Anthropic CCA-F (Claude Certified Architect - Foundations (CCAR-F)) Practice Test 36

Question 350

An audit of an order-status triage agent identifies two unmet requirements: temporary service faults can be retried selectively and a report separates documented findings from its own inference. Select TWO actions that remedy those particular deficiencies.

  1. label direct source facts separately from analytical conclusions
  2. include a retryable classification and bounded retry advice
  3. retry every error with the same request indefinitely
  4. present every deduction as a direct source quotation

Explanation

Correct Answers

A:

For an order-status triage agent, distinguishing evidence from interpretation prevents inference being misattributed to a document, which directly establishes that a report separates documented findings from its own inference; the separate requirement that temporary service faults can be retried selectively is handled by a different action.

B:

For an order-status triage agent, typed transient failures allow backoff without treating permanent errors the same way, which directly establishes that temporary service faults can be retried selectively; the separate requirement that a report separates documented findings from its own inference is handled by a different action.

Incorrect Answers

C:

For an order-status triage agent, permission and validation failures will not heal through repetition, so this option cannot establish that temporary service faults can be retried selectively; correctly addressing a report separates documented findings from its own inference elsewhere does not repair that deficiency.

D:

For an order-status triage agent, unsupported attribution misrepresents what the source actually says, so this option cannot establish that a report separates documented findings from its own inference; correctly addressing temporary service faults can be retried selectively elsewhere does not repair that deficiency.

Question 351

When designing a backend API maintenance team, the requirement is not satisfied unless a high-risk workflow only invokes specifically approved capabilities and secrets are not embedded in a shared MCP setup. Which implementation addresses the two independent conditions?

  1. rely on a sentence asking the agent not to use risky tools; and place long-lived access tokens directly in .mcp.json
  2. declare allowed-tools for the skill and enforce executor permissions; and place long-lived access tokens directly in .mcp.json
  3. declare allowed-tools for the skill and enforce executor permissions; and inject credentials through approved environment or secret-management mechanisms
  4. rely on a sentence asking the agent not to use risky tools; and inject credentials through approved environment or secret-management mechanisms

Explanation

Correct Answer

C:

In a backend API maintenance team, tool restrictions narrow the agent's operational capability while the skill runs; alongside this, separating credentials from versioned configuration limits accidental disclosure, satisfying the distinct requirements that a high-risk workflow only invokes specifically approved capabilities and that secrets are not embedded in a shared MCP setup.

Incorrect Answers

A:

In a backend API maintenance team, instruction-only boundaries cannot substitute for actual tool authorization and repository history can expose tokens to everyone with code access; the design would fail both conditions: a high-risk workflow only invokes specifically approved capabilities and secrets are not embedded in a shared MCP setup.

B:

In a backend API maintenance team, tool restrictions narrow the agent's operational capability while the skill runs, but repository history can expose tokens to everyone with code access; therefore the separate requirement that secrets are not embedded in a shared MCP setup would remain unsatisfied.

D:

In a backend API maintenance team, separating credentials from versioned configuration limits accidental disclosure, but instruction-only boundaries cannot substitute for actual tool authorization; therefore the separate requirement that a high-risk workflow only invokes specifically approved capabilities would remain unsatisfied.

Question 352

Production readiness for an unfamiliar monorepo investigation depends on ensuring that a delegated reviewer has no production write capability and that path-scoped rules are loaded for all intended source directories. Select the most appropriate combined approach.

  1. provide only the read-oriented tools required by that reviewer; and update the rule's paths glob to include the actual target paths
  2. give the reviewer every administrative tool and ask for restraint; and add more prose telling the model to remember the missing rule
  3. provide only the read-oriented tools required by that reviewer; and add more prose telling the model to remember the missing rule
  4. give the reviewer every administrative tool and ask for restraint; and update the rule's paths glob to include the actual target paths

Explanation

Correct Answer

A:

In an unfamiliar monorepo investigation, agent-scoped tool exposure prevents a write path in a read-only role; alongside this, correct glob coverage determines whether conditional guidance is in context, satisfying the distinct requirements that a delegated reviewer has no production write capability and that path-scoped rules are loaded for all intended source directories.

Incorrect Answers

B:

In an unfamiliar monorepo investigation, prompt restraint is weaker than withholding execution capability and instructions outside the matched scope still will not be loaded; the design would fail both conditions: a delegated reviewer has no production write capability and path-scoped rules are loaded for all intended source directories.

C:

In an unfamiliar monorepo investigation, agent-scoped tool exposure prevents a write path in a read-only role, but instructions outside the matched scope still will not be loaded; therefore the separate requirement that path-scoped rules are loaded for all intended source directories would remain unsatisfied.

D:

In an unfamiliar monorepo investigation, correct glob coverage determines whether conditional guidance is in context, but prompt restraint is weaker than withholding execution capability; therefore the separate requirement that a delegated reviewer has no production write capability would remain unsatisfied.

Question 353

Tests for a research PDF metadata pipeline will pass only if an unusual null-value case is handled without inventing data and if a user-facing chat answer returns with low latency. Which proposed design directly satisfies both checks?

  1. include a demonstration showing how an absent field is represented; and use synchronous Messages requests for the interactive path
  2. show only records where every field is present; and submit the one message to an offline batch and block for completion
  3. show only records where every field is present; and use synchronous Messages requests for the interactive path
  4. include a demonstration showing how an absent field is represented; and submit the one message to an offline batch and block for completion

Explanation

Correct Answer

A:

In a research PDF metadata pipeline, explicit missing-value examples clarify that omissions are not permission to hallucinate; alongside this, the caller needs an immediate result rather than an asynchronously retrieved batch output, satisfying the distinct requirements that an unusual null-value case is handled without inventing data and that a user-facing chat answer returns with low latency.

Incorrect Answers

B:

In a research PDF metadata pipeline, the model lacks a model of the required missing-data behavior and batch scheduling is a poor fit for a live conversational deadline; the design would fail both conditions: an unusual null-value case is handled without inventing data and a user-facing chat answer returns with low latency.

C:

In a research PDF metadata pipeline, the caller needs an immediate result rather than an asynchronously retrieved batch output, but the model lacks a model of the required missing-data behavior; therefore the separate requirement that an unusual null-value case is handled without inventing data would remain unsatisfied.

D:

In a research PDF metadata pipeline, explicit missing-value examples clarify that omissions are not permission to hallucinate, but batch scheduling is a poor fit for a live conversational deadline; therefore the separate requirement that a user-facing chat answer returns with low latency would remain unsatisfied.

Question 354

During review of a repository maintenance copilot, an unmet criterion states that workload growth does not cause uncontrolled parallel fan-out. Which measure resolves this specific gap?

  1. launch all possible child tasks at once without resource controls
  2. divide work along independent outputs and run those stages in parallel
  3. use one focused tool-assisted agent for the bounded task
  4. limit concurrency and schedule independent jobs against execution budgets

Explanation

Correct Answer

D:

For a repository maintenance copilot, bounded orchestration balances throughput with tool quotas and downstream service capacity; this directly establishes that workload growth does not cause uncontrolled parallel fan-out instead of relying on an unrelated workflow control.

Incorrect Answers

A:

In a repository maintenance copilot, unbounded fan-out can overwhelm tools and blow through cost budgets; as a result, the stated requirement that workload growth does not cause uncontrolled parallel fan-out would still not be met.

B:

Although dependency-aware decomposition avoids serial work when branches do not consume one another, that only supports the distinct goal that independent stages of a complex job run efficiently, not the requirement that workload growth does not cause uncontrolled parallel fan-out in a repository maintenance copilot.

C:

Although orchestration overhead is not justified when there is no useful specialization or concurrency, that only supports the distinct goal that a straightforward lookup avoids unnecessary agent coordination, not the requirement that workload growth does not cause uncontrolled parallel fan-out in a repository maintenance copilot.

Question 355

The release checklist for a warranty claims assistant explicitly checks that client-side tool requests trigger another model turn after execution; it also checks that an unresolved claim is qualified without inventing certainty. What coordinated solution meets the checklist?

  1. end the loop whenever assistant text appears; and state a firm conclusion because the report needs a complete ending
  2. end the loop whenever assistant text appears; and express the uncertainty and identify what verification is missing
  3. branch on stop_reason tool_use and execute returned tool_use blocks; and express the uncertainty and identify what verification is missing
  4. branch on stop_reason tool_use and execute returned tool_use blocks; and state a firm conclusion because the report needs a complete ending

Explanation

Correct Answer

C:

In a warranty claims assistant, the response explicitly signals pending client execution and the loop must provide results before another model decision; alongside this, calibrated language communicates a bounded evidence gap honestly, satisfying the distinct requirements that client-side tool requests trigger another model turn after execution and that an unresolved claim is qualified without inventing certainty.

Incorrect Answers

A:

In a warranty claims assistant, text may accompany a tool request and is not the completion signal and a polished narrative cannot supply missing corroboration; the design would fail both conditions: client-side tool requests trigger another model turn after execution and an unresolved claim is qualified without inventing certainty.

B:

In a warranty claims assistant, calibrated language communicates a bounded evidence gap honestly, but text may accompany a tool request and is not the completion signal; therefore the separate requirement that client-side tool requests trigger another model turn after execution would remain unsatisfied.

D:

In a warranty claims assistant, the response explicitly signals pending client execution and the loop must provide results before another model decision, but a polished narrative cannot supply missing corroboration; therefore the separate requirement that an unresolved claim is qualified without inventing certainty would remain unsatisfied.

Question 356

An incident postmortem for a source-navigation tool for new hires requires controls so that separate code modules can be inspected in parallel and that a conversational task does not invoke tools needlessly. Which paired remedy covers both failure modes?

  1. make the second specialist wait for the first even though their modules are independent; and force a tool on every message regardless of intent
  2. spawn independent specialists in the same delegation stage; and force a tool on every message regardless of intent
  3. spawn independent specialists in the same delegation stage; and use automatic tool choice when a call is optional
  4. make the second specialist wait for the first even though their modules are independent; and use automatic tool choice when a call is optional

Explanation

Correct Answer

C:

In a source-navigation tool for new hires, concurrent subagents can examine non-overlapping modules before the coordinator merges their reports; alongside this, auto allows the model to answer directly when no external side effect or fresh data is needed, satisfying the distinct requirements that separate code modules can be inspected in parallel and that a conversational task does not invoke tools needlessly.

Incorrect Answers

A:

In a source-navigation tool for new hires, serialization wastes time without supplying a dependency and mandatory calls add latency and can invoke irrelevant operations; the design would fail both conditions: separate code modules can be inspected in parallel and a conversational task does not invoke tools needlessly.

B:

In a source-navigation tool for new hires, concurrent subagents can examine non-overlapping modules before the coordinator merges their reports, but mandatory calls add latency and can invoke irrelevant operations; therefore the separate requirement that a conversational task does not invoke tools needlessly would remain unsatisfied.

D:

In a source-navigation tool for new hires, auto allows the model to answer directly when no external side effect or fresh data is needed, but serialization wastes time without supplying a dependency; therefore the separate requirement that separate code modules can be inspected in parallel would remain unsatisfied.

Question 357

When designing a research PDF metadata pipeline, the requirement is not satisfied unless a partially failed batch is repaired efficiently and the agent can recover exact policy facts after compaction. Which implementation addresses the two independent conditions?

  1. identify failed records and resubmit only safe retry candidates; and retrieve authoritative facts from a durable store when needed
  2. rerun the entire completed batch whenever one record fails; and treat the compressed summary as an immutable source of truth
  3. rerun the entire completed batch whenever one record fails; and retrieve authoritative facts from a durable store when needed
  4. identify failed records and resubmit only safe retry candidates; and treat the compressed summary as an immutable source of truth

Explanation

Correct Answer

A:

In a research PDF metadata pipeline, selective retry avoids paying again for already successful items; alongside this, re-reading source state restores precision that a condensed history cannot guarantee, satisfying the distinct requirements that a partially failed batch is repaired efficiently and that the agent can recover exact policy facts after compaction.

Incorrect Answers

B:

In a research PDF metadata pipeline, successful work would be duplicated unnecessarily and summaries may simplify caveats or omit an exception; the design would fail both conditions: a partially failed batch is repaired efficiently and the agent can recover exact policy facts after compaction.

C:

In a research PDF metadata pipeline, re-reading source state restores precision that a condensed history cannot guarantee, but successful work would be duplicated unnecessarily; therefore the separate requirement that a partially failed batch is repaired efficiently would remain unsatisfied.

D:

In a research PDF metadata pipeline, selective retry avoids paying again for already successful items, but summaries may simplify caveats or omit an exception; therefore the separate requirement that the agent can recover exact policy facts after compaction would remain unsatisfied.

Question 358

For a vendor onboarding document pipeline, a reviewer identifies two independent obligations: an uncertain high-impact decision is routed for review instead of automatic execution and a handoff retains only action-relevant context. Which proposal should be approved?

  1. proceed whenever the assistant's prose sounds certain; and pass a compact structured record of facts, decisions, pending tasks, and sources
  2. set a clear evidence or confidence boundary and require review below it; and pass a compact structured record of facts, decisions, pending tasks, and sources
  3. proceed whenever the assistant's prose sounds certain; and send every previous chat turn to each new worker
  4. set a clear evidence or confidence boundary and require review below it; and send every previous chat turn to each new worker

Explanation

Correct Answer

B:

In a vendor onboarding document pipeline, risk-weighted gates limit autonomous actions when support is inadequate; alongside this, focused handoffs reduce context size while protecting continuity-critical details, satisfying the distinct requirements that an uncertain high-impact decision is routed for review instead of automatic execution and that a handoff retains only action-relevant context.

Incorrect Answers

A:

In a vendor onboarding document pipeline, focused handoffs reduce context size while protecting continuity-critical details, but fluent language is not a calibrated reliability measure; therefore the separate requirement that an uncertain high-impact decision is routed for review instead of automatic execution would remain unsatisfied.

C:

In a vendor onboarding document pipeline, fluent language is not a calibrated reliability measure and large histories include irrelevant or outdated information; the design would fail both conditions: an uncertain high-impact decision is routed for review instead of automatic execution and a handoff retains only action-relevant context.

D:

In a vendor onboarding document pipeline, risk-weighted gates limit autonomous actions when support is inadequate, but large histories include irrelevant or outdated information; therefore the separate requirement that a handoff retains only action-relevant context would remain unsatisfied.

Question 359

Tests for an automated CI review service will pass only if the team obtains a read-only survey of an unfamiliar module and if the reviewer does not simply echo the generator's confidence. Which proposed design directly satisfies both checks?

  1. run write tools to discover how the module behaves; and provide the generator's score as the sole review criterion
  2. investigate in plan-oriented or read-only mode first; and provide the generator's score as the sole review criterion
  3. run write tools to discover how the module behaves; and give the reviewer the source and rubric, not only the draft's self-rating
  4. investigate in plan-oriented or read-only mode first; and give the reviewer the source and rubric, not only the draft's self-rating

Explanation

Correct Answer

D:

In an automated CI review service, inspection can establish architecture and change scope without side effects; alongside this, independent evidence lets a critic challenge unsupported assertions, satisfying the distinct requirements that the team obtains a read-only survey of an unfamiliar module and that the reviewer does not simply echo the generator's confidence.

Incorrect Answers

A:

In an automated CI review service, mutation is unnecessary for gaining initial understanding and anchoring on a self-assigned grade hides actual defects; the design would fail both conditions: the team obtains a read-only survey of an unfamiliar module and the reviewer does not simply echo the generator's confidence.

B:

In an automated CI review service, inspection can establish architecture and change scope without side effects, but anchoring on a self-assigned grade hides actual defects; therefore the separate requirement that the reviewer does not simply echo the generator's confidence would remain unsatisfied.

C:

In an automated CI review service, independent evidence lets a critic challenge unsupported assertions, but mutation is unnecessary for gaining initial understanding; therefore the separate requirement that the team obtains a read-only survey of an unfamiliar module would remain unsatisfied.

Question 360

The release gate for a financial statement data extractor requires that an extractor always returns the identifiers required by downstream code; it also requires that a classification uses a limited approved category set. Which TWO actions meet the specified checks? Choose TWO.

  1. define precise output labels and mapping rules
  2. declare mandatory schema fields in the required list
  3. allow a free-form category for each case
  4. treat every identifier as an optional free-form note

Explanation

Correct Answers

A:

For a financial statement data extractor, controlled categories reduce schema drift and make downstream analytics comparable, which directly establishes that a classification uses a limited approved category set; the separate requirement that an extractor always returns the identifiers required by downstream code is handled by a different action.

B:

For a financial statement data extractor, required properties prevent accidental omission of key attributes, which directly establishes that an extractor always returns the identifiers required by downstream code; the separate requirement that a classification uses a limited approved category set is handled by a different action.

Incorrect Answers

C:

For a financial statement data extractor, unbounded labels fragment the intended taxonomy, so this option cannot establish that a classification uses a limited approved category set; correctly addressing an extractor always returns the identifiers required by downstream code elsewhere does not repair that deficiency.

D:

For a financial statement data extractor, the consumer cannot reliably rely on a field that may be omitted, so this option cannot establish that an extractor always returns the identifiers required by downstream code; correctly addressing a classification uses a limited approved category set elsewhere does not repair that deficiency.

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