Anthropic CCAR-P (Claude Certified Architect - Professional) Practice Test 93

Question 920

A manufacturing quality service owner is reviewing Token-economics telemetry after a canary rollout. The critical regression would occur if request totals alone cannot isolate token-heavy work. What observable acceptance signal belongs in the dashboard?

  1. Track carried-forward constraint accuracy across long conversations
  2. Track answer support recall under a fixed token budget
  3. Track context-overflow rate at high-percentile request size
  4. Track tokens and dollars per accepted outcome by category

Explanation

Correct Answer

D:

An appropriate gate measures tokens and dollars per accepted outcome by category. It tests the mechanism in which the model configuration can record uncached input, cache reads, output and tool-related usage per task, rather than a cosmetic demonstration. That distinction matters because request totals alone cannot isolate token-heavy work.

Incorrect Answers

A:

Where a long case management dialogue continually replays stale turns, a relevant check would establish whether the model configuration can preserve verified decisions and current state while pruning redundant history; by contrast, the manufacturing quality release needs evidence from tokens and dollars per accepted outcome by category.

B:

The proposed measure would suit a setting where a case includes more background than the model can productively use, since the model configuration can rank excerpts by task relevance and retain traceable source references; however, that proxy would not demonstrate tokens and dollars per accepted outcome by category for the selected intervention.

C:

Where templates unexpectedly exceed request limits after tool results are appended, a relevant check would establish whether the model configuration can estimate input, tool-result and output allowances before sending requests; but the acceptance gate must instead inspect tokens and dollars per accepted outcome by category to detect the targeted failure.

Question 921

The logistics coordination release team needs a numerical gate for Exception-first task splitting. Rather than equating deployment with success, it wants to catch cases in which an average-case prompt can hide high-impact failure cases. Which indicator should it track?

  1. Audit failed handoffs caused by missing upstream artifacts
  2. Audit exception detection recall and unsafe bypass counts
  3. Audit retained key-fact recall and final contradiction rate
  4. Audit citation support rate on final findings

Explanation

Correct Answer

B:

The acceptance measure focuses on exception detection recall and unsafe bypass counts at the task or workflow boundary. Model fluency would not establish whether the architecture can separate normal and exception branches with stronger checks for the latter, and it could conceal cases where an average-case prompt can hide high-impact failure cases.

Incorrect Answers

A:

The proposed measure would suit a setting where downstream analyses sometimes run before upstream data cleansing finishes, since the architecture can model the task as a DAG and execute nodes only when prerequisites are satisfied; however, that proxy would not demonstrate exception detection recall and unsafe bypass counts for the selected intervention.

C:

Where thousands of independent notes exceed practical context budgets, a relevant check would establish whether the architecture can summarize bounded chunks then aggregate their structured findings; by contrast, the logistics coordination release needs evidence from exception detection recall and unsafe bypass counts.

D:

The alternative indicator checks whether the architecture can gather and validate evidence before asking Claude to draw conclusions when complex risk reports contain claims from many separate source systems; however, that proxy would not demonstrate exception detection recall and unsafe bypass counts for the selected intervention.

Question 922

A healthcare operations audit challenges the need for Reproduction-first debugging. Which risk would remain unmanaged if its function were not provided elsewhere?

  1. The danger that changing the system prompt cannot repair invalid connector payloads
  2. The danger that debugging in place keeps all users exposed to the broken state
  3. The danger that a one-off patch without tests allows the same bug to return
  4. The danger that top-level request durations cannot identify the responsible dependency
  5. The danger that guessing from one screenshot hides true upstream causes

Explanation

Correct Answer

E:

The relevant failure mode is that guessing from one screenshot hides true upstream causes. The existing design can prevent it when the engineering workflow can capture minimal reproducible requests and traces before changing behavior. The team needs redacted traces and environment version details and must monitor reproducible incident rate and time to confirmed root cause to support any claim of equivalent protection.

Incorrect Answers

A:

When end users see failures only when one MCP connector returns malformed results, changing the system prompt cannot repair invalid connector payloads; whereas removing Reproduction-first debugging primarily exposes the risk that guessing from one screenshot hides true upstream causes.

B:

If a newly enabled hook or agent config breaks developer workflows, debugging in place keeps all users exposed to the broken state; but this is not the immediate loss associated with omitting Reproduction-first debugging: guessing from one screenshot hides true upstream causes.

C:

If a production agent repeatedly reintroduces a previously fixed defect, a one-off patch without tests allows the same bug to return; however, the named safeguard addresses the distinct hazard that guessing from one screenshot hides true upstream causes.

D:

The other exposure arises when a slow multi-tool run obscures its longest step: top-level request durations cannot identify the responsible dependency. Omitting the requested control instead leaves the more immediate risk that guessing from one screenshot hides true upstream causes; whereas removing Reproduction-first debugging primarily exposes the risk that guessing from one screenshot hides true upstream causes.

Question 923

A media licensing release candidate fails review after assessors observe that blindly increasing concurrency can overload downstream quotas. What intervention should take priority?

  1. Enforce Cache reuse measurement
  2. Enforce Adaptive tool timeout budget
  3. Enforce Parallelizable workflow optimization
  4. Enforce Result-driven model escalation

Explanation

Correct Answer

C:

This is the causal correction: the evaluation process can parallelize work while joining outputs under a bounded concurrency cap. With independence proof and capacity testing in place, the team can test overall task latency and quota error rate and show that the problematic case in which blindly increasing concurrency can overload downstream quotas has been addressed.

Incorrect Answers

A:

Where a common long policy prefix appears in nearly every request, the evaluation process can measure effective cache reads and preserve stable prefix ordering; but this remedy operates on another boundary, not the case where blindly increasing concurrency can overload downstream quotas.

B:

When one stalled low-priority tool can dominate overall response time, the evaluation process can use per-dependency deadlines and continue only if required evidence is available; whereas the media licensing defect follows the path in which blindly increasing concurrency can overload downstream quotas.

D:

This addresses the failure where swapping every request to a premium tier increases cost without matching benefit, because the evaluation process can raise model capacity selectively after evidence-based difficulty checks; but this remedy operates on another boundary, not the case where blindly increasing concurrency can overload downstream quotas.

Question 924

A legal operations incident exposes a specific weakness: duplicating bespoke connectors across clients multiplies maintenance. Which architectural correction addresses the cause rather than masking the symptom?

  1. Introduce Agent-to-agent task delegation
  2. Introduce MCP server as standardized tool boundary
  3. Introduce Direct Messages API service integration
  4. Introduce CLI invocation in controlled automation

Explanation

Correct Answer

B:

The remediation acts at the relevant design boundary: the integration can expose reusable typed tools and resources through a controlled MCP server. That counteracts the observed hazard that duplicating bespoke connectors across clients multiplies maintenance. Rollout should include server ownership, tool schemas and access control, with regression evidence from authorized tool compatibility and change success across clients.

Incorrect Answers

A:

Where separate services own distinct autonomous subworkflows and outputs, the integration can communicate through explicit task and artifact contracts between agents; whereas the legal operations defect follows the path in which duplicating bespoke connectors across clients multiplies maintenance.

C:

When adding an agent protocol for a single inference boundary adds unnecessary complexity, the integration can call the model through the application service with controlled retries and validation; but this remedy operates on another boundary, not the case where duplicating bespoke connectors across clients multiplies maintenance.

D:

The alternative reduces the risk that a persistent cross-agent protocol is excess infrastructure for one local job because the integration can use a noninteractive CLI process with bounded permissions and captured exit status; however, the legal operations defect follows the path in which duplicating bespoke connectors across clients multiplies maintenance.

Question 925

The defining problem for a healthcare operations pilot is that stakeholders cannot tell which architectural control satisfies which need. Which proposed approach addresses it most precisely?

  1. Implement Interface and error contract
  2. Implement End-to-end data-flow diagram
  3. Implement Prompt and model configuration manifest
  4. Implement Decision-to-requirement traceability

Explanation

Correct Answer

D:

The intervention acts on the system behavior directly: the delivery team can link decisions to acceptance tests and authoritative requirement IDs. Before rollout, verify stable requirement registry and testing ownership. The relevant success test is requirements with assigned controls and passing evidence; otherwise documentation volume without traceability is hard to audit or change.

Incorrect Answers

A:

When separate teams implement tool providers and consumers inconsistently, the delivery team can publish request schemas, response semantics, error categories and retries; the alternative cannot replace this key requirement: the delivery team can link decisions to acceptance tests and authoritative requirement IDs.

B:

When privacy reviewers cannot see where source documents or traces travel, the delivery team can show all data stores, transformations, transfers and trust boundaries; whereas the healthcare operations case must guard against the outcome that documentation volume without traceability is hard to audit or change.

C:

For a workload where multiple environments produce divergent behavior despite using one codebase, the delivery team can version prompts, model choices, tools and safety settings together; that does not provide the capability actually needed: the delivery team can link decisions to acceptance tests and authoritative requirement IDs.

Question 926

In a research administration modernization project, classification only needs a narrow subset of full case records. Which approach best fits this specific operating constraint?

  1. Retention and erasure workflow
  2. Data residency deployment review
  3. FedRAMP authorization boundary check
  4. Least-data-required architecture

Explanation

Correct Answer

D:

The architectural distinction is that the control layer can transform and suppress unnecessary personal fields before invoking tools. The decision should be implemented with purpose-specific field inventory and data handlers and evaluated against minimization compliance and utility impact by field; otherwise, sending complete files for convenience increases protected data exposure.

Incorrect Answers

A:

If a user requests deletion of personal data from model-related application stores, the control layer can map and process erasure across prompts, logs, caches and derived indexes as applicable; whereas the research administration case must guard against the outcome that sending complete files for convenience increases protected data exposure.

B:

For a workload where contract terms restrict where customer records may be processed or stored, the control layer can verify regional processing, logs, backups and support access as a whole; the missing operational behavior is specific: the control layer can transform and suppress unnecessary personal fields before invoking tools.

C:

If a US federal workload has explicit authorized-environment constraints, the control layer can validate the complete deployment and service boundary against agency requirements; the current decision instead depends on a specific capability: the control layer can transform and suppress unnecessary personal fields before invoking tools.

Question 927

Following a regression in enterprise knowledge management, the team establishes that an average-case prompt can hide high-impact failure cases. Which proposed redesign most directly prevents recurrence?

  1. Enforce Exception-first task splitting
  2. Enforce Map-reduce summarization
  3. Enforce Dependency graph scheduling
  4. Enforce Hierarchical work-breakdown structure
  5. Enforce Reconciliation of conflicting sources
  6. Enforce Evidence-gathering before synthesis

Explanation

Correct Answer

A:

The reported failure calls for the control under which the architecture can separate normal and exception branches with stronger checks for the latter. A prompt-only reminder or unrelated metric would not address the risk that an average-case prompt can hide high-impact failure cases. Evidence of resolution requires an exception taxonomy and differentiated routing policy and an improvement in exception detection recall and unsafe bypass counts.

Incorrect Answers

B:

Although the alternative reduces the risk that concatenating every note increases cost and can truncate important evidence because the architecture can summarize bounded chunks then aggregate their structured findings; yet it would not correct the observed weakness: an average-case prompt can hide high-impact failure cases.

C:

This addresses the failure where independent worker retries cannot fix an invalid dependency order, because the architecture can model the task as a DAG and execute nodes only when prerequisites are satisfied; but this remedy operates on another boundary, not the case where an average-case prompt can hide high-impact failure cases.

D:

This addresses the failure where one giant request conceals intermediate failures and unclear ownership, because the architecture can break the outcome into testable components with explicit dependencies; whereas the enterprise knowledge management defect follows the path in which an average-case prompt can hide high-impact failure cases.

E:

When direct summarization may arbitrarily select one side without noting dispute, the architecture can separate conflict detection and resolution from answer drafting; yet it would not correct the observed weakness: an average-case prompt can hide high-impact failure cases.

F:

Where complex risk reports contain claims from many separate source systems, the architecture can gather and validate evidence before asking Claude to draw conclusions; but this remedy operates on another boundary, not the case where an average-case prompt can hide high-impact failure cases.

Question 928

A procurement services pilot cannot be accepted because request totals alone cannot isolate token-heavy work. The owner can change one primary architectural control; what should it be?

  1. Enforce Exact token budget accounting
  2. Enforce Source-priority truncation policy
  3. Enforce Rolling conversation summary with state
  4. Enforce Token-economics telemetry

Explanation

Correct Answer

D:

This is the causal correction: the model configuration can record uncached input, cache reads, output and tool-related usage per task. With request-level usage accounting and trace identifiers in place, the team can test tokens and dollars per accepted outcome by category and show that the problematic case in which request totals alone cannot isolate token-heavy work has been addressed.

Incorrect Answers

A:

Where templates unexpectedly exceed request limits after tool results are appended, the model configuration can estimate input, tool-result and output allowances before sending requests; yet it would not correct the observed weakness: request totals alone cannot isolate token-heavy work.

B:

Where a long research input must retain governing rules and current evidence, the model configuration can truncate lowest-priority material while protecting essential instructions and citations; but this remedy operates on another boundary, not the case where request totals alone cannot isolate token-heavy work.

C:

Where a long case management dialogue continually replays stale turns, the model configuration can preserve verified decisions and current state while pruning redundant history; whereas the procurement services defect follows the path in which request totals alone cannot isolate token-heavy work.

Question 929

The manufacturing quality delivery group must make the following claim executable: the evaluation process can parallelize work while joining outputs under a bounded concurrency cap. Which TWO measures pair the right intervention with the right readiness condition? Choose TWO.

  1. Implement Parallelizable workflow optimization
  2. Establish relevance evaluation and source-preserving deduplication
  3. Establish independence proof and capacity testing
  4. Track accepted outcome cost by escalation bucket
  5. Implement Inference request batching where latency permits

Explanation

Correct Answers

A:

The architecture component is justified by its behavior: the evaluation process can parallelize work while joining outputs under a bounded concurrency cap. It targets the exposure that blindly increasing concurrency can overload downstream quotas, while the complementary selected actions provide readiness and, for a three-part package, evidence.

C:

The prerequisite is independence proof and capacity testing. It provides the operating preparation for a system in which the evaluation process can parallelize work while joining outputs under a bounded concurrency cap; without this setup, the planned safeguard has not been demonstrated as deployable.

Incorrect Answers

B:

The alternative setup enables a different mechanism: the evaluation process can deduplicate and rank passages before model invocation; whereas this intervention requires independence proof and capacity testing as its operating precondition.

D:

This test would validate a different benefit: the evaluation process can raise model capacity selectively after evidence-based difficulty checks; but that indicator does not verify overall task latency and quota error rate for the selected architecture.

E:

This design choice is aimed at cases where offline processing runs thousands of independent requests nightly; it is effective because the evaluation process can group asynchronous requests using the available batch workflow. The rollout described is instead governed by the case where; however, the intended intervention requires that the evaluation process can parallelize work while joining outputs under a bounded concurrency cap.

Question 930

A legal operations rollout must satisfy this design premise: the integration can expose reusable typed tools and resources through a controlled MCP server. Choose TWO linked items that form a workable architecture change and its essential preparation. Choose TWO.

  1. Implement CLI invocation in controlled automation
  2. Track event-to-action latency and duplicate-event rate
  3. Establish server ownership, tool schemas and access control
  4. Establish identity propagation and contract versioning
  5. Implement MCP server as standardized tool boundary

Explanation

Correct Answers

C:

This step creates the enabling basis of server ownership, tool schemas and access control. That preparation supports the mechanism that the integration can expose reusable typed tools and resources through a controlled MCP server; merely choosing the control does not prove its configuration or authority is ready.

E:

The control itself makes the following capability possible: the integration can expose reusable typed tools and resources through a controlled MCP server. That matters when duplicating bespoke connectors across clients multiplies maintenance; it does not replace the required enabling condition or outcome test.

Incorrect Answers

A:

This design choice is aimed at cases where a development workflow needs a reproducible one-off command in CI; it is effective because the integration can use a noninteractive CLI process with bounded permissions and captured exit status. The rollout described is instead governed by the case where many Claude clients need access to the same governed business capability and requires a different primary intervention; however, the intended intervention requires that the integration can expose reusable typed tools and resources through a controlled MCP server.

B:

This test would validate a different benefit: the integration can accept validated inbound events and enqueue bounded downstream tasks; but that indicator does not verify authorized tool compatibility and change success across clients for the selected architecture.

D:

The alternative setup enables a different mechanism: the integration can communicate through explicit task and artifact contracts between agents; whereas this intervention requires server ownership, tool schemas and access control as its operating precondition.

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