Anthropic CCA-F (Claude Certified Architect - Foundations (CCAR-F)) Practice Test 45

Question 440

In a continuous integration repair assistant, the design must ensure that a conclusion supported by model agreement is also tested against source evidence and that tool output is normalized before it enters the agent context. Choose TWO independent changes that directly meet these two requirements.

  1. use a PostToolUse processing hook to standardize the returned data
  2. preprocess only the original user message
  3. promote agreement alone to a guaranteed correctness signal
  4. require source and business-rule checks regardless of model agreement

Explanation

Correct Answers

A:

For a continuous integration repair assistant, post-execution interception can clean result fields and attach useful status without changing the model's tool selection, which directly establishes that tool output is normalized before it enters the agent context; the separate requirement that a conclusion supported by model agreement is also tested against source evidence is handled by a different action.

D:

For a continuous integration repair assistant, correlated model errors can survive consensus without external verification, which directly establishes that a conclusion supported by model agreement is also tested against source evidence; the separate requirement that tool output is normalized before it enters the agent context is handled by a different action.

Incorrect Answers

B:

For a continuous integration repair assistant, input normalization does not correct inconsistent tool outputs, so this option cannot establish that tool output is normalized before it enters the agent context; correctly addressing a conclusion supported by model agreement is also tested against source evidence elsewhere does not repair that deficiency.

C:

For a continuous integration repair assistant, shared biases can produce identical wrong conclusions, so this option cannot establish that a conclusion supported by model agreement is also tested against source evidence; correctly addressing tool output is normalized before it enters the agent context elsewhere does not repair that deficiency.

Question 441

While preparing a subscription support service for deployment, engineers must ensure that an order adjustment uses the current order record; they must also ensure that partial data is not presented as a complete successful result. Which combination best satisfies those requirements?

  1. require the order lookup result before calculating or applying a change; and return an explicit partial status with missing-item identifiers
  2. issue the update using details inferred only from conversation; and return an explicit partial status with missing-item identifiers
  3. require the order lookup result before calculating or applying a change; and silently omit records that failed to load
  4. issue the update using details inferred only from conversation; and silently omit records that failed to load

Explanation

Correct Answer

A:

In a subscription support service, the dependent operation needs authoritative state and must not guess from the customer's description; alongside this, failure metadata preserves the difference between a sparse valid response and incomplete execution, satisfying the distinct requirements that an order adjustment uses the current order record and that partial data is not presented as a complete successful result.

Incorrect Answers

B:

In a subscription support service, failure metadata preserves the difference between a sparse valid response and incomplete execution, but unverified user text can reference the wrong order or outdated state; therefore the separate requirement that an order adjustment uses the current order record would remain unsatisfied.

C:

In a subscription support service, the dependent operation needs authoritative state and must not guess from the customer's description, 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.

D:

In a subscription support service, unverified user text can reference the wrong order or outdated state and the coordinator could publish an inaccurate completeness claim; the design would fail both conditions: an order adjustment uses the current order record and partial data is not presented as a complete successful result.

Question 442

An audit of a supplier invoice extractor identifies two unmet requirements: several requested client tools are returned in one response and a classification uses a limited approved category set. Select TWO actions that remedy those particular deficiencies.

  1. execute every tool_use block and return a result with each corresponding ID
  2. allow a free-form category for each case
  3. define precise output labels and mapping rules
  4. execute only the first returned tool call

Explanation

Correct Answers

A:

For a supplier invoice extractor, a multi-tool assistant turn can request independent actions and must receive all matching outcomes, which directly establishes that several requested client tools are returned in one response; the separate requirement that a classification uses a limited approved category set is handled by a different action.

C:

For a supplier invoice 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 several requested client tools are returned in one response is handled by a different action.

Incorrect Answers

B:

For a supplier invoice extractor, unbounded labels fragment the intended taxonomy, so this option cannot establish that a classification uses a limited approved category set; correctly addressing several requested client tools are returned in one response elsewhere does not repair that deficiency.

D:

For a supplier invoice extractor, the unanswered tool uses remain unresolved and later reasoning loses evidence, so this option cannot establish that several requested client tools are returned in one response; correctly addressing a classification uses a limited approved category set elsewhere does not repair that deficiency.

Question 443

A readiness review for a compliance interpretation research team requires that a delegated agent receives the background it needs to act; that dependent analysis begins only when source retrieval is complete; and that a conversational task does not invoke tools needlessly. Which THREE actions directly implement those distinct requirements? Choose THREE.

  1. force a tool on every message regardless of intent
  2. pass the necessary task context and references explicitly at invocation
  3. start the analysis and retrieval agents together without shared results
  4. run retrieval before analysis and pass its validated artifacts
  5. use automatic tool choice when a call is optional
  6. assume the child agent sees all previous coordinator turns

Explanation

Correct Answers

B:

For a compliance interpretation research team, subagents do not automatically inherit the coordinator's entire conversational history, which directly establishes that a delegated agent receives the background it needs to act; the separate requirement that dependent analysis begins only when source retrieval is complete is handled by a different action.

D:

For a compliance interpretation research team, a downstream agent needs actual upstream findings rather than an assumed future result, which directly establishes that dependent analysis begins only when source retrieval is complete; the separate requirement that a conversational task does not invoke tools needlessly is handled by a different action.

E:

For a compliance interpretation research team, auto allows the model to answer directly when no external side effect or fresh data is needed, which directly establishes that a conversational task does not invoke tools needlessly; the separate requirement that a delegated agent receives the background it needs to act is handled by a different action.

Incorrect Answers

A:

For a compliance interpretation research team, mandatory calls add latency and can invoke irrelevant operations, so this option cannot establish that a conversational task does not invoke tools needlessly; correctly addressing a delegated agent receives the background it needs to act elsewhere does not repair that deficiency.

C:

For a compliance interpretation research team, the analysis branch cannot evaluate evidence it has not received, so this option cannot establish that dependent analysis begins only when source retrieval is complete; correctly addressing a conversational task does not invoke tools needlessly elsewhere does not repair that deficiency.

F:

For a compliance interpretation research team, implicit context inheritance makes delegation fragile and incomplete, so this option cannot establish that a delegated agent receives the background it needs to act; correctly addressing dependent analysis begins only when source retrieval is complete elsewhere does not repair that deficiency.

Question 444

In a supplier invoice extractor, the engineering team must guarantee that context reduction does not erase unresolved obligations. What is the most appropriate control?

  1. track pending approvals and required next steps in durable task state
  2. mark verified, user-reported, and unknown state explicitly
  3. retrieve authoritative facts from a durable store when needed
  4. assume the agent will recall every open action from memory

Explanation

Correct Answer

A:

For a supplier invoice extractor, persisted obligations survive when natural-language history is shortened; this directly establishes that context reduction does not erase unresolved obligations instead of relying on an unrelated workflow control.

Incorrect Answers

B:

Although epistemic labels stop a tentative suggestion becoming a confirmed action parameter, that only supports the distinct goal that a long workflow preserves confirmed facts separately from speculation, not the requirement that context reduction does not erase unresolved obligations in a supplier invoice extractor.

C:

Although re-reading source state restores precision that a condensed history cannot guarantee, that only supports the distinct goal that the agent can recover exact policy facts after compaction, not the requirement that context reduction does not erase unresolved obligations in a supplier invoice extractor.

D:

In a supplier invoice extractor, window truncation can drop an outstanding approval or escalation; as a result, the stated requirement that context reduction does not erase unresolved obligations would still not be met.

Question 445

The release checklist for an unfamiliar monorepo investigation explicitly checks that a large research job combines specialized findings into one answer; it also checks that a truncated response is not mistaken for successful completion. What coordinated solution meets the checklist?

  1. have all subagents independently publish the final answer; and interpret every non-tool stop reason as a successful final answer
  2. use a coordinator that delegates focused tasks and owns final synthesis; and interpret every non-tool stop reason as a successful final answer
  3. use a coordinator that delegates focused tasks and owns final synthesis; and handle max_tokens as truncation rather than end_turn
  4. have all subagents independently publish the final answer; and handle max_tokens as truncation rather than end_turn

Explanation

Correct Answer

C:

In an unfamiliar monorepo investigation, the hub can integrate distinct subagent outputs and resolve overlap without peer-to-peer chatter; alongside this, reaching an output limit is a stop condition but does not establish task completion, satisfying the distinct requirements that a large research job combines specialized findings into one answer and that a truncated response is not mistaken for successful completion.

Incorrect Answers

A:

In an unfamiliar monorepo investigation, independent final outputs offer no single point to reconcile contradictions and a length-limited result may be incomplete or unusable; the design would fail both conditions: a large research job combines specialized findings into one answer and a truncated response is not mistaken for successful completion.

B:

In an unfamiliar monorepo investigation, the hub can integrate distinct subagent outputs and resolve overlap without peer-to-peer chatter, but a length-limited result may be incomplete or unusable; therefore the separate requirement that a truncated response is not mistaken for successful completion would remain unsatisfied.

D:

In an unfamiliar monorepo investigation, reaching an output limit is a stop condition but does not establish task completion, but independent final outputs offer no single point to reconcile contradictions; therefore the separate requirement that a large research job combines specialized findings into one answer would remain unsatisfied.

Question 446

Tests for a repository change-review job will pass only if a coding assistant's output passes explicit acceptance checks and if long project guidance is divided into maintainable source files. Which proposed design directly satisfies both checks?

  1. accept an implementation because the agent describes it as complete; and use CLAUDE.md imports to organize referenced instructions
  2. accept an implementation because the agent describes it as complete; and duplicate full instruction paragraphs in every directory
  3. run the specified tests and compare results against acceptance criteria; and use CLAUDE.md imports to organize referenced instructions
  4. run the specified tests and compare results against acceptance criteria; and duplicate full instruction paragraphs in every directory
  5. implement only run the specified tests and compare results against acceptance criteria; make no corresponding change for the second requirement

Explanation

Correct Answer

C:

In a repository change-review job, objective checks support a pass decision better than self-reported confidence; alongside this, modular imports keep a short entry point while preserving reviewed details in separate files, satisfying the distinct requirements that a coding assistant's output passes explicit acceptance checks and that long project guidance is divided into maintainable source files.

Incorrect Answers

A:

In a repository change-review job, modular imports keep a short entry point while preserving reviewed details in separate files, but model assertions cannot replace build and behavior evidence; therefore the separate requirement that a coding assistant's output passes explicit acceptance checks would remain unsatisfied.

B:

In a repository change-review job, model assertions cannot replace build and behavior evidence and copying guidance encourages drift when policies change; the design would fail both conditions: a coding assistant's output passes explicit acceptance checks and long project guidance is divided into maintainable source files.

D:

In a repository change-review job, objective checks support a pass decision better than self-reported confidence, but copying guidance encourages drift when policies change; therefore the separate requirement that long project guidance is divided into maintainable source files would remain unsatisfied.

E:

In a repository change-review job, objective checks support a pass decision better than self-reported confidence, but the independent mechanism for the other criterion is absent: modular imports keep a short entry point while preserving reviewed details in separate files; as written, the design does not ensure that long project guidance is divided into maintainable source files.

Question 447

For a source-navigation tool for new hires, a reviewer identifies two independent obligations: specialists return comparable artifacts for aggregation and a destructive tool receives human authorization at execution time. Which proposal should be approved?

  1. merge narrative responses by concatenating their text; and collect a general approval at the start of the conversation
  2. give subagents a common results contract including claim, evidence, and status; and interrupt or deny the tool until an authorization callback approves it
  3. merge narrative responses by concatenating their text; and interrupt or deny the tool until an authorization callback approves it
  4. give subagents a common results contract including claim, evidence, and status; and collect a general approval at the start of the conversation

Explanation

Correct Answer

B:

In a source-navigation tool for new hires, a typed handoff lets the coordinator compare like fields and detect missing support; alongside this, execution gates make the approval binding on the selected action and inputs, satisfying the distinct requirements that specialists return comparable artifacts for aggregation and that a destructive tool receives human authorization at execution time.

Incorrect Answers

A:

In a source-navigation tool for new hires, unstructured concatenation hides provenance and duplicate or conflicting claims and blanket consent does not verify the later high-risk operation; the design would fail both conditions: specialists return comparable artifacts for aggregation and a destructive tool receives human authorization at execution time.

C:

In a source-navigation tool for new hires, execution gates make the approval binding on the selected action and inputs, but unstructured concatenation hides provenance and duplicate or conflicting claims; therefore the separate requirement that specialists return comparable artifacts for aggregation would remain unsatisfied.

D:

In a source-navigation tool for new hires, a typed handoff lets the coordinator compare like fields and detect missing support, but blanket consent does not verify the later high-risk operation; therefore the separate requirement that a destructive tool receives human authorization at execution time would remain unsatisfied.

Question 448

In a technical review of a customer entitlement assistant, the architect must address the following constraints together: a failed extraction is retried with specific corrective information; a function call accepts only valid argument shapes. Which change set is most defensible?

  1. feed field-level validation errors back in a bounded retry; and allow arbitrary keys and silently coerce incorrect types
  2. repeat the same vague prompt after each failure; and define the tool input schema, enable strict tool use where available, and validate inputs in the handler
  3. feed field-level validation errors back in a bounded retry; and define the tool input schema, enable strict tool use where available, and validate inputs in the handler
  4. repeat the same vague prompt after each failure; and allow arbitrary keys and silently coerce incorrect types

Explanation

Correct Answer

C:

In a customer entitlement assistant, targeted feedback identifies what the next attempt must correct; alongside this, strict tool inputs constrain generation, and handler validation provides a final enforceable argument boundary, satisfying the distinct requirements that a failed extraction is retried with specific corrective information and that a function call accepts only valid argument shapes.

Incorrect Answers

A:

In a customer entitlement assistant, targeted feedback identifies what the next attempt must correct, but unvalidated coercion can change intended operations or silently drop important fields; therefore the separate requirement that a function call accepts only valid argument shapes would remain unsatisfied.

B:

In a customer entitlement assistant, strict tool inputs constrain generation, and handler validation provides a final enforceable argument boundary, but unchanged requests do not reveal which constraint was violated; therefore the separate requirement that a failed extraction is retried with specific corrective information would remain unsatisfied.

D:

In a customer entitlement assistant, unchanged requests do not reveal which constraint was violated and unvalidated coercion can change intended operations or silently drop important fields; the design would fail both conditions: a failed extraction is retried with specific corrective information and a function call accepts only valid argument shapes.

Question 449

The release checklist for a tender-response data extractor explicitly checks that a large batch shares a consistent evaluation rubric; it also checks that confidence scores reflect available evidence and uncertainty. What coordinated solution meets the checklist?

  1. freeze the prompt version and schema for the submission set; and set confidence to high whenever the explanation is detailed
  2. change the classification criteria mid-batch without version tracking; and set confidence to high whenever the explanation is detailed
  3. change the classification criteria mid-batch without version tracking; and calibrate confidence against known outcomes and source support
  4. freeze the prompt version and schema for the submission set; and calibrate confidence against known outcomes and source support
  5. implement only freeze the prompt version and schema for the submission set; make no corresponding change for the second requirement

Explanation

Correct Answer

D:

In a tender-response data extractor, versioned common instructions make independently processed records comparable; alongside this, observed error rates and evidence quality are better than a fluent self-rating, satisfying the distinct requirements that a large batch shares a consistent evaluation rubric and that confidence scores reflect available evidence and uncertainty.

Incorrect Answers

A:

In a tender-response data extractor, versioned common instructions make independently processed records comparable, but verbal detail is not a validated reliability estimate; therefore the separate requirement that confidence scores reflect available evidence and uncertainty would remain unsatisfied.

B:

In a tender-response data extractor, mixed rubrics undermine consistent downstream aggregation and verbal detail is not a validated reliability estimate; the design would fail both conditions: a large batch shares a consistent evaluation rubric and confidence scores reflect available evidence and uncertainty.

C:

In a tender-response data extractor, observed error rates and evidence quality are better than a fluent self-rating, but mixed rubrics undermine consistent downstream aggregation; therefore the separate requirement that a large batch shares a consistent evaluation rubric would remain unsatisfied.

E:

In a tender-response data extractor, versioned common instructions make independently processed records comparable, but the independent mechanism for the other criterion is absent: observed error rates and evidence quality are better than a fluent self-rating; as written, the design does not ensure that confidence scores reflect available evidence and uncertainty.

Question 450

In a subscription support service, the acceptance criteria require that a destructive tool receives human authorization at execution time and that a truncated response is not mistaken for successful completion. Which paired implementation meets both?

  1. implement only interrupt or deny the tool until an authorization callback approves it; make no corresponding change for the second requirement
  2. collect a general approval at the start of the conversation; and interpret every non-tool stop reason as a successful final answer
  3. collect a general approval at the start of the conversation; and handle max_tokens as truncation rather than end_turn
  4. interrupt or deny the tool until an authorization callback approves it; and interpret every non-tool stop reason as a successful final answer
  5. interrupt or deny the tool until an authorization callback approves it; and handle max_tokens as truncation rather than end_turn

Explanation

Correct Answer

E:

In a subscription support service, execution gates make the approval binding on the selected action and inputs; alongside this, reaching an output limit is a stop condition but does not establish task completion, satisfying the distinct requirements that a destructive tool receives human authorization at execution time and that a truncated response is not mistaken for successful completion.

Incorrect Answers

A:

In a subscription support service, execution gates make the approval binding on the selected action and inputs, but the independent mechanism for the other criterion is absent: reaching an output limit is a stop condition but does not establish task completion; as written, the design does not ensure that a truncated response is not mistaken for successful completion.

B:

In a subscription support service, blanket consent does not verify the later high-risk operation and a length-limited result may be incomplete or unusable; the design would fail both conditions: a destructive tool receives human authorization at execution time and a truncated response is not mistaken for successful completion.

C:

In a subscription support service, reaching an output limit is a stop condition but does not establish task completion, but blanket consent does not verify the later high-risk operation; therefore the separate requirement that a destructive tool receives human authorization at execution time would remain unsatisfied.

D:

In a subscription support service, execution gates make the approval binding on the selected action and inputs, but a length-limited result may be incomplete or unusable; therefore the separate requirement that a truncated response is not mistaken for successful completion would remain unsatisfied.

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