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

Question 60

A research task asks for five unrelated competitor profiles and one comparative summary. Which decomposition yields the clearest critical path?

  1. Write the comparison before any profiles have been retrieved
  2. Assign all workers the entire comparison and publish every output
  3. Execute profiles serially despite no rate or resource constraint
  4. Gather profiles in parallel; synthesize after verification

Explanation

Correct Answer

D:

The profiles have no prerequisite dependency on one another. Aggregation waits for their evidence and resolves differences centrally.

Incorrect Answers

A:

Synthesis cannot be grounded without the component findings. This reverses the information dependency.

B:

Duplicating the final task creates redundant answers with no integration owner. Workers should have nonoverlapping responsibilities.

C:

Unnecessary serialization extends latency. Independent read-only subtasks can safely proceed concurrently.

Question 61

A refund flow requires identity validation, balance lookup and then a financial mutation. Which plan respects the dependency structure?

  1. Run mutation and verification in parallel
  2. Split each action into dozens of unbounded worker invocations
  3. Skip balance lookup whenever the customer sounds confident
  4. Verify identity, then choose an action based on account state

Explanation

Correct Answer

D:

A sensitive side effect depends on a confirmed identity and valid account state. The graph should not execute actions before prerequisites are satisfied.

Incorrect Answers

A:

A transfer could occur before identity failure is discovered. Parallelism is unsafe across a mandatory authorization dependency.

B:

Overfragmentation increases coordination cost without improving prerequisite enforcement.

C:

Customer assertions do not establish the backend balance needed for correct payment decisions.

Question 62

A legal-document pipeline has a fixed deterministic conversion step and one ambiguous legal interpretation. Which task should use model judgment?

  1. Convert the document before checking mandatory source integrity
  2. Use bounded agent review but deterministic conversion
  3. Ask a shell script to decide legal intent from syntax alone
  4. Use a large autonomous agent to rename every temporary file

Explanation

Correct Answer

B:

Known transformations should use conventional code while genuinely ambiguous language benefits from bounded reasoning and review.

Incorrect Answers

A:

The plan must preserve prerequisites such as source validation. Irreversible or lossy processing should follow verification.

C:

A simple deterministic converter cannot resolve domain ambiguity without a specified rule. Human or model-assisted interpretation is the reasoning step.

D:

Deterministic housekeeping gains little from model planning and increases failure modes.

Question 63

A coordinator repeatedly spawns workers to evaluate tiny formatting differences. Which redesign best reduces overhead?

  1. Combine related low-risk checks into bounded validation
  2. Remove all formatting checks to eliminate cost
  3. Spawn a worker per punctuation mark
  4. Make every formatting worker independently alter the whole file

Explanation

Correct Answer

A:

Agent setup and context transfer can dominate trivial jobs. Grouping compatible checks improves efficiency while preserving explicit criteria.

Incorrect Answers

B:

The requirement still includes correctness and readability. Reducing overhead need not mean dropping validation.

C:

Excessive granularity multiplies overhead without introducing distinct expertise. A validator can assess many related constraints at once.

D:

Concurrent broad writes create collisions and make faults harder to isolate. Validation and edits should be coordinated.

Question 64

A team's investigation requires fetching evidence, drafting a recommendation and obtaining an independent critique. Which ordering is sound?

  1. Ask all agents to edit the same final draft simultaneously
  2. Critique a draft before the writer has produced it
  3. Gather evidence, draft, then review against requirements
  4. Draft without evidence and ask the critic to make the facts true

Explanation

Correct Answer

C:

The stages form a dependency chain. A critic can identify unsupported claims only after seeing the completed draft and its sources.

Incorrect Answers

A:

Uncoordinated writes can overwrite corrections. A sequential review and revision process is easier to trace.

B:

A reviewer cannot assess an absent artifact. It may review requirements earlier, but the final draft still needs evaluation.

D:

Review cannot convert unsupported statements into reliable findings. Factual retrieval precedes responsible synthesis.

Question 65

A request contains three document classes with different validation requirements. Which decomposition pattern is appropriate?

  1. Use worker completion speed as the document classifier
  2. Route each class to its documented extraction and validation procedure
  3. Ask every parser to process every document and average results
  4. Use one permissive parser and disregard type-specific constraints

Explanation

Correct Answer

B:

Different input types merit specialized paths selected by a routing decision. Each path can reuse precise checks suited to that format.

Incorrect Answers

A:

Runtime performance does not establish document type. Routing should inspect document characteristics and requirements.

C:

Averaging incompatible interpretations is not a validation strategy. Route to the right specialized workflow.

D:

A single unconstrained path can miss legally required fields and silently accept malformed outputs.

Question 66

A multi-agent build system needs a test run after code generation, with revisions only on failures. Which workflow pattern fits?

  1. Iterate with an evaluator, bounded retries and test feedback
  2. Replace tests with the writer's assertion that the code works
  3. Run the release step before tests to reduce latency
  4. Regenerate code forever even after all tests pass

Explanation

Correct Answer

A:

Evaluation generates concrete defects that guide targeted repair. The loop stops on passing criteria or a defined retry limit.

Incorrect Answers

B:

Self-confidence is not executable verification. The evaluator needs an independent signal.

C:

Publication before verification violates the requirement that only passing code be delivered.

D:

Unbounded optimization wastes resources and can regress a working solution. A clear pass condition should terminate the loop.

Question 67

A coordinator must analyze ten independent datasets but can run at most three workers simultaneously. What schedule is most appropriate?

  1. Start all ten workers and hope the platform queues them safely
  2. Run a single worker for all ten even though three are permitted
  3. Use a bounded worker pool of three and aggregate completed results
  4. Have workers recursively spawn more workers beyond the cap

Explanation

Correct Answer

C:

A concurrency cap meets resource limits while still allowing parallel progress. The parent should track outstanding jobs and completeness.

Incorrect Answers

A:

Ignoring the explicit cap can exceed resource quotas or overload dependencies. Capacity needs active orchestration.

B:

Fully serial execution is needlessly slow when independent datasets support bounded parallelism.

D:

Nested spawning can bypass intended accounting and violate resource constraints. Fan-out should remain globally bounded.

Question 68

An assistant has to classify support intent and then call one of four external services. Which process minimizes unnecessary tool use?

  1. Ask each service to classify the request after it executes
  2. Select a service at random and correct mistakes later
  3. Classify the request first and route only to the matching service
  4. Run every external service before classifying the requested intent
  5. Call all four services on every customer message

Explanation

Correct Answer

C:

Routing avoids indiscriminate calls and maps user needs to a narrow action path. The chosen service can still validate required inputs.

Incorrect Answers

A:

Post-action classification is too late when services mutate data. Decide the path before side effects.

B:

Random routing ignores information available from the request. Retries may be expensive or risky.

D:

Early broadcast can trigger irrelevant reads or writes. Routing should choose the necessary interface before executing side effects.

E:

Broadcasting generates irrelevant operations and can create accidental side effects. Only the relevant specialist should run.

Question 69

A report writer spends most of its context on duplicate search results. Which decomposition change addresses the root cause?

  1. Increase generation length so duplicate results get more discussion
  2. Drop all source references before drafting
  3. Deduplicate search hits before giving evidence to the writer
  4. Give every search result its own writing agent without consolidation

Explanation

Correct Answer

C:

Separating retrieval preparation from synthesis reduces context noise. The writer receives diverse supporting sources rather than repeated snippets.

Incorrect Answers

A:

Longer prose amplifies redundant evidence and does not improve selection quality.

B:

Provenance is needed to verify claims. Efficient deduplication should retain citations to unique evidence.

D:

More writers reproduce the duplication at larger scale. A curated aggregation step is still missing.

Question 70

A deterministic XML-to-JSON conversion is tested and stable, but an engineer wants a separate autonomous agent for every field. Which advice is strongest?

  1. Use agents because more model turns necessarily increase accuracy
  2. Ask the coordinator to infer field mappings from unrelated conversation history
  3. Discard the stable converter to make the architecture appear more advanced
  4. Keep the transformation in ordinary code and reserve agents for actual judgment

Explanation

Correct Answer

D:

Known transformations benefit from predictable logic and cheap tests. Agent delegation adds latency, cost and unnecessary uncertainty.

Incorrect Answers

A:

Repeated generations do not guarantee faithful deterministic conversion. They can introduce additional errors.

B:

A stable schema mapping should be explicit, not inferred from noisy context. This would reduce reproducibility.

C:

Complexity is not a goal in itself. A proven deterministic component should be retained when it satisfies the contract.

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