Anthropic CCAR-P (Claude Certified Architect - Professional) Practice Test 22

Question 210

A client receives different failures from external calls. Which cases generally warrant considering a bounded retry after checking the service contract? Choose TWO.

  1. An authorization denial for the current identity
  2. A transient service-unavailable response
  3. A validation error for a missing required field
  4. A rate-limit response with retry guidance
  5. A request for an unsupported operation version

Explanation

Correct Answers

B:

Temporary unavailability may resolve, so a bounded retry can be appropriate. The policy still needs deadlines and protection against repeating unsafe side effects.

D:

The service explicitly signals a capacity condition and often a delay. Respecting that guidance can support recovery without an immediate retry storm.

Incorrect Answers

A:

Repeated requests do not grant missing permission. The identity or approved access arrangement must be corrected before the operation can succeed.

C:

An unchanged malformed request is likely to fail again. The caller must correct the payload rather than retry it blindly.

E:

An unsupported contract is not repaired by waiting. The client must use a supported interface or handle the incompatibility explicitly.

Question 211

One user request launches 100 independent retrieval calls and exceeds the downstream service's allowed concurrency. What should the integration implement?

  1. A cost estimate calculated after all downstream requests are launched
  2. A bounded worker pool respecting downstream limits
  3. Sequential retries of all 100 after every failure
  4. One new connection per call with no shared limit

Explanation

Correct Answer

B:

Independent work can still require admission control. Concurrency limits preserve useful parallelism without overwhelming the dependency or exhausting local resources.

Incorrect Answers

A:

An estimate can inform planning but does not enforce concurrent admission. The downstream limit must constrain how many calls are active at once.

C:

Restarting all work duplicates successful calls and can create large recovery delays. The executor should bound and track individual operations.

D:

Additional connections do not remove the downstream constraint. They can increase contention and trigger more failures.

Question 212

A supplier tool returns 1,234.50 in one locale and 1.234,50 in another. What should the integration contract do?

  1. Strip all punctuation and use the remaining integer
  2. Assume every comma is a decimal separator
  3. Use a canonical numeric representation with explicit currency
  4. Ask the model to select the larger interpretation

Explanation

Correct Answer

C:

Locale-dependent punctuation can change interpretation. A machine contract should define the amount representation and currency rather than rely on display conventions.

Incorrect Answers

A:

Removing separators loses scale information and can multiply the amount incorrectly. The parser needs the actual representation rules.

B:

That interpretation is wrong for one of the examples. A universal punctuation guess does not handle locale-dependent formatting safely.

D:

Choosing by magnitude is not evidence of the intended value. Numeric semantics must come from the tool's documented contract.

Question 213

A new payment-status tool is being integrated. Which checks are necessary before relying on it? Choose THREE.

  1. Require pending and settled payments to return the same status value
  2. Test only the tool's discovery metadata and omit execution cases
  3. Skip negative cases when the happy path works
  4. Exercise timeouts and unavailable-service responses
  5. Test authorization with permitted and denied identities
  6. Verify argument validation and result schema behavior

Explanation

Correct Answers

D:

Failure behavior is part of the integration contract. The application must distinguish unknown status from a confirmed payment result.

E:

Successful privileged calls alone do not establish access control. Negative cases show whether the service rejects requests outside the caller's entitlement.

F:

A usable integration must handle valid and invalid inputs consistently. The client needs evidence that it can interpret the returned fields and errors.

Incorrect Answers

A:

These states have different business meanings. Collapsing them would prevent the caller from using the tool to determine actual payment status.

B:

Discovery confirms that the tool is advertised, not that its arguments, authority checks, or failure behavior work. Acceptance needs relevant execution evidence.

C:

Invalid input and denied access are common failure boundaries. A working success case cannot establish their correctness.

Question 214

A retriever sends confidential candidate passages to an unapproved third-party reranker. Final answers stay inside the company. What is the integration issue?

  1. Only the reranker's response length needs limiting
  2. Renaming source documents makes all passage content nonconfidential
  3. The reranking step itself discloses protected content externally
  4. There is no issue because the final answer stays internal

Explanation

Correct Answer

C:

Data flow includes intermediate services, not only final output destinations. The candidate text must be permitted for every processor that receives it.

Incorrect Answers

A:

Length controls returned data, not the confidential passages sent to the service. The outbound processing relationship is the relevant boundary.

B:

Removing filenames does not necessarily remove sensitive facts from the text. The integration must assess the actual information being transmitted.

D:

The external transmission already occurred before the answer was generated. Final-output location does not erase that intermediate disclosure.

Question 215

An inventory tool returns a quantity without saying when it was observed. Users need near-current stock. What should its response contract include?

  1. The length of the database query
  2. The model version that formats the response
  3. An observation timestamp or explicit freshness guarantee
  4. Only the time when the adapter formatted the response

Explanation

Correct Answer

C:

The consumer needs evidence that the quantity meets the service's freshness requirement. A bare number cannot establish whether it reflects current inventory.

Incorrect Answers

A:

Query length says nothing about how current the returned value is. It cannot substitute for timestamp or source freshness semantics.

B:

Formatting provenance does not establish the age of the underlying stock observation. Data freshness belongs to the source-result contract.

D:

Formatting time can be newer than the underlying observation. Freshness must identify when the source state was actually measured or its valid guarantee.

Question 216

An operations workflow must run a documented local utility in an isolated environment with no remote endpoint. Which mechanism best matches the available surface?

  1. A constrained CLI adapter using the utility's supported interface
  2. Browser automation assuming the command-line utility also exposes a GUI
  3. An HTTP adapter configured from an assumed URL for the utility
  4. Replace the local utility with generated output in the same text format

Explanation

Correct Answer

A:

The required capability already exists locally. An adapter can enforce arguments, permissions, and structured result handling without inventing a remote service.

Incorrect Answers

B:

A local executable does not imply a browser-accessible user interface. The documented surface in this scenario is its CLI.

C:

The scenario explicitly provides no remote endpoint. Assuming one exists creates an integration contract unsupported by the available utility.

D:

Matching the utility's output appearance does not perform its required operation. The integration must invoke the available supported mechanism and verify its result.

Question 217

An evaluation randomly splits pages, placing pages from the same customer contract in both development and test sets. What leakage control is needed?

  1. Stratify pages by answer label while allowing contracts in both sets
  2. Shuffle the page order again
  3. Deduplicate exact page text but leave different pages of one contract in both sets
  4. Split by contract or customer group as appropriate

Explanation

Correct Answer

D:

Related pages share facts and structure. Group separation prevents familiarity with one document from inflating apparent performance on its near-identical companion pages.

Incorrect Answers

A:

Label balance does not remove shared contract content across partitions. Group boundaries must be respected to avoid that source of leakage.

B:

Another shuffle preserves the same cross-set relationships. Random order does not remove information shared by pages from one contract.

C:

Distinct pages can still share contract-specific facts and structure. Exact-text deduplication alone does not enforce the needed group separation.

Question 218

A forecasting assistant is tested on records that predate many examples used during development. Production must handle future policy changes. Which split better tests that requirement?

  1. Randomly mix all dates without checking chronology
  2. Test only the oldest available records
  3. Remove every timestamp before splitting
  4. Develop on earlier periods and evaluate on later unseen periods

Explanation

Correct Answer

D:

A temporal holdout approximates deployment on future inputs. It helps expose reliance on patterns or policies that change over time.

Incorrect Answers

A:

Random mixing can let future information influence development. That can overstate performance for a genuinely forward-looking deployment.

B:

Old examples may be useful for compatibility, but they do not represent the future-period challenge identified in the requirement.

C:

Deleting dates prevents enforcing the temporal boundary. It does not eliminate the underlying chronological relationships between the records.

Question 219

Two experts assign different correct categories to many evaluation cases. What should happen before using the scores as a release gate?

  1. Resolve rubric ambiguity and adjudicate disputed labels
  2. Choose whichever expert gives higher model scores
  3. Average categorical labels numerically
  4. Discard every disagreement without recording it

Explanation

Correct Answer

A:

An unstable reference makes model scores difficult to interpret. Disagreement may reveal unclear business rules rather than a model defect.

Incorrect Answers

B:

Selecting labels to favor the candidate biases the evaluation. The reference should represent the agreed task standard independently of model performance.

C:

Category names are not necessarily ordered quantities. A numeric average may not correspond to any valid adjudicated outcome.

D:

Removing hard cases can hide an important requirement gap. Disputes should be investigated and their disposition documented transparently.

Question 220

An evaluation's expected answers have no source records or revision dates. Several policies have changed. What is needed before trusting failures against that set?

  1. Replace every expected answer with the candidate's response
  2. Verify references against authoritative versions and record provenance
  3. Assume the oldest answer key is always authoritative
  4. Ignore all failures involving policy questions

Explanation

Correct Answer

B:

A model can disagree with an outdated answer while following current policy. Evaluation truth needs its own maintained evidence and version history.

Incorrect Answers

A:

That makes the model its own answer key. It removes the independent standard the evaluation is supposed to apply.

C:

Age does not establish current correctness. Reference material must match the intended task version and applicable policy period.

D:

Excluding the relevant domain avoids the verification problem by abandoning coverage. The reference labels should be repaired instead.

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