Anthropic CCA-F (Claude Certified Architect - Foundations (CCAR-F)) Practice Test 90

Question 890

An architecture review of a structured case-file ingestion workflow finds two missing guarantees: each batch result is matched to its originating record; a machine consumer receives a schema-compliant JSON object. What two-part change closes both gaps?

  1. match returned results to input rows only by array position; and rely only on a sentence requesting JSON with no validation
  2. match returned results to input rows only by array position; and use supported structured JSON output with a declared schema
  3. assign stable custom IDs and map results by those IDs; and rely only on a sentence requesting JSON with no validation
  4. assign stable custom IDs and map results by those IDs; and use supported structured JSON output with a declared schema

Explanation

Correct Answer

D:

In a structured case-file ingestion workflow, asynchronous completion order need not match submission order; alongside this, constrained output enforces parseable structure and required field types at generation, satisfying the distinct requirements that each batch result is matched to its originating record and that a machine consumer receives a schema-compliant JSON object.

Incorrect Answers

A:

In a structured case-file ingestion workflow, out-of-order or failed items can corrupt record association and prompt-only formatting can still produce malformed or incomplete results; the design would fail both conditions: each batch result is matched to its originating record and a machine consumer receives a schema-compliant JSON object.

B:

In a structured case-file ingestion workflow, constrained output enforces parseable structure and required field types at generation, but out-of-order or failed items can corrupt record association; therefore the separate requirement that each batch result is matched to its originating record would remain unsatisfied.

C:

In a structured case-file ingestion workflow, asynchronous completion order need not match submission order, but prompt-only formatting can still produce malformed or incomplete results; therefore the separate requirement that a machine consumer receives a schema-compliant JSON object would remain unsatisfied.

Question 891

While preparing a build verification pipeline for deployment, engineers must ensure that an automation expects a defined JSON result schema; they must also ensure that a common security rule remains active across the project. Which combination best satisfies those requirements?

  1. use the CLI structured-output schema option for the response; and scope the security instruction only to one unrelated test directory
  2. ask for valid JSON in prose but accept any text result; and keep project-wide security guidance in shared instructions rather than a narrow path glob
  3. use the CLI structured-output schema option for the response; and keep project-wide security guidance in shared instructions rather than a narrow path glob
  4. ask for valid JSON in prose but accept any text result; and scope the security instruction only to one unrelated test directory

Explanation

Correct Answer

C:

In a build verification pipeline, schema-constrained output is more robust for downstream integrations than prompt-only formatting; alongside this, cross-cutting obligations should not disappear when an agent changes directories, satisfying the distinct requirements that an automation expects a defined JSON result schema and that a common security rule remains active across the project.

Incorrect Answers

A:

In a build verification pipeline, schema-constrained output is more robust for downstream integrations than prompt-only formatting, but other code paths would lack the universal control; therefore the separate requirement that a common security rule remains active across the project would remain unsatisfied.

B:

In a build verification pipeline, cross-cutting obligations should not disappear when an agent changes directories, but a natural-language instruction does not guarantee parseable or complete fields; therefore the separate requirement that an automation expects a defined JSON result schema would remain unsatisfied.

D:

In a build verification pipeline, a natural-language instruction does not guarantee parseable or complete fields and other code paths would lack the universal control; the design would fail both conditions: an automation expects a defined JSON result schema and a common security rule remains active across the project.

Question 892

In a legacy service decomposition, the acceptance criteria require that a handoff retains only action-relevant context and that the agent executes tests and examines their exit status. Which paired implementation meets both?

  1. pass a compact structured record of facts, decisions, pending tasks, and sources; and use Bash with an approved test command
  2. send every previous chat turn to each new worker; and inspect the test configuration only with Read and claim tests passed
  3. pass a compact structured record of facts, decisions, pending tasks, and sources; and inspect the test configuration only with Read and claim tests passed
  4. send every previous chat turn to each new worker; and use Bash with an approved test command

Explanation

Correct Answer

A:

In a legacy service decomposition, focused handoffs reduce context size while protecting continuity-critical details; alongside this, the shell can run build tools and return execution feedback that Read cannot supply, satisfying the distinct requirements that a handoff retains only action-relevant context and that the agent executes tests and examines their exit status.

Incorrect Answers

B:

In a legacy service decomposition, large histories include irrelevant or outdated information and reading configuration does not execute a test suite; the design would fail both conditions: a handoff retains only action-relevant context and the agent executes tests and examines their exit status.

C:

In a legacy service decomposition, focused handoffs reduce context size while protecting continuity-critical details, but reading configuration does not execute a test suite; therefore the separate requirement that the agent executes tests and examines their exit status would remain unsatisfied.

D:

In a legacy service decomposition, the shell can run build tools and return execution feedback that Read cannot supply, but large histories include irrelevant or outdated information; therefore the separate requirement that a handoff retains only action-relevant context would remain unsatisfied.

Question 893

For a supplier invoice extractor, a reviewer identifies two independent obligations: partial data is not presented as a complete successful result and a corrected attempt preserves a record of prior validation findings. Which proposal should be approved?

  1. return an explicit partial status with missing-item identifiers; and replace all earlier outputs without recording the validation history
  2. silently omit records that failed to load; and replace all earlier outputs without recording the validation history
  3. return an explicit partial status with missing-item identifiers; and retain the original source, response version, and retry feedback
  4. silently omit records that failed to load; and retain the original source, response version, and retry feedback

Explanation

Correct Answer

C:

In a supplier invoice extractor, failure metadata preserves the difference between a sparse valid response and incomplete execution; alongside this, traceable attempts allow auditors to distinguish correction from silent overwrite, satisfying the distinct requirements that partial data is not presented as a complete successful result and that a corrected attempt preserves a record of prior validation findings.

Incorrect Answers

A:

In a supplier invoice extractor, failure metadata preserves the difference between a sparse valid response and incomplete execution, but the reason for a changed value becomes untraceable; therefore the separate requirement that a corrected attempt preserves a record of prior validation findings would remain unsatisfied.

B:

In a supplier invoice extractor, the coordinator could publish an inaccurate completeness claim and the reason for a changed value becomes untraceable; the design would fail both conditions: partial data is not presented as a complete successful result and a corrected attempt preserves a record of prior validation findings.

D:

In a supplier invoice extractor, traceable attempts allow auditors to distinguish correction from silent overwrite, but the coordinator could publish an inaccurate completeness claim; therefore the separate requirement that partial data is not presented as a complete successful result would remain unsatisfied.

Question 894

An architecture review of a build verification pipeline finds two missing guarantees: the repository can review and version its custom workflows; CI failures provide actionable evidence to maintainers. What two-part change closes both gaps?

  1. distribute the canonical skill only through private direct messages; and persist structured errors, relevant logs, and a bounded diagnostic summary
  2. store project skills in the repository's .claude/skills structure; and discard the runner logs and keep only the assistant's final sentence
  3. store project skills in the repository's .claude/skills structure; and persist structured errors, relevant logs, and a bounded diagnostic summary
  4. distribute the canonical skill only through private direct messages; and discard the runner logs and keep only the assistant's final sentence

Explanation

Correct Answer

C:

In a build verification pipeline, version-controlled skills evolve alongside tests and conventions with code review; alongside this, stable artifacts help maintainers reproduce the problem without rerunning a chat, satisfying the distinct requirements that the repository can review and version its custom workflows and that CI failures provide actionable evidence to maintainers.

Incorrect Answers

A:

In a build verification pipeline, stable artifacts help maintainers reproduce the problem without rerunning a chat, but unversioned copies lose the team's approved definition; therefore the separate requirement that the repository can review and version its custom workflows would remain unsatisfied.

B:

In a build verification pipeline, version-controlled skills evolve alongside tests and conventions with code review, but the missing trace impedes diagnosis of command or infrastructure failures; therefore the separate requirement that CI failures provide actionable evidence to maintainers would remain unsatisfied.

D:

In a build verification pipeline, unversioned copies lose the team's approved definition and the missing trace impedes diagnosis of command or infrastructure failures; the design would fail both conditions: the repository can review and version its custom workflows and CI failures provide actionable evidence to maintainers.

Question 895

In a customs paperwork processor, the immediate design requirement is that one agent's draft is reviewed independently for errors. Which change most directly establishes that behavior?

  1. give the reviewer the source and rubric, not only the draft's self-rating
  2. separate factual, structural, and policy review passes
  3. use a separate review pass with explicit criteria
  4. ask the draft generator to declare its own work perfect

Explanation

Correct Answer

C:

For a customs paperwork processor, independent criticism exposes omissions that unstructured self-praise can miss; this directly establishes that one agent's draft is reviewed independently for errors instead of relying on an unrelated workflow control.

Incorrect Answers

A:

Although independent evidence lets a critic challenge unsupported assertions, that only supports the distinct goal that the reviewer does not simply echo the generator's confidence, not the requirement that one agent's draft is reviewed independently for errors in a customs paperwork processor.

B:

Although specialized criteria can detect different failure modes without one overloaded evaluation instruction, that only supports the distinct goal that a complex document receives focused quality checks, not the requirement that one agent's draft is reviewed independently for errors in a customs paperwork processor.

D:

In a customs paperwork processor, self-assessment without a second criterion-based pass is weak evidence; as a result, the stated requirement that one agent's draft is reviewed independently for errors would still not be met.

Question 896

For a dependency upgrade initiative, a reviewer identifies two independent obligations: an existing file receives a precise localized replacement and untrusted project MCP definitions receive approval in interactive Claude Code sessions. Which proposal should be approved?

  1. use Edit for an exact targeted change; and automatically approve every MCP definition found in a cloned repository
  2. overwrite the whole file with a newly generated copy; and review and approve project-scoped server configuration before connecting interactively
  3. use Edit for an exact targeted change; and review and approve project-scoped server configuration before connecting interactively
  4. overwrite the whole file with a newly generated copy; and automatically approve every MCP definition found in a cloned repository

Explanation

Correct Answer

C:

In a dependency upgrade initiative, surgical edits preserve unrelated lines and provide reviewable diffs; alongside this, interactive project trust checks limit execution of unfamiliar server commands and endpoints, satisfying the distinct requirements that an existing file receives a precise localized replacement and that untrusted project MCP definitions receive approval in interactive Claude Code sessions.

Incorrect Answers

A:

In a dependency upgrade initiative, surgical edits preserve unrelated lines and provide reviewable diffs, but repository content can include unexpected executable commands and unattended sessions require explicit policy controls; therefore the separate requirement that untrusted project MCP definitions receive approval in interactive Claude Code sessions would remain unsatisfied.

B:

In a dependency upgrade initiative, interactive project trust checks limit execution of unfamiliar server commands and endpoints, but full replacement risks losing unrelated code and formatting; therefore the separate requirement that an existing file receives a precise localized replacement would remain unsatisfied.

D:

In a dependency upgrade initiative, full replacement risks losing unrelated code and formatting and repository content can include unexpected executable commands and unattended sessions require explicit policy controls; the design would fail both conditions: an existing file receives a precise localized replacement and untrusted project MCP definitions receive approval in interactive Claude Code sessions.

Question 897

For a typed API generator project, a reviewer identifies two independent obligations: reviewer feedback points to a specific missed requirement and a broad tool catalog does not overwhelm a small task. Which proposal should be approved?

  1. repeat the original prompt verbatim and expect a different outcome; and load all available MCP tools into every specialist turn
  2. turn the feedback into a concrete test or change request; and supply only context-relevant tools for the current agent
  3. turn the feedback into a concrete test or change request; and load all available MCP tools into every specialist turn
  4. repeat the original prompt verbatim and expect a different outcome; and supply only context-relevant tools for the current agent

Explanation

Correct Answer

B:

In a typed API generator project, actionable feedback narrows the next edit and makes improvement measurable; alongside this, reducing irrelevant tool definitions improves selection and saves context, satisfying the distinct requirements that reviewer feedback points to a specific missed requirement and that a broad tool catalog does not overwhelm a small task.

Incorrect Answers

A:

In a typed API generator project, unchanged inputs do not target the identified deficiency and tool bloat makes the available actions harder to discriminate; the design would fail both conditions: reviewer feedback points to a specific missed requirement and a broad tool catalog does not overwhelm a small task.

C:

In a typed API generator project, actionable feedback narrows the next edit and makes improvement measurable, but tool bloat makes the available actions harder to discriminate; therefore the separate requirement that a broad tool catalog does not overwhelm a small task would remain unsatisfied.

D:

In a typed API generator project, reducing irrelevant tool definitions improves selection and saves context, but unchanged inputs do not target the identified deficiency; therefore the separate requirement that reviewer feedback points to a specific missed requirement would remain unsatisfied.

Question 898

Before expanding a returns-policy assistant, the owner must demonstrate that only one approved write tool is callable in a sensitive stage, while also demonstrating that a status field can contain only approved business states. Which design provides both controls?

  1. let the model select from a broad global list of tools; and accept any model-generated status description
  2. let the model select from a broad global list of tools; and enforce the approved status enum with structured output and validate it in the consumer
  3. restrict the stage's allowed tool set and verify permissions in the executor; and accept any model-generated status description
  4. restrict the stage's allowed tool set and verify permissions in the executor; and enforce the approved status enum with structured output and validate it in the consumer

Explanation

Correct Answer

D:

In a returns-policy assistant, capability scoping narrows possible side effects before model selection; alongside this, a constrained schema expresses the allowed labels and downstream validation rejects out-of-contract values, satisfying the distinct requirements that only one approved write tool is callable in a sensitive stage and that a status field can contain only approved business states.

Incorrect Answers

A:

In a returns-policy assistant, excess capabilities create avoidable routing and authorization risk and free-form status phrases can violate a fixed business-state contract; the design would fail both conditions: only one approved write tool is callable in a sensitive stage and a status field can contain only approved business states.

B:

In a returns-policy assistant, a constrained schema expresses the allowed labels and downstream validation rejects out-of-contract values, but excess capabilities create avoidable routing and authorization risk; therefore the separate requirement that only one approved write tool is callable in a sensitive stage would remain unsatisfied.

C:

In a returns-policy assistant, capability scoping narrows possible side effects before model selection, but free-form status phrases can violate a fixed business-state contract; therefore the separate requirement that a status field can contain only approved business states would remain unsatisfied.

Question 899

In a technical review of a returns-and-billing assistant, the architect must address the following constraints together: a failed retrieval agent does not masquerade as having found no matches; contradictory evidence is reviewed before action. Which change set is most defensible?

  1. convert every retrieval failure into an empty list; and select the most recent-looking sentence without checking its source
  2. return a typed failure status with its origin to the coordinator; and select the most recent-looking sentence without checking its source
  3. return a typed failure status with its origin to the coordinator; and mark the inconsistency and request resolution or human verification
  4. convert every retrieval failure into an empty list; and mark the inconsistency and request resolution or human verification

Explanation

Correct Answer

C:

In a returns-and-billing assistant, an operational error must remain distinguishable from a valid empty result; alongside this, conflicting source claims are a knowledge gap, not authorization to guess, satisfying the distinct requirements that a failed retrieval agent does not masquerade as having found no matches and that contradictory evidence is reviewed before action.

Incorrect Answers

A:

In a returns-and-billing assistant, the final report could incorrectly claim no sources exist and recency alone cannot establish which contradictory record is correct; the design would fail both conditions: a failed retrieval agent does not masquerade as having found no matches and contradictory evidence is reviewed before action.

B:

In a returns-and-billing assistant, an operational error must remain distinguishable from a valid empty result, but recency alone cannot establish which contradictory record is correct; therefore the separate requirement that contradictory evidence is reviewed before action would remain unsatisfied.

D:

In a returns-and-billing assistant, conflicting source claims are a knowledge gap, not authorization to guess, but the final report could incorrectly claim no sources exist; therefore the separate requirement that a failed retrieval agent does not masquerade as having found no matches would remain unsatisfied.

Question 900

Tests for a code policy conformance job will pass only if partial completion of a multi-step operation is handled safely and if secrets are not embedded in a shared MCP setup. Which proposed design directly satisfies both checks?

  1. implement only inject credentials through approved environment or secret-management mechanisms; make no corresponding change for the first requirement
  2. track individual stage outcomes and use explicit recovery or compensating actions; and inject credentials through approved environment or secret-management mechanisms
  3. treat the first successful tool call as proof that the entire workflow succeeded; and place long-lived access tokens directly in .mcp.json
  4. treat the first successful tool call as proof that the entire workflow succeeded; and inject credentials through approved environment or secret-management mechanisms
  5. track individual stage outcomes and use explicit recovery or compensating actions; and place long-lived access tokens directly in .mcp.json

Explanation

Correct Answer

B:

In a code policy conformance job, an orchestrated workflow must not report overall success while a required step failed; alongside this, separating credentials from versioned configuration limits accidental disclosure, satisfying the distinct requirements that partial completion of a multi-step operation is handled safely and that secrets are not embedded in a shared MCP setup.

Incorrect Answers

A:

In a code policy conformance job, separating credentials from versioned configuration limits accidental disclosure, but the independent mechanism for the other criterion is absent: an orchestrated workflow must not report overall success while a required step failed; as written, the design does not ensure that partial completion of a multi-step operation is handled safely.

C:

In a code policy conformance job, later actions can fail after an earlier side effect and repository history can expose tokens to everyone with code access; the design would fail both conditions: partial completion of a multi-step operation is handled safely and secrets are not embedded in a shared MCP setup.

D:

In a code policy conformance job, separating credentials from versioned configuration limits accidental disclosure, but later actions can fail after an earlier side effect; therefore the separate requirement that partial completion of a multi-step operation is handled safely would remain unsatisfied.

E:

In a code policy conformance job, an orchestrated workflow must not report overall success while a required step failed, 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.

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