Anthropic CCAR-P (Claude Certified Architect - Professional) Practice Test 14

Question 130

A team trims system instructions to save tokens and then sees authorization regressions. Which approach should the design team choose?

  1. At this decision point, summarize completed observations and discard raw outputs no longer needed for decisions.
  2. As a candidate control, use targeted retrieval and evaluate positional recall rather than relying on window size.
  3. In the initial solution, protect safety and identity constraints while compressing lower-priority redundant prose.
  4. In the proposed design, restructure prompts with stable reusable prefixes and appropriate caching when supported.

Explanation

Correct Answer

C:

Token savings must not weaken the governing rules that make the workflow safe; the controlled test needs to find that adversarial tests confirm authorization behavior is unchanged after compaction.

Incorrect Answers

A:

Tool-result hygiene keeps context focused on actionable evidence and current state; this makes sense when a multi-tool agent accumulates irrelevant tool outputs after each step, rather than the risk that a team trims system instructions to save tokens and then sees authorization regressions.

B:

Long-context placement effects require evidence selection and careful testing of factual grounding; this technique fits where a user asks about a value in a huge document and the model misses it near the center, rather than the condition that a team trims system instructions to save tokens and then sees authorization regressions.

D:

Reuse of unchanged large context can reduce repetitive processing while keeping request-specific data separate; the design helps when a cost review shows repeated static reference pages transmitted on every request, but the present constraint is that a team trims system instructions to save tokens and then sees authorization regressions.

Question 131

A multi-tool agent accumulates irrelevant tool outputs after each step. Where should the solution architect focus the implementation?

  1. During architecture review, restructure prompts with stable reusable prefixes and appropriate caching when supported.
  2. In the proposed implementation, retain structured persistent task state and summarize disposable tool chatter.
  3. At this decision point, compact older dialogue into validated state while preserving active constraints verbatim.
  4. For the operating team, use targeted retrieval and evaluate positional recall rather than relying on window size.
  5. In the initial solution, summarize completed observations and discard raw outputs no longer needed for decisions.

Explanation

Correct Answer

E:

Tool-result hygiene keeps context focused on actionable evidence and current state; validation needs to show that long-run tests remain under the token budget without losing decision-critical facts.

Incorrect Answers

A:

Reuse of unchanged large context can reduce repetitive processing while keeping request-specific data separate; the other workload is one where a cost review shows repeated static reference pages transmitted on every request, not the risk where a multi-tool agent accumulates irrelevant tool outputs after each step.

B:

State compaction separates durable decisions from transient execution details; this applies when a conversation needs running project state but old tool results overwhelm the prompt, not this situation where a multi-tool agent accumulates irrelevant tool outputs after each step.

C:

Context management must prioritize current obligations and unresolved facts over conversational history bulk; the alternative helps if a long chat transcript pushes the newest safety-critical instruction out of the available context, whereas this scenario involves a multi-tool agent accumulates irrelevant tool outputs after each step.

D:

Long-context placement effects require evidence selection and careful testing of factual grounding; it is justified if a user asks about a value in a huge document and the model misses it near the center, but here a multi-tool agent accumulates irrelevant tool outputs after each step.

Question 132

A support conversation grows until a later turn omits an active restriction carried from the first request. What result discriminates a working safeguard from a demo?

  1. The release check confirms that a replay after compaction preserves active user permissions and outstanding tasks.
  2. The reviewer observes that adversarial tests confirm authorization behavior is unchanged after compaction.
  3. The release report verifies that after ten turns the assistant retains approvals and unresolved actions without stale logs.
  4. The controlled replay verifies that needle-in-document evaluations preserve cited accuracy across document positions.

Explanation

Correct Answer

A:

Context management must prioritize current obligations and unresolved facts over conversational history bulk; the control should hold even if a long chat transcript pushes the newest safety-critical instruction out of the available context.

Incorrect Answers

B:

Token savings must not weaken the governing rules that make the workflow safe; the other case is when a team trims system instructions to save tokens and then sees authorization regressions, but the actual risk is that a support conversation grows until a later turn omits an active restriction carried from the first request.

C:

State compaction separates durable decisions from transient execution details; this makes sense when a conversation needs running project state but old tool results overwhelm the prompt, but the requested boundary concerns a support conversation grows until a later turn omits an active restriction carried from the first request.

D:

Long-context placement effects require evidence selection and careful testing of factual grounding; this technique fits where a user asks about a value in a huge document and the model misses it near the center, not the present case where a support conversation grows until a later turn omits an active restriction carried from the first request.

Question 133

A retrieval response includes dozens of pages irrelevant to the requested subclause, inflating token usage. What outcome must the reviewer see before proceeding?

  1. The regression suite verifies that selected snippets cover answer-bearing sections while irrelevant pages are excluded.
  2. A production-like test shows that adversarial tests confirm authorization behavior is unchanged after compaction.
  3. The release check confirms that cache telemetry shows eligible hits and lower effective input-processing cost.
  4. The reviewer observes that long-run tests remain under the token budget without losing decision-critical facts.
  5. The release report verifies that needle-in-document evaluations preserve cited accuracy across document positions.

Explanation

Correct Answer

A:

More context can dilute relevance and increase cost without improving grounded answers; the test is necessary where a retrieval assistant sends every matching manual page to Claude even for narrow queries.

Incorrect Answers

B:

Token savings must not weaken the governing rules that make the workflow safe; that is defensible if a team trims system instructions to save tokens and then sees authorization regressions, while the immediate question involves a retrieval response includes dozens of pages irrelevant to the requested subclause, inflating token usage.

C:

Reuse of unchanged large context can reduce repetitive processing while keeping request-specific data separate; the other workload is one where a cost review shows repeated static reference pages transmitted on every request, but the present constraint is that a retrieval response includes dozens of pages irrelevant to the requested subclause, inflating token usage.

D:

Tool-result hygiene keeps context focused on actionable evidence and current state; this applies when a multi-tool agent accumulates irrelevant tool outputs after each step, whereas the current issue is that a retrieval response includes dozens of pages irrelevant to the requested subclause, inflating token usage.

E:

Long-context placement effects require evidence selection and careful testing of factual grounding; the alternative helps if a user asks about a value in a huge document and the model misses it near the center, whereas the release criterion addresses a retrieval response includes dozens of pages irrelevant to the requested subclause, inflating token usage.

Question 134

A long dossier test hides one critical counterexample between many similar examples. Which result should the pilot review prioritize?

  1. The end-to-end check confirms that a replay after compaction preserves active user permissions and outstanding tasks.
  2. The observed outcome is that selected snippets cover answer-bearing sections while irrelevant pages are excluded.
  3. The regression suite verifies that needle-in-document evaluations preserve cited accuracy across document positions.
  4. A production-like test shows that after ten turns the assistant retains approvals and unresolved actions without stale logs.

Explanation

Correct Answer

C:

Long-context placement effects require evidence selection and careful testing of factual grounding; the test covers the original risk that a user asks about a value in a huge document and the model misses it near the center.

Incorrect Answers

A:

Context management must prioritize current obligations and unresolved facts over conversational history bulk; that belongs to situations where a long chat transcript pushes the newest safety-critical instruction out of the available context, not this situation where a long dossier test hides one critical counterexample between many similar examples.

B:

More context can dilute relevance and increase cost without improving grounded answers; it is relevant if a retrieval assistant sends every matching manual page to Claude even for narrow queries, whereas this scenario involves a long dossier test hides one critical counterexample between many similar examples.

D:

State compaction separates durable decisions from transient execution details; the other case is when a conversation needs running project state but old tool results overwhelm the prompt, instead of when a long dossier test hides one critical counterexample between many similar examples.

Question 135

An investigation resumes after a pause but loses which tool calls were already completed. What result would directly validate the necessary safeguard?

  1. The measured evidence confirms that needle-in-document evaluations preserve cited accuracy across document positions.
  2. The scenario test demonstrates that after ten turns the assistant retains approvals and unresolved actions without stale logs.
  3. The end-to-end check confirms that long-run tests remain under the token budget without losing decision-critical facts.
  4. The observed outcome is that adversarial tests confirm authorization behavior is unchanged after compaction.

Explanation

Correct Answer

B:

State compaction separates durable decisions from transient execution details; a valid sign-off must address the risk that a conversation needs running project state but old tool results overwhelm the prompt.

Incorrect Answers

A:

Long-context placement effects require evidence selection and careful testing of factual grounding; this control is needed when a user asks about a value in a huge document and the model misses it near the center, but the actual risk is that an investigation resumes after a pause but loses which tool calls were already completed.

C:

Tool-result hygiene keeps context focused on actionable evidence and current state; it serves a setting where a multi-tool agent accumulates irrelevant tool outputs after each step, not the present case where an investigation resumes after a pause but loses which tool calls were already completed.

D:

Token savings must not weaken the governing rules that make the workflow safe; that is defensible if a team trims system instructions to save tokens and then sees authorization regressions, while this problem involves an investigation resumes after a pause but loses which tool calls were already completed.

Question 136

A pricing review shows a stable reference prefix re-sent unchanged to the model on every request. What evidence would reveal whether the control truly works?

  1. The negative-case evaluation confirms that long-run tests remain under the token budget without losing decision-critical facts.
  2. A controlled trial finds that after ten turns the assistant retains approvals and unresolved actions without stale logs.
  3. The readiness assessment shows that needle-in-document evaluations preserve cited accuracy across document positions.
  4. The scenario test demonstrates that cache telemetry shows eligible hits and lower effective input-processing cost.

Explanation

Correct Answer

D:

Reuse of unchanged large context can reduce repetitive processing while keeping request-specific data separate; the evidence must be robust when a cost review shows repeated static reference pages transmitted on every request.

Incorrect Answers

A:

Tool-result hygiene keeps context focused on actionable evidence and current state; it targets cases where a multi-tool agent accumulates irrelevant tool outputs after each step, while the immediate question involves a pricing review shows a stable reference prefix re-sent unchanged to the model on every request.

B:

State compaction separates durable decisions from transient execution details; the approach helps when a conversation needs running project state but old tool results overwhelm the prompt, but the present constraint is that a pricing review shows a stable reference prefix re-sent unchanged to the model on every request.

C:

Long-context placement effects require evidence selection and careful testing of factual grounding; that belongs to situations where a user asks about a value in a huge document and the model misses it near the center, whereas the current issue is that a pricing review shows a stable reference prefix re-sent unchanged to the model on every request.

Question 137

An aggressive prompt-reduction experiment removes a control instruction and breaks authorization behavior. Which result proves the requested control, not just configuration?

  1. The recorded test establishes that cache telemetry shows eligible hits and lower effective input-processing cost.
  2. The representative case test finds that adversarial tests confirm authorization behavior is unchanged after compaction.
  3. The staged rollout demonstrates that long-run tests remain under the token budget without losing decision-critical facts.
  4. A controlled trial finds that a replay after compaction preserves active user permissions and outstanding tasks.

Explanation

Correct Answer

B:

Token savings must not weaken the governing rules that make the workflow safe; the test covers the original risk that a team trims system instructions to save tokens and then sees authorization regressions.

Incorrect Answers

A:

Reuse of unchanged large context can reduce repetitive processing while keeping request-specific data separate; the design helps when a cost review shows repeated static reference pages transmitted on every request, whereas this scenario involves an aggressive prompt-reduction experiment removes a control instruction and breaks authorization behavior.

C:

Tool-result hygiene keeps context focused on actionable evidence and current state; this control is needed when a multi-tool agent accumulates irrelevant tool outputs after each step, instead of when an aggressive prompt-reduction experiment removes a control instruction and breaks authorization behavior.

D:

Context management must prioritize current obligations and unresolved facts over conversational history bulk; that would be useful if a long chat transcript pushes the newest safety-critical instruction out of the available context, not the case in which an aggressive prompt-reduction experiment removes a control instruction and breaks authorization behavior.

Question 138

A help-desk app sends the same lengthy policy manual before each unique ticket. How should the architect revise the proposed solution?

  1. For the primary workstream, put stable policy material in a cacheable prefix and append variable ticket data after it.
  2. For this workflow, review Skill resources, tool grants and invocation permissions under least privilege.
  3. For the affected service, keep volatile account state outside stable cache blocks and refresh it per request.
  4. In the project scope, extract common governed instructions into reusable prompt modules under version control.
  5. Within the rollout design, package maintained instructions and resources as a documented Skill for the relevant environment.

Explanation

Correct Answer

A:

Caching works best when repeated prefix content stays unchanged between eligible calls; a clear sign-off signal is that repeated requests show cache hits while different tickets still receive correct answers.

Incorrect Answers

B:

Reusable instructions are not a substitute for sandboxing and authorization of real tools; its scope includes cases where a security team worries Skills can execute actions beyond their intended role, not the present case where a help-desk app sends the same lengthy policy manual before each unique ticket.

C:

Caching stale transactional details risks incorrect answers even if it reduces repeated tokens; it targets cases where a cached prefix includes current account balances that change between requests, while this problem involves a help-desk app sends the same lengthy policy manual before each unique ticket.

D:

Modularity ensures a policy revision can be reviewed and propagated consistently; the approach helps when a company maintains a dozen prompts with copied global safety policy and inconsistent updates, and not where a help-desk app sends the same lengthy policy manual before each unique ticket.

E:

Skills organize repeatable contextual procedures without requiring every chat to restate the entire playbook; that belongs to situations where a developer wants a reusable guided workflow for recurring repository migrations, rather than the risk that a help-desk app sends the same lengthy policy manual before each unique ticket.

Question 139

A company maintains a dozen prompts with copied global safety policy and inconsistent updates. Which control should govern the first implementation?

  1. Within the delivery plan, use named prompt versions with deployment pinning and controlled promotion.
  2. At the system boundary, put stable policy material in a cacheable prefix and append variable ticket data after it.
  3. For the primary workstream, canonicalize stable prompt section ordering before measuring cache effectiveness.
  4. For this workflow, extract common governed instructions into reusable prompt modules under version control.

Explanation

Correct Answer

D:

Modularity ensures a policy revision can be reviewed and propagated consistently; the solution is testable when a change inventory shows each application uses the new approved module version.

Incorrect Answers

A:

Reusable templates need lifecycle controls so experiments do not silently affect every consuming system; this technique fits where a team calls a 200-line prompt from many services but cannot roll back one experiment, but the present constraint is that a company maintains a dozen prompts with copied global safety policy and inconsistent updates.

B:

Caching works best when repeated prefix content stays unchanged between eligible calls; the trigger would be that a help-desk app sends the same lengthy policy manual before each unique ticket, whereas the current issue is that a company maintains a dozen prompts with copied global safety policy and inconsistent updates.

C:

Cacheability depends on repeated identical content, not merely on semantically similar instructions; the design helps when a cache optimization changes header order per request, yielding no reused prefixes, whereas the release criterion addresses a company maintains a dozen prompts with copied global safety policy and inconsistent updates.

Question 140

An engineering organization uses Claude Code to guide recurring repository migrations, and its shared, lengthy instruction prefix is intended for caching. Developers currently hand-copy migration steps and reorder static context in each request. Which TWO changes improve reusable process guidance and stable cache reuse? Choose TWO.

  1. For this production path, review Skill resources, tool grants and invocation permissions under least privilege.
  2. For the intended users, put stable policy material in a cacheable prefix and append variable ticket data after it.
  3. Within the delivery plan, canonicalize stable prompt section ordering before measuring cache effectiveness.
  4. At the system boundary, package maintained instructions and resources as a documented Skill for the relevant environment.
  5. For the primary workstream, use named prompt versions with deployment pinning and controlled promotion.

Explanation

Correct Answers

C:

Cacheability depends on repeated identical content, not merely on semantically similar instructions; one operational check should confirm that telemetry records improved eligible cache hits after consistent prefix construction; another obligation arises when a developer wants a reusable guided workflow for recurring repository migrations.

D:

Skills organize repeatable contextual procedures without requiring every chat to restate the entire playbook; the intended outcome is that independent developers invoke the Skill and reproduce the approved migration checks; another obligation arises when a cache optimization changes header order per request, yielding no reused prefixes.

Incorrect Answers

A:

Reusable instructions are not a substitute for sandboxing and authorization of real tools; this applies when a security team worries Skills can execute actions beyond their intended role, but here a developer wants a reusable guided workflow for recurring repository migrations; another obligation arises when a cache optimization changes header order per request, yielding no reused prefixes.

B:

Caching works best when repeated prefix content stays unchanged between eligible calls; the alternative helps if a help-desk app sends the same lengthy policy manual before each unique ticket, instead of when a cache optimization changes header order per request, yielding no reused prefixes; another obligation arises when a developer wants a reusable guided workflow for recurring repository migrations.

E:

Reusable templates need lifecycle controls so experiments do not silently affect every consuming system; it targets cases where a team calls a 200-line prompt from many services but cannot roll back one experiment, but this decision concerns a developer wants a reusable guided workflow for recurring repository migrations; another obligation arises when a cache optimization changes header order per request, yielding no reused prefixes.

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