McAfee Secure

Exam Code: CCA-F

Exam Name: Claude Certified Architect - Foundations

Certification Provider: Anthropic

Questions & Answers

Anthropic CCA-F Questions & Answers

Study with Up-To-Date REAL Exam Questions and Answers from the ACTUAL Test

Questions & Answers

134 Questions & Answers with Testing Engine

"Claude Certified Architect - Foundations Exam", also known as CCA-F exam, is a Anthropic certification exam.

Pass your tests with the always up-to-date CCA-F Exam Engine. Your CCA-F training materials keep you at the head of the pack!

Satisfaction Guaranteed

Money Back Guarantee

Testking provides hassle-free money back guarantee with our products. That is because we have 100% trust in the abilities of our professional and experience product team, and our record is a proof of that.

99.6% PASS RATE
Was: $137.49
Now: $124.99

Anthropic CCA-F Practice Test 31

Question 300

A team enables Claude's strict JSON output mode and document citation blocks simultaneously and receives an API error. How should the architect adapt the design?

  1. Concatenate fake citation strings into arbitrary JSON fields without checking their sources.
  2. Keep sending the incompatible combination and assume a larger token budget will fix it.
  3. Disable all citations and still claim the final report has verified page support.
  4. Separate citation-bearing document analysis from the strict JSON output phase, preserving verified source IDs in application data and respecting the current compatibility rules.

Explanation

Correct Answer

D:

Strict JSON and interleaved native citation blocks have incompatible output constraints in the documented API; a staged approach preserves both goals.

Incorrect Answers

A:

Syntactic citation-looking text does not establish that the claims were grounded.

B:

A feature-compatibility error is not a token-length problem.

C:

Dropping evidence while claiming citations creates false provenance.

Question 301

A platform exposes `extract_customer`, `get_order`, and `create_refund`, but their input schemas all reuse a generic object with an untyped `payload` field. The model routinely routes calls incorrectly. What contract change is most useful?

  1. Make output fields optional across all tools to prevent parse failures.
  2. Rename every tool `perform_action` so the naming scheme is uniform.
  3. Define distinct purpose-specific tool names and typed parameters, with read-versus-write permissions and result schemas matching each operation.
  4. Expose all business operations through one unrestricted JSON string and rely on prompt prose for validation.

Explanation

Correct Answer

C:

Discriminating typed contracts reduce ambiguity and make authorization enforceable.

Incorrect Answers

A:

Removing meaningful contracts can conceal missing inputs and operational errors.

B:

Identical generic names increase selection ambiguity.

D:

A free-form field hides required parameters and weakens backend guardrails.

Question 302

An expense extractor reads a credit note showing '-125.00' but emits `amount:125.00` with a valid JSON schema. Which validation catches the real error?

  1. Accept the amount because the JSON parses as a valid number.
  2. Compare the sign and transaction type to the source evidence, rejecting a positive charge where the document records a credit.
  3. Exclude credit notes from reconciliation because negative values complicate matching.
  4. Always multiply every extracted value by negative one.

Explanation

Correct Answer

B:

Schema compliance cannot detect a semantically reversed financial value without source-based checks.

Incorrect Answers

A:

A numeric type does not preserve the direction of a financial adjustment.

C:

Credits are relevant financial evidence, not invalid records. The pipeline must preserve signs and correct rather than remove contradictory amounts.

D:

A global sign flip would corrupt ordinary debit invoices.

Question 303

A structured intake service accepts a date string '2026-02-30' because it matches YYYY-MM-DD. Which extra check should the pipeline apply?

  1. Discard the whole batch of unrelated valid records.
  2. Use deterministic calendar-date validation and return a precise field-level error for the impossible day.
  3. Trust any string that matches the regex as a real calendar date.
  4. Correct the impossible date to the last valid date of that month without returning a validation error.

Explanation

Correct Answer

B:

A string pattern validates formatting only; February does not contain that date.

Incorrect Answers

A:

One invalid field should be handled as a targeted validation failure.

C:

Syntax cannot establish calendar validity.

D:

A silent correction guesses what the submitter intended. Calendar validity needs checking, and the invalid source date should be rejected or clarified.

Question 304

An invoice states a total of 1,190 but the extracted line items add to 1,180. What should a structured extraction system output for reconciliation?

  1. Preserve stated_total and calculated_total separately and set a discrepancy flag with source evidence.
  2. Replace the stated total with the sum and label it as printed on the invoice.
  3. Keep only 1,190 and suppress the arithmetic disagreement.
  4. Assume a missing 10 unit tax without source evidence.

Explanation

Correct Answer

A:

Keeping both values reveals the inconsistency instead of silently falsifying either the source or the arithmetic.

Incorrect Answers

B:

A computed figure must not be misrepresented as the document's value.

C:

The mismatch is relevant for accounting review and should be disclosed.

D:

Inventing a tax line masks the discrepancy rather than resolving it.

Question 305

An OCR invoice contains an unreadable customer reference but clearly shows the supplier, invoice date and totals. The schema allows a nullable reference. Which action is appropriate?

  1. Treat the blank field as the literal string 'verified'.
  2. Declare every field invalid because one character is unreadable.
  3. Generate a realistic customer reference from the supplier's name.
  4. Return the reliably extracted fields, leave customer_reference null and flag that field's uncertainty for focused review.

Explanation

Correct Answer

D:

Independent verified fields remain useful, while the ambiguous one is not fabricated.

Incorrect Answers

A:

A placeholder is not proof that a source value exists.

B:

Localized uncertainty does not justify discarding all useful source evidence.

C:

A guessed identifier can link the invoice to the wrong account.

Question 306

A JSON extraction request fails because `line_items[2].quantity` is a string rather than an integer. The model retries with the same malformed field. What feedback is most effective?

  1. Remove the line_items array from the required schema.
  2. Change every quantity to 1 to make the schema pass.
  3. Send only 'invalid output' without a location or expected type.
  4. Return the exact JSON path, expected integer type, observed invalid value and correction requirement for that specific item.

Explanation

Correct Answer

D:

Precise machine-readable feedback gives the next attempt an actionable target.

Incorrect Answers

A:

Deleting needed information avoids rather than fixes the constraint violation.

B:

An arbitrary default invents transaction data.

C:

Generic errors often cause identical malformed retries.

Question 307

A claims-extraction pipeline produces a correct amount and entity name but loses the originating document ID when records enter a warehouse. What should the ingestion validation require?

  1. Use the worker's nickname as the only source reference.
  2. Preserve document ID, page or span reference and stable record ID with each extracted claim before ingestion.
  3. Accept a blank document ID when the model says its confidence is high.
  4. Assign a new ingestion ID and label it the originating document ID without retaining source linkage.

Explanation

Correct Answer

B:

Source-linked fields allow auditing and later correction of disputed claims.

Incorrect Answers

A:

Agent labels cannot locate evidence in original documents.

C:

Model confidence does not replace provenance.

D:

An internal row ID can track a record but cannot stand in for missing source provenance. The original document reference must be recovered or flagged unresolved.

Question 308

A classification schema accepts only `approved`, `pending` or `denied`, but the model returns `awaiting_review`. What should feedback to the next attempt include?

  1. Silently convert all unknown values to `approved`.
  2. Identify the exact enum field and its three allowed values, and require a mapped status only if justified by the case evidence.
  3. Allow any spelling and claim the strict schema passed.
  4. Delete the status field and treat omission as a valid classification.

Explanation

Correct Answer

B:

Specific constraints guide correction without fabricating a decision.

Incorrect Answers

A:

A permissive default can authorize actions without evidence.

C:

Unbounded statuses break downstream processing contracts.

D:

Removing a required decision status hides unresolved work.

Question 309

A scanned application has two overlapping account numbers and neither can be matched reliably to the account label. What should the extraction assistant do?

  1. Mark the affected identifier ambiguous and route its source span for human review while retaining other supported fields.
  2. Combine half of each number into a new candidate.
  3. Choose whichever number appears first in OCR text.
  4. Submit both numbers as one verified account identifier.

Explanation

Correct Answer

A:

A conflict in high-impact identity data should remain unresolved rather than be guessed.

Incorrect Answers

B:

A synthetic hybrid is not present in the document.

C:

Reading order is not proof of the correct account label.

D:

Concatenation conceals ambiguity and can cause misrouting.

Question 310

A structured-output pipeline gives a model three correction attempts for a field, but all three violate the same invariant. What should happen when the bound is reached?

  1. Stop the loop, record the unresolved field and error history, and route the item to defined fallback or review handling.
  2. Suppress the validation result and mark the item complete.
  3. Replace the entire document with a generated document that fits the schema.
  4. Continue indefinitely until the model eventually guesses a valid value.

Explanation

Correct Answer

A:

A bounded failure outcome prevents runaway cost while preserving evidence of the unsolved defect.

Incorrect Answers

B:

That would silently admit invalid business data.

C:

The deliverable must reflect the actual source, not invented evidence.

D:

Unbounded retries can consume resources without improving accuracy.

Claude Certified Architect - Foundations (CCA-F) Practice Tests