Anthropic CCAR-P (Claude Certified Architect - Professional) Practice Test 20

Question 190

A tool's full API spec is stable but its permissions change by user role. Which implementation decision is the strongest fit?

  1. For this production path, retrieve product-specific instructions on demand after resolving the user's actual target.
  2. For the intended users, add semantic capability descriptions with constrained discovery and an explicit selection step.
  3. Within the delivery plan, load the reporting procedure when intent matches instead of putting it in every system prompt.
  4. At the system boundary, filter discoverable capabilities by current role before exposing details to the model.

Explanation

Correct Answer

D:

Discovery itself must respect authorization; hiding a schema cannot replace backend checks; the implementation should produce evidence that lower-privilege sessions neither discover nor invoke restricted functions.

Incorrect Answers

A:

Demand-driven context limits irrelevant guidance and prevents stale instructions dominating the prompt; this makes sense when a research agent loads entire vendor manuals before knowing which product the user selected, rather than the risk that a tool's full API spec is stable but its permissions change by user role.

B:

Well-described discovery entries help choose the correct integration without dumping every parameter; this technique fits where a catalog search must choose among many connectors whose names are difficult to distinguish, rather than the condition that a tool's full API spec is stable but its permissions change by user role.

C:

Conditional instruction loading improves focus for routine tasks without losing specialized capability; the trigger would be that an assistant needs a rarely used financial-reporting skill only for month-end requests, while the immediate question involves a tool's full API spec is stable but its permissions change by user role.

Question 191

A multistep agent keeps obsolete tool outputs and bloats the reasoning context. What should be prioritized during architecture review?

  1. In the proposed design, provide a compact tool index and reveal full schemas only after relevant selection.
  2. Under the reviewed approach, keep discovery concise but expand precise argument schemas before the selected tool call.
  3. For this production path, persist conclusions and retrieve earlier raw outputs only when the next step needs them.
  4. For the intended users, filter discoverable capabilities by current role before exposing details to the model.

Explanation

Correct Answer

C:

Lazy retrieval keeps active context relevant instead of appending all historical observations; acceptance evidence must confirm that a long trace remains within token limits while preserving decision-critical checkpoints.

Incorrect Answers

A:

Progressive disclosure reduces wasted context and distractors while preserving access to needed capabilities; the other workload is one where an assistant sees 300 enterprise tool schemas although most requests use only three, not the risk where a multistep agent keeps obsolete tool outputs and bloats the reasoning context.

B:

Progressive disclosure is not the removal of validation contracts required for correct execution; this applies when a team compresses all tool schemas into vague one-line descriptions and invocation accuracy falls, not this situation where a multistep agent keeps obsolete tool outputs and bloats the reasoning context.

D:

Discovery itself must respect authorization; hiding a schema cannot replace backend checks; it is justified if a tool's full API spec is stable but its permissions change by user role, but here a multistep agent keeps obsolete tool outputs and bloats the reasoning context.

Question 192

An assistant needs a rarely used financial-reporting skill only for month-end requests. Which response follows from the actual operational constraint?

  1. As a candidate control, filter discoverable capabilities by current role before exposing details to the model.
  2. In the initial solution, retrieve product-specific instructions on demand after resolving the user's actual target.
  3. In the proposed design, load the reporting procedure when intent matches instead of putting it in every system prompt.
  4. Under the reviewed approach, provide a compact tool index and reveal full schemas only after relevant selection.

Explanation

Correct Answer

C:

Conditional instruction loading improves focus for routine tasks without losing specialized capability; a clear sign-off signal is that normal requests omit the reporting payload; eligible month-end requests retrieve its checked instructions.

Incorrect Answers

A:

Discovery itself must respect authorization; hiding a schema cannot replace backend checks; it serves conditions where a tool's full API spec is stable but its permissions change by user role, but this decision concerns an assistant needs a rarely used financial-reporting skill only for month-end requests.

B:

Demand-driven context limits irrelevant guidance and prevents stale instructions dominating the prompt; the other case is when a research agent loads entire vendor manuals before knowing which product the user selected, but the actual risk is that an assistant needs a rarely used financial-reporting skill only for month-end requests.

D:

Progressive disclosure reduces wasted context and distractors while preserving access to needed capabilities; this technique fits where an assistant sees 300 enterprise tool schemas although most requests use only three, not the present case where an assistant needs a rarely used financial-reporting skill only for month-end requests.

Question 193

A team compresses all tool schemas into vague one-line descriptions and invocation accuracy falls. Which change targets the constraint rather than a proxy metric?

  1. In the proposed implementation, add semantic capability descriptions with constrained discovery and an explicit selection step.
  2. At this decision point, persist conclusions and retrieve earlier raw outputs only when the next step needs them.
  3. As a candidate control, load the reporting procedure when intent matches instead of putting it in every system prompt.
  4. For this deployment, retrieve product-specific instructions on demand after resolving the user's actual target.
  5. In the proposed design, keep discovery concise but expand precise argument schemas before the selected tool call.

Explanation

Correct Answer

E:

Progressive disclosure is not the removal of validation contracts required for correct execution; one measurable outcome is that argument-schema tests pass after expansion, even with the compact initial catalog.

Incorrect Answers

A:

Well-described discovery entries help choose the correct integration without dumping every parameter; it serves a setting where a catalog search must choose among many connectors whose names are difficult to distinguish, rather than the condition that a team compresses all tool schemas into vague one-line descriptions and invocation accuracy falls.

B:

Lazy retrieval keeps active context relevant instead of appending all historical observations; that is defensible if a multistep agent keeps obsolete tool outputs and bloats the reasoning context, while the immediate question involves a team compresses all tool schemas into vague one-line descriptions and invocation accuracy falls.

C:

Conditional instruction loading improves focus for routine tasks without losing specialized capability; the other workload is one where an assistant needs a rarely used financial-reporting skill only for month-end requests, but the present constraint is that a team compresses all tool schemas into vague one-line descriptions and invocation accuracy falls.

D:

Demand-driven context limits irrelevant guidance and prevents stale instructions dominating the prompt; this applies when a research agent loads entire vendor manuals before knowing which product the user selected, whereas the current issue is that a team compresses all tool schemas into vague one-line descriptions and invocation accuracy falls.

Question 194

A frequently used support operation spends more context tokens describing unavailable tools than answering requests. Which measurement is the most relevant release criterion?

  1. The reviewer observes that simple requests use a small schema budget and still find the correct specialist tool.
  2. The release report verifies that a product change causes the next retrieval to load only the newly relevant documentation.
  3. The controlled replay verifies that argument-schema tests pass after expansion, even with the compact initial catalog.
  4. A reproducible run shows that lower-privilege sessions neither discover nor invoke restricted functions.

Explanation

Correct Answer

A:

Progressive disclosure reduces wasted context and distractors while preserving access to needed capabilities; the condition under test is that an assistant sees 300 enterprise tool schemas although most requests use only three.

Incorrect Answers

B:

Demand-driven context limits irrelevant guidance and prevents stale instructions dominating the prompt; it is relevant if a research agent loads entire vendor manuals before knowing which product the user selected, whereas this scenario involves a frequently used support operation spends more context tokens describing unavailable tools than answering requests.

C:

Progressive disclosure is not the removal of validation contracts required for correct execution; it serves conditions where a team compresses all tool schemas into vague one-line descriptions and invocation accuracy falls, but here a frequently used support operation spends more context tokens describing unavailable tools than answering requests.

D:

Discovery itself must respect authorization; hiding a schema cannot replace backend checks; the other case is when a tool's full API spec is stable but its permissions change by user role, instead of when a frequently used support operation spends more context tokens describing unavailable tools than answering requests.

Question 195

A product-specific agent reads all manuals during startup even before a product identifier has been selected. Which production-like test finding is most convincing?

  1. A production-like test shows that simple requests use a small schema budget and still find the correct specialist tool.
  2. The release check confirms that ambiguous requests elicit clarification or a bounded candidate list rather than random tool use.
  3. The reviewer observes that a product change causes the next retrieval to load only the newly relevant documentation.
  4. The release report verifies that normal requests omit the reporting payload; eligible month-end requests retrieve its checked instructions.
  5. The controlled replay verifies that lower-privilege sessions neither discover nor invoke restricted functions.

Explanation

Correct Answer

C:

Demand-driven context limits irrelevant guidance and prevents stale instructions dominating the prompt; this is relevant for cases where a research agent loads entire vendor manuals before knowing which product the user selected.

Incorrect Answers

A:

Progressive disclosure reduces wasted context and distractors while preserving access to needed capabilities; this control is needed when an assistant sees 300 enterprise tool schemas although most requests use only three, but the actual risk is that a product-specific agent reads all manuals during startup even before a product identifier has been selected.

B:

Well-described discovery entries help choose the correct integration without dumping every parameter; that would be useful if a catalog search must choose among many connectors whose names are difficult to distinguish, but the requested boundary concerns a product-specific agent reads all manuals during startup even before a product identifier has been selected.

D:

Conditional instruction loading improves focus for routine tasks without losing specialized capability; that is defensible if an assistant needs a rarely used financial-reporting skill only for month-end requests, while this problem involves a product-specific agent reads all manuals during startup even before a product identifier has been selected.

E:

Discovery itself must respect authorization; hiding a schema cannot replace backend checks; the other workload is one where a tool's full API spec is stable but its permissions change by user role, and not where a product-specific agent reads all manuals during startup even before a product identifier has been selected.

Question 196

A connector directory returns relevant functions, but ambiguous descriptions cause wrong-tool invocations. Which finding should the release gate require?

  1. The observed outcome is that ambiguous requests elicit clarification or a bounded candidate list rather than random tool use.
  2. The regression suite verifies that a product change causes the next retrieval to load only the newly relevant documentation.
  3. A production-like test shows that normal requests omit the reporting payload; eligible month-end requests retrieve its checked instructions.
  4. The release check confirms that simple requests use a small schema budget and still find the correct specialist tool.

Explanation

Correct Answer

A:

Well-described discovery entries help choose the correct integration without dumping every parameter; it tests the operating case in which a catalog search must choose among many connectors whose names are difficult to distinguish.

Incorrect Answers

B:

Demand-driven context limits irrelevant guidance and prevents stale instructions dominating the prompt; the approach helps when a research agent loads entire vendor manuals before knowing which product the user selected, but the present constraint is that a connector directory returns relevant functions, but ambiguous descriptions cause wrong-tool invocations.

C:

Conditional instruction loading improves focus for routine tasks without losing specialized capability; that belongs to situations where an assistant needs a rarely used financial-reporting skill only for month-end requests, whereas the current issue is that a connector directory returns relevant functions, but ambiguous descriptions cause wrong-tool invocations.

D:

Progressive disclosure reduces wasted context and distractors while preserving access to needed capabilities; it is relevant if an assistant sees 300 enterprise tool schemas although most requests use only three, whereas the release criterion addresses a connector directory returns relevant functions, but ambiguous descriptions cause wrong-tool invocations.

Question 197

A tool registry advertises a single static permission list even though each user has different grants. Which verified outcome most directly addresses the reported risk?

  1. The scenario test demonstrates that argument-schema tests pass after expansion, even with the compact initial catalog.
  2. The end-to-end check confirms that lower-privilege sessions neither discover nor invoke restricted functions.
  3. The recorded test establishes that ambiguous requests elicit clarification or a bounded candidate list rather than random tool use.
  4. The regression suite verifies that a long trace remains within token limits while preserving decision-critical checkpoints.

Explanation

Correct Answer

B:

Discovery itself must respect authorization; hiding a schema cannot replace backend checks; the test covers the original risk that a tool's full API spec is stable but its permissions change by user role.

Incorrect Answers

A:

Progressive disclosure is not the removal of validation contracts required for correct execution; the design helps when a team compresses all tool schemas into vague one-line descriptions and invocation accuracy falls, whereas this scenario involves a tool registry advertises a single static permission list even though each user has different grants.

C:

Well-described discovery entries help choose the correct integration without dumping every parameter; this control is needed when a catalog search must choose among many connectors whose names are difficult to distinguish, instead of when a tool registry advertises a single static permission list even though each user has different grants.

D:

Lazy retrieval keeps active context relevant instead of appending all historical observations; that would be useful if a multistep agent keeps obsolete tool outputs and bloats the reasoning context, not the case in which a tool registry advertises a single static permission list even though each user has different grants.

Question 198

A research assistant continues treating old search results as decision evidence after their validity expires. What independently verifiable observation is needed?

  1. A controlled trial finds that ambiguous requests elicit clarification or a bounded candidate list rather than random tool use.
  2. The measured evidence confirms that lower-privilege sessions neither discover nor invoke restricted functions.
  3. The scenario test demonstrates that a long trace remains within token limits while preserving decision-critical checkpoints.
  4. The end-to-end check confirms that normal requests omit the reporting payload; eligible month-end requests retrieve its checked instructions.

Explanation

Correct Answer

C:

Lazy retrieval keeps active context relevant instead of appending all historical observations; the verification boundary is the one where a multistep agent keeps obsolete tool outputs and bloats the reasoning context.

Incorrect Answers

A:

Well-described discovery entries help choose the correct integration without dumping every parameter; it is justified if a catalog search must choose among many connectors whose names are difficult to distinguish, but the requested boundary concerns a research assistant continues treating old search results as decision evidence after their validity expires.

B:

Discovery itself must respect authorization; hiding a schema cannot replace backend checks; its scope includes cases where a tool's full API spec is stable but its permissions change by user role, not the present case where a research assistant continues treating old search results as decision evidence after their validity expires.

D:

Conditional instruction loading improves focus for routine tasks without losing specialized capability; the approach helps when an assistant needs a rarely used financial-reporting skill only for month-end requests, and not where a research assistant continues treating old search results as decision evidence after their validity expires.

Question 199

Legal documents are split every 500 characters, severing definitions from their exceptions. Which action is justified by the scenario?

  1. In the project scope, deduplicate and filter by authoritative version before reranking retrieved evidence.
  2. Within the rollout design, add verified text extraction for scans with quality checks and preserve original file provenance.
  3. For the next architecture revision, chunk along semantic headings with controlled overlap and parent-document references.
  4. For the current integration, index metadata for tenant, classification and effective dates and apply filters before ranking.
  5. Under the system plan, version and atomically refresh the index with deletion handling and source timestamps.

Explanation

Correct Answer

C:

Structure-aware chunks preserve enough local meaning for retrieval and legally important qualifiers; one measurable outcome is that retrieval tests find both the governing clause and its exception for matched queries.

Incorrect Answers

A:

Duplicate and stale passages waste context and can bias synthesis toward conflicting historic content; this technique fits where search results show duplicate near-identical chunks from many versions of the same policy, but the present constraint is that legal documents are split every 500 characters, severing definitions from their exceptions.

B:

Non-text sources must be converted reliably before embedding; unreadable text cannot be retrieved; the trigger would be that an ingestion process skips scanned forms because the PDF text layer is empty, whereas the current issue is that legal documents are split every 500 characters, severing definitions from their exceptions.

D:

Vector similarity alone cannot enforce source entitlement or policy validity; that fits cases where a corpus contains both general guidance and document-specific restrictions that should not cross tenants, while the required safeguard concerns legal documents are split every 500 characters, severing definitions from their exceptions.

E:

Freshness requires consistent ingestion and invalidation rather than prompt reminders about current policy; this control is needed when a knowledge base is refreshed nightly but answers still cite superseded policies, not the risk where legal documents are split every 500 characters, severing definitions from their exceptions.

Question 200

A knowledge base is refreshed nightly but answers still cite superseded policies. What change directly resolves this production-design issue?

  1. For this workflow, version and atomically refresh the index with deletion handling and source timestamps.
  2. For the affected service, index metadata for tenant, classification and effective dates and apply filters before ranking.
  3. In the project scope, rework passage indexing and reranking on labeled question-to-evidence pairs.
  4. For the intended users, add verified text extraction for scans with quality checks and preserve original file provenance.

Explanation

Correct Answer

A:

Freshness requires consistent ingestion and invalidation rather than prompt reminders about current policy; a suitable release result is that revoked passages disappear and new documents become searchable after the declared freshness window.

Incorrect Answers

B:

Vector similarity alone cannot enforce source entitlement or policy validity; the alternative helps if a corpus contains both general guidance and document-specific restrictions that should not cross tenants, instead of when a knowledge base is refreshed nightly but answers still cite superseded policies.

C:

Document-level relevance is insufficient when the model needs precise passages to ground claims; it is justified if an assistant retrieves relevant file names but not the answer-bearing paragraphs, not the case in which a knowledge base is refreshed nightly but answers still cite superseded policies.

D:

Non-text sources must be converted reliably before embedding; unreadable text cannot be retrieved; its scope includes cases where an ingestion process skips scanned forms because the PDF text layer is empty, whereas the problem presented is that a knowledge base is refreshed nightly but answers still cite superseded policies.

Claude Certified Architect - Professional (CCAR-P) Practice Tests