Anthropic CCAR-P (Claude Certified Architect - Professional) Practice Test 15

Question 140

A legal operations change board is considering removing Rate and spend limits by tenant to simplify a release. Which outcome is the primary residual risk if that protection is absent?

  1. The risk that a global cost cap allows one tenant to crowd out others
  2. The risk that after-the-fact logs cannot reverse a completed sensitive transfer
  3. The risk that upstream input minimization alone may not catch derived disclosures
  4. The risk that instruction blending permits external authors to steer agent behavior

Explanation

Correct Answer

A:

The loss of the control permits the situation where a global cost cap allows one tenant to crowd out others. Its governing mechanism is that the control layer can enforce bounded usage budgets and anomaly controls per tenant; without an equivalent mechanism, a favorable demonstration cannot establish safety or quality.

Incorrect Answers

B:

If proposed external transfers could produce irreversible customer harm, after-the-fact logs cannot reverse a completed sensitive transfer; however, the named safeguard addresses the distinct hazard that a global cost cap allows one tenant to crowd out others.

C:

When generated summaries may echo sensitive identifiers unnecessary for recipients, upstream input minimization alone may not catch derived disclosures; whereas removing Rate and spend limits by tenant primarily exposes the risk that a global cost cap allows one tenant to crowd out others.

D:

The other exposure arises when supplier documents include embedded instructions masquerading as system messages: instruction blending permits external authors to steer agent behavior. Omitting the requested control instead leaves the more immediate risk that a global cost cap allows one tenant to crowd out others; but this is not the immediate loss associated with omitting Rate and spend limits by tenant: a global cost cap allows one tenant to crowd out others.

Question 141

The insurance processing deployment uses Capped tool-loop execution, but executive reports currently count only successful model responses. The residual concern is that uncontrolled retries exhaust quotas and conceal a failed precondition. Which acceptance measure would reveal real progress?

  1. Evaluate successful read tasks without elevated write privilege
  2. Evaluate prohibited-call rejection and schema-compliant invocation rates
  3. Evaluate attempts per failed case and bounded-failure completion rate
  4. Evaluate tool-definition tokens per resolved user task
  5. Evaluate unnecessary tool-call rate and task success under the reduced set

Explanation

Correct Answer

C:

Tracking attempts per failed case and bounded-failure completion rate provides direct evidence that the system delivers its intended effect: the integration can enforce retry, token and call budgets with a deliberate stop path. The measure is more informative than prompt counts because the failure being measured is that uncontrolled retries exhaust quotas and conceal a failed precondition.

Incorrect Answers

A:

The alternative indicator checks whether the integration can expose read-only access independently from privileged changes when an assistant must read audit events but rarely modify audit settings; however, that proxy would not demonstrate attempts per failed case and bounded-failure completion rate for the selected intervention.

B:

When a generic execute-any-query tool lets a model submit unreviewed commands, the integration can replace broad verbs with narrow parameterized operations and validation; but the acceptance gate must instead inspect attempts per failed case and bounded-failure completion rate to detect the targeted failure.

D:

When a complex connector ecosystem cannot fit full tool descriptions in each request, the integration can advertise compact discovery metadata and load detailed definitions on demand; however, that proxy would not demonstrate attempts per failed case and bounded-failure completion rate for the selected intervention.

E:

When a research agent is offered dozens of irrelevant mutation tools, the integration can expose only tools needed for the active task and user permission scope; but the acceptance gate must instead inspect attempts per failed case and bounded-failure completion rate to detect the targeted failure.

Question 142

A lean redesign in media licensing would retire Unsafe-action escape rate while leaving the rest of the service unchanged. Which operational exposure is most directly increased?

  1. An exposure in which citation formatting alone does not prove support for the claim
  2. An exposure in which general answer accuracy can hide rare severe tool abuse
  3. An exposure in which median latency can hide the failures users notice most
  4. An exposure in which accuracy on answered cases alone can reward refusing everything

Explanation

Correct Answer

B:

The risk is not a generic model inconvenience: general answer accuracy can hide rare severe tool abuse. The proposed safeguard changes that boundary because the evaluation process can measure disallowed actions that reach a protected side-effecting endpoint, and the release would need policy test harness and action audit events plus unauthorized side effects per adversarial test run to show the risk remains contained.

Incorrect Answers

A:

The competing hazard is that citation formatting alone does not prove support for the claim in a workload where the system summarizes regulated policies with citations; whereas removing Unsafe-action escape rate primarily exposes the risk that general answer accuracy can hide rare severe tool abuse.

C:

The competing hazard is that median latency can hide the failures users notice most in a workload where interactive sessions frequently exceed their response deadline; however, the named safeguard addresses the distinct hazard that general answer accuracy can hide rare severe tool abuse.

D:

When incomplete evidence should yield escalation rather than invented certainty, accuracy on answered cases alone can reward refusing everything; whereas removing Unsafe-action escape rate primarily exposes the risk that general answer accuracy can hide rare severe tool abuse.

Question 143

Following a regression in software platform, the team establishes that failover to an untested model can change schema and decision quality. Which proposed redesign most directly prevents recurrence?

  1. Deploy Fallback model under failure
  2. Deploy Context-capacity fit before model selection
  3. Deploy Reasoning-capable model for complex synthesis
  4. Deploy Small-model first-pass routing

Explanation

Correct Answer

A:

The remediation acts at the relevant design boundary: the model configuration can route to an approved fallback model with compatible output contracts. That counteracts the observed hazard that failover to an untested model can change schema and decision quality. Rollout should include model contract compatibility tests and outage policy, with regression evidence from safe fallback completion rate under simulated outages.

Incorrect Answers

B:

This addresses the failure where nominal context limits alone do not prove reliable use of every passage, because the model configuration can test effective context utilization and recall before choosing a model configuration; whereas the software platform defect follows the path in which failover to an untested model can change schema and decision quality.

C:

Where multi-document arbitration requires several interdependent inferences, a stronger model can justify tradeoffs across heterogeneous evidence; yet it would not correct the observed weakness: failover to an untested model can change schema and decision quality.

D:

Although the alternative reduces the risk that always using the strongest model increases cost and wait time because a fast adequate tier handles easy cases and escalates only uncertain ones; but this remedy operates on another boundary, not the case where failover to an untested model can change schema and decision quality.

Question 144

A cost-cutting proposal for telecommunications support eliminates Baseline workflow benchmarking. Which undesirable consequence would become most credible after that change?

  1. An exposure in which a polished demo can hide unacceptable long-tail failures
  2. An exposure in which a portfolio of speculative features can crowd out working high-value uses
  3. An exposure in which deployment can appear successful while adding hidden review effort
  4. An exposure in which automating the last step could bypass required accountability

Explanation

Correct Answer

C:

This is the direct residual hazard: deployment can appear successful while adding hidden review effort. The omitted intervention matters because measuring existing effort establishes a credible incremental-value estimate. A reviewer should demand a baseline observation period and matching task definitions and measure end-to-end labor minutes per accepted outcome before treating the risk as controlled.

Incorrect Answers

A:

The competing hazard is that a polished demo can hide unacceptable long-tail failures in a workload where the proposed automation encounters exceptions that have never been tested; however, the named safeguard addresses the distinct hazard that deployment can appear successful while adding hidden review effort.

B:

The other exposure arises when a team wants to fund a large platform before validating which use case helps: a portfolio of speculative features can crowd out working high-value uses. Omitting the requested control instead leaves the more immediate risk that deployment can appear successful while adding hidden review effort; whereas removing Baseline workflow benchmarking primarily exposes the risk that deployment can appear successful while adding hidden review effort.

D:

The competing hazard is that automating the last step could bypass required accountability in a workload where caseworkers must retain authority for consequential final decisions; however, the named safeguard addresses the distinct hazard that deployment can appear successful while adding hidden review effort.

Question 145

The post-incident analysis for legal operations concludes that uncontrolled retries exhaust quotas and conceal a failed precondition. Which remediation targets the relevant failure boundary?

  1. Apply Progressive tool discovery
  2. Apply Read-write capability separation
  3. Apply Tool ownership and lifecycle review
  4. Apply Capped tool-loop execution

Explanation

Correct Answer

D:

The engineering fix changes the mechanism rather than the presentation: the integration can enforce retry, token and call budgets with a deliberate stop path. It requires loop policy and classified tool-error responses; a regression test should examine attempts per failed case and bounded-failure completion rate in cases where uncontrolled retries exhaust quotas and conceal a failed precondition.

Incorrect Answers

A:

When eager inclusion of the entire catalog wastes context and confuses selection, the integration can advertise compact discovery metadata and load detailed definitions on demand; yet it would not correct the observed weakness: uncontrolled retries exhaust quotas and conceal a failed precondition.

B:

When mixing reading and mutation in one tool overgrants routine sessions, the integration can expose read-only access independently from privileged changes; but this remedy operates on another boundary, not the case where uncontrolled retries exhaust quotas and conceal a failed precondition.

C:

The alternative reduces the risk that overlapping names lead to arbitrary selection and inconsistent answers because the integration can retire or consolidate redundant capabilities with a responsible owner; however, the legal operations defect follows the path in which uncontrolled retries exhaust quotas and conceal a failed precondition.

Question 146

A compliance assurance team is selecting the next improvement for its AI service; teams propose many attractive pilots without common selection criteria. What design choice has the strongest direct justification?

  1. Introduce Workflow observation interviews
  2. Introduce Use-case prioritization workshop
  3. Introduce Stakeholder and decision-rights mapping
  4. Introduce Functional versus nonfunctional split

Explanation

Correct Answer

B:

Under the stated constraints, the delivery team can jointly rank candidates by value, feasibility, harm and data readiness. That directly addresses the operational failure without relying on a generic prompt reminder. Production depends on scoring rubric and representatives from delivery and risk teams; the result to monitor is agreed priority scores and expected benefits for shortlisted uses.

Incorrect Answers

A:

When sponsors describe an ideal process that differs from daily operations, the delivery team can observe actual exception handling and frontline work before mapping automation; whereas the compliance assurance case must guard against the outcome that political priority alone may select a low-return high-risk deployment.

C:

When product, security and legal teams disagree on who can approve release, the delivery team can define accountable decision owners and consultation rights early; the current decision instead depends on a specific capability: the delivery team can jointly rank candidates by value, feasibility, harm and data readiness.

D:

For a workload where a proposal lists features but omits response time and recovery requirements, the delivery team can capture distinct behavior, latency, security, availability and data constraints; whereas the compliance assurance case must guard against the outcome that political priority alone may select a low-return high-risk deployment.

Question 147

In a logistics coordination modernization project, proposed external transfers could produce irreversible customer harm. Which approach best fits this specific operating constraint?

  1. Deploy Output privacy filter
  2. Deploy Untrusted retrieved-content isolation
  3. Deploy High-risk human approval gate
  4. Deploy Tool-execution policy gate

Explanation

Correct Answer

C:

The mechanism is specific: the control layer can stop before execution and require authorized human confirmation. A credible implementation requires reviewer authority, timeout path and approval evidence, while its acceptance test monitors unapproved high-risk actions prevented during simulations. The alternative approaches leave the risk that after-the-fact logs cannot reverse a completed sensitive transfer.

Incorrect Answers

A:

When generated summaries may echo sensitive identifiers unnecessary for recipients, the control layer can enforce recipient-specific redaction and validated output fields; this rollout still needs the following guarantee: the control layer can stop before execution and require authorized human confirmation.

B:

For a workload where supplier documents include embedded instructions masquerading as system messages, the control layer can treat retrieved text as evidence and prohibit it from changing tool policy; the stated requirement remains unmet: the control layer can stop before execution and require authorized human confirmation.

D:

When assistants propose privileged write operations in customer systems, the control layer can require server-side authorization and business-policy checks before each side effect; this rollout still needs the following guarantee: the control layer can stop before execution and require authorized human confirmation.

Question 148

A financial operations technical leadership group is designing a Claude-supported workflow because agent integration must respect access restrictions for multiple tenants. Which architectural decision most directly meets that requirement?

  1. Security policy violation rate
  2. Cost per resolved ticket
  3. Unsafe-action escape rate
  4. Grounded-claim verification rate

Explanation

Correct Answer

A:

This control changes the relevant system boundary: the evaluation process can evaluate denied and erroneously permitted requests against authoritative policy. The team should establish labeled authorization test cases before enabling it, then evaluate false-allow and false-deny rates by scope so that a successful business task may still leak restricted records is not hidden by fluent outputs.

Incorrect Answers

B:

If model usage drops while human rework unexpectedly grows, the evaluation process can include review, retries and inference in outcome-based unit economics; whereas the financial operations case must guard against the outcome that a successful business task may still leak restricted records.

C:

When an assistant can invoke sensitive operations through tools, the evaluation process can measure disallowed actions that reach a protected side-effecting endpoint; the alternative leaves the central requirement unproven: the evaluation process can evaluate denied and erroneously permitted requests against authoritative policy.

D:

When the system summarizes regulated policies with citations, the evaluation process can score whether factual claims are supported by cited source spans; it does not ensure the required behavior: the evaluation process can evaluate denied and erroneously permitted requests against authoritative policy.

Question 149

The defining problem for an enterprise knowledge management pilot is that agent-assisted fixes work on one laptop but fail in CI. Which proposed approach addresses it most precisely?

  1. Deploy Hook-based policy checks
  2. Deploy Shared MCP connector governance
  3. Deploy Reproducible development environment
  4. Deploy Repository-level project instructions

Explanation

Correct Answer

C:

The proposed control fits because the engineering workflow can pin dependencies, tooling and environment assumptions across contributors. Its production prerequisites include versioned setup scripts and build reproducibility checks; check agent-generated changes passing clean-environment builds on representative tasks. Without this boundary, ambient local settings conceal incompatibilities in shared builds.

Incorrect Answers

A:

When code generation sometimes bypasses mandatory lint and secret scans, the engineering workflow can run deterministic checks at controlled workflow boundaries; whereas the enterprise knowledge management case must guard against the outcome that ambient local settings conceal incompatibilities in shared builds.

B:

For a workload where development teams connect tools with overlapping credentials and contracts, the engineering workflow can register reviewed MCP servers with clear ownership and allowed scopes; the stated scenario depends on a different mechanism: the engineering workflow can pin dependencies, tooling and environment assumptions across contributors.

D:

If developers get inconsistent coding behavior across branches and members, the engineering workflow can version shared Claude Code guidance alongside the project and review changes; whereas the enterprise knowledge management case must guard against the outcome that ambient local settings conceal incompatibilities in shared builds.

Question 150

The education services project must justify retaining Capped tool-loop execution to a skeptical review committee. What specific failure does this control prevent?

  1. The possibility that uncontrolled retries exhaust quotas and conceal a failed precondition
  2. The possibility that eager inclusion of the entire catalog wastes context and confuses selection
  3. The possibility that overlapping names lead to arbitrary selection and inconsistent answers
  4. The possibility that broad tool visibility increases selection error and attack surface

Explanation

Correct Answer

A:

This is the direct residual hazard: uncontrolled retries exhaust quotas and conceal a failed precondition. The omitted intervention matters because the integration can enforce retry, token and call budgets with a deliberate stop path. A reviewer should demand loop policy and classified tool-error responses and measure attempts per failed case and bounded-failure completion rate before treating the risk as controlled.

Incorrect Answers

B:

The other exposure arises when a complex connector ecosystem cannot fit full tool descriptions in each request: eager inclusion of the entire catalog wastes context and confuses selection; whereas removing Capped tool-loop execution primarily exposes the risk that uncontrolled retries exhaust quotas and conceal a failed precondition.

C:

If several plugins expose overlapping outdated customer lookup functions, overlapping names lead to arbitrary selection and inconsistent answers; but this is not the immediate loss associated with omitting Capped tool-loop execution: uncontrolled retries exhaust quotas and conceal a failed precondition.

D:

The other exposure arises when a research agent is offered dozens of irrelevant mutation tools: broad tool visibility increases selection error and attack surface; however, the named safeguard addresses the distinct hazard that uncontrolled retries exhaust quotas and conceal a failed precondition.

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