. Here, however, search and synthesis agents disagree over which paper version they cited.

Question 68

An organization publishes MCP tools for searching internal technical documentation. A tool named query has no indication of which index it searches. What is the most defensible next step?

  1. Validate and constrain readable paths in the tool implementation
  2. Specify its corpus, query shape, and result guarantees in the description
  3. Describe deterministic lookup versus semantic search use cases explicitly
  4. Document optional fields, cursor behavior, and what confidence actually represents

Explanation

Correct Answer

B:

Model selection depends on clear capability metadata; an opaque name leaves the assistant guessing which source will answer the user's question.

Incorrect Answers

A:

It would be appropriate when a file-read tool accepts any path string and can access outside the workspace; descriptions help selection but do not provide security; a server-side allowlist must protect unauthorized files. Here, a tool named query has no indication of which index it searches, so a different mechanism is required.

C:

This choice is relevant when two tools share a capability, but one is optimized for exact identifier lookups; a direct identifier lookup gives precise retrieval when a key is known; embedding search is better for conceptual matching, not exact ID verification. Here, however, a tool named query has no indication of which index it searches.

D:

Typed, well-described results let the agent distinguish absent values from negative findings and paginate without assumptions; the alternative problem occurs when a results object has optional confidence and pagination fields. The present scenario instead requires the system to make tool discoverability self-describing.

Question 69

A Claude Code team shares MCP connections across several repositories. A repository requires a custom issue tracker MCP server not needed anywhere else. Which approach satisfies the stated constraint?

  1. Verify each effective user, project, and local configuration layer
  2. Audit server provenance, tool definitions, and permission prompts before enablement
  3. Inject the secret through a managed environment or credential store
  4. Store the connection in the supported project-scoped configuration

Explanation

Correct Answer

D:

A project-level definition keeps repo-specific capabilities with the project instead of exposing them to unrelated workspaces.

Incorrect Answers

A:

This choice is relevant when a team updates server configuration but different developers still see inconsistent tools; effective tool availability can differ when local overrides or older definitions remain; inspecting the active scope resolves the discrepancy. Here, however, a repository requires a custom issue tracker MCP server not needed anywhere else.

B:

External servers can expose surprising side effects or injected descriptions; installation is not authorization for every capability; the alternative problem occurs when a new MCP server exposes a tool that should not be trusted automatically. The present scenario instead requires the system to limit the server configuration to one project.

C:

It would be appropriate when a secret API token is needed by a configured MCP server; project configuration can be shared and reviewed, but credentials belong in protected storage rather than plaintext repository files. Here, a repository requires a custom issue tracker MCP server not needed anywhere else, so a different mechanism is required.

Question 70

An agent runs a bounded support workflow with multiple optional tool calls. A tool-driven conversation keeps calling the same read-only endpoint without progress. Which mechanism should be selected?

  1. Check cancellation state before dispatching queued tool work
  2. Return a bounded handoff or clarification rather than forcing more autonomous turns
  3. Use rate-aware concurrency limits and jittered backoff
  4. Log authorized tool names, request IDs, outcomes, and decision context
  5. Enforce a maximum tool-call or iteration budget and a fallback outcome

Explanation

Correct Answer

E:

Without a host-side bound, repeatedly proposing the same action can consume resources without resolving the customer's issue.

Incorrect Answers

A:

A cancellation flag in application state can block new operations; telling the model to stop is not sufficient for queued execution; it protects against the case where a user cancels a request while a queued tool operation has not started, rather than the failure described here: a tool-driven conversation keeps calling the same read-only endpoint without progress.

B:

This choice is relevant when a response contains no tool calls but the task is still unresolved and requires human judgment; when prerequisites are missing, another iteration does not create evidence; explicit unresolved state protects correctness and enables review. Here, however, a tool-driven conversation keeps calling the same read-only endpoint without progress.

C:

It would be appropriate when a backend becomes rate limited while many agents share it; application-side scheduling controls aggregate demand; independent immediate retries can amplify throttling across sessions. Here, a tool-driven conversation keeps calling the same read-only endpoint without progress, so a different mechanism is required.

D:

Structured execution logs establish which external action occurred and on what evidence without assuming that model prose is a reliable audit record; the alternative problem occurs when support needs a record of why an agent issued each sensitive call. The present scenario instead requires the system to stop a nonproductive tool loop.

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