Anthropic CCDV-F (Claude Certified Developer - Foundations) Practice Test 10

Question 90

A planner assigns database writes to a read-only specialist. What should coordination do?

  1. Treat the planner’s assignment as a permission grant.
  2. Tell the specialist to bypass its read-only credentials.
  3. Pretend the specialist completed the write after reading data.
  4. Check worker capabilities before assignment and route authorized write work appropriately.

Explanation

Correct Answer

D:

A role description must match executable permissions. Capability-aware assignment prevents plans that cannot be carried out by their designated workers.

Incorrect Answers

A:

Planning intent cannot override the worker’s actual access restrictions.

B:

That violates the intended boundary rather than correcting the assignment.

C:

Reading does not perform mutation, and reporting otherwise fabricates an outcome.

Question 91

A team changes agent frameworks while retaining Claude. What should its compatibility review cover?

  1. Message conversion, tool contracts, state semantics, and error handling.
  2. Nothing, because framework abstractions are interchangeable.
  3. Only the spelling of the model family.
  4. Only whether both packages use Python.

Explanation

Correct Answer

A:

Frameworks can differ in orchestration and representation even when they call the same model. Those boundaries need explicit tests.

Incorrect Answers

B:

Abstraction APIs can conceal important differences; migration still needs verification of the application’s required behavior.

C:

Keeping the model name does not preserve the framework’s execution semantics.

D:

A shared language does not guarantee compatible messages, persistence, or control-flow behavior.

Question 92

A pure risk function takes category and region. A cache keys results only by category, so two regions receive the same result incorrectly. What fixes the memoization contract?

  1. Include every result-determining argument in the cache key.
  2. Shorten the cache lifetime while keeping the same key.
  3. Use the most recently requested region for all cached categories.
  4. Add the request timestamp to the key but continue omitting region.

Explanation

Correct Answer

A:

A cached value is reusable only for inputs equivalent under the function's contract. Adding region separates computations that can legitimately produce different answers even when category matches.

Incorrect Answers

B:

A shorter lifetime reduces how long an incorrect entry remains, but requests for different regions can still collide during that interval.

C:

That makes results depend on unrelated traffic rather than the current function inputs. It does not restore the function's deterministic input-output relationship.

D:

A timestamp does not encode the omitted function argument. It can reduce reuse without establishing the correct category-region mapping.

Question 93

A workflow’s “payment settled” branch must run only after settlement is confirmed. Which condition should control that branch?

  1. The user’s positive tone after submitting payment.
  2. A fixed delay assumed to be sufficient for every payment.
  3. A verified settlement status from the payment system.
  4. The assistant’s statement that payment should succeed.

Explanation

Correct Answer

C:

The transition depends on an external business event. Explicit status evidence is stronger than a model prediction about what probably happened.

Incorrect Answers

A:

Tone does not prove the payment system completed the transaction.

B:

Elapsed time is not guaranteed settlement; failures or pending states can persist beyond the delay.

D:

An expectation is not confirmation of settlement and cannot satisfy the branch precondition.

Question 94

A service averages one second but frequently takes ten seconds for some customers. Which additional requirement captures this problem?

  1. A percentile latency target under representative load.
  2. Only a larger daily request quota.
  3. Only a lower average token count.
  4. A guarantee that every answer has equal length.

Explanation

Correct Answer

A:

Tail metrics expose slow experiences hidden by averages and should be measured with realistic concurrency. A percentile target defines what proportion of users may experience delays above the threshold, which an average alone cannot express.

Incorrect Answers

B:

Quota describes allowed volume, not the time users wait for individual responses.

C:

Token volume can influence latency but does not directly measure the slow customer experiences described.

D:

Uniform text length neither measures nor guarantees acceptable response latency.

Question 95

Misrouting urgent cases is far more costly than sending ordinary cases for review. What should evaluation reflect?

  1. Only the number of labels available.
  2. Identical business impact for every wrong prediction.
  3. Separate error costs and recall requirements for urgent cases.
  4. Only total accuracy across all labels.

Explanation

Correct Answer

C:

A cost-sensitive evaluation reflects the asymmetric harm rather than hiding critical misses inside overall accuracy.

Incorrect Answers

A:

Label count says nothing about which errors matter or how often they occur.

B:

That contradicts the stated asymmetry and can select an unsuitable classifier.

D:

A dominant ordinary class can produce high accuracy while urgent cases are frequently missed.

Question 96

An extraction service routes uncertain outputs to humans. What operational requirement must accompany this design?

  1. A rule suppressing uncertainty to keep queues empty.
  2. A requirement that every input uses maximum reasoning.
  3. A review-volume and turnaround capacity target.
  4. An assumption that review is instantaneous.

Explanation

Correct Answer

C:

Escalation only works when reviewers can handle the expected workload within the required time. A review queue can become the system's bottleneck even when inference is fast, so staffing and escalation volume belong in the capacity plan.

Incorrect Answers

A:

Hiding uncertainty improves queue statistics by admitting unsupported results, not by improving quality.

B:

More inference does not establish available human-review capacity or turnaround.

D:

Human queues introduce real capacity and delay constraints that the application must account for.

Question 97

A customer restricts where data may be processed. What must be clarified before selecting an integration?

  1. Only the developer’s physical location.
  2. The permitted processing locations and all relevant provider/tool data paths.
  3. Only the browser’s language setting.
  4. A region name inserted into the prompt.

Explanation

Correct Answer

B:

The restriction applies to actual processing flows, including external tools, rather than merely the frontend’s hosting location.

Incorrect Answers

A:

The developer’s location does not identify the deployed services’ processing boundaries.

C:

Display language does not establish where backend inference or tool processing occurs.

D:

Text instructions do not configure infrastructure placement or provider routing.

Question 98

A model outage must not prevent customers from submitting support tickets. Which requirement guides the fallback?

  1. Invent summaries and claim the model succeeded.
  2. Retry indefinitely before accepting a ticket.
  3. Preserve ticket intake independently and defer optional AI enrichment.
  4. Reject all submissions when summarization fails.

Explanation

Correct Answer

C:

The essential business operation can continue while model-dependent assistance is temporarily unavailable. Specify which features remain usable and how unavailable actions are communicated; a fallback should preserve the promised minimum behavior.

Incorrect Answers

A:

Fabricated enrichment hides the outage and introduces unsupported ticket information.

B:

Unbounded waiting blocks the required submission capability during the outage.

D:

This couples basic intake to an optional enrichment and violates the stated availability goal.

Question 99

A pricing assistant must use catalog changes within five minutes. What requirement follows?

  1. Choose a model whose name sounds recent.
  2. Refresh the browser page while reusing the same stale pricing snapshot.
  3. Increase response creativity.
  4. Define retrieval and cache freshness against that five-minute boundary.

Explanation

Correct Answer

D:

The data pipeline must make updates visible within the allowed age, not rely on model training knowledge. Define the maximum acceptable age of the pricing evidence and how the application refreshes it before answering time-sensitive questions.

Incorrect Answers

A:

Model branding does not guarantee awareness of private catalog updates.

B:

A UI refresh does not update the evidence if the backend continues returning an outdated snapshot.

C:

Sampling behavior does not refresh catalog data or establish its timestamp.

Question 100

An assistant supports employees and external customers. What should requirements specify before exposing the same tools to both?

  1. That external users receive shorter answers.
  2. Which operations and records each authenticated role may access.
  3. That every user receives the same retrieved document set.
  4. That users describe their own privileges in prompts.

Explanation

Correct Answer

B:

Role-specific authorization defines the actual capability boundary and informs tool exposure and backend checks.

Incorrect Answers

A:

Answer length does not restrict the underlying operations or records they can access.

C:

Uniform retrieval can expose documents beyond individual entitlements; access must follow the authenticated user.

D:

Self-asserted authority is untrusted input and cannot establish actual entitlements.

Claude Certified Developer - Foundations (CCDV-F) Practice Tests