Free CCAR-F Exam Braindumps
Claude Certified Architect - Foundations Questions 31–35 (page: 8)

Updated On: 17-Nov-2026

The system processes product reviews using tool use with a defined schema: rating (integer 1-5), pros (string array), cons (string array), and overall_sentiment (enum: positive, ne Testing reveals two issues with brief or ambiguous reviews (-20% of the dataset): (1) for reviews like "Great product!", Claude fabricates specific pros and cons rather than Indica Information isn't explicitly stated, and (2) for sarcastic reviews like "Well that was.. interesting", Claude picks sentiment arbitrarily since there's no option for ambiguous cases. W modification best addresses both issues?

  1. Make pros and cons optional fields, and add "neutral" and "unclear" to the sentiment enum
  2. Allow empty arrays for pros/cons as valid output, and add "unclear" ss the sentiment enum
  3. Add an extraction_confidence field (0.0-1.0) for each value, and filter outputs where any confidence falls below a threshold.
  4. Allow null values for pros/cons, and add "unclear" to the sentiment earum.

Answer(s): B

Explanation:

A: Make pros and cons optional fields, and add "neutral" and "unclear" to the sentiment enum Incorrect.
Making fields optional can lead to inconsistent outputs, and “neutral” doesn’t solve ambiguity—it’s different from “unclear”.
B: Allow empty arrays for pros/cons as valid output, and add "unclear" as the sentiment enum Correct.
This prevents fabrication by allowing explicitly empty outputs when no details are present, and “unclear” handles ambiguous or sarcastic sentiment appropriately.
C: Add an extraction_confidence field (0.0–1.0) for each value, and filter outputs where any confidence falls below a threshold. Incorrect.
This adds complexity but doesn’t prevent fabrication or resolve ambiguity in outputs.
D: Allow null values for pros/cons, and add "unclear" to the sentiment enum. Incorrect.
Nulls are less consistent than empty arrays for structured outputs and can complicate downstream processing.



Your extraction system implements automatic retries when validation fails. On each retry, the specific validation error is appended to the prompt. This retry-with-error-feedback approach resolves most failures within 2-3 attempts. For which failure pattern would additional retries be LEAST effective?

  1. The model extracts "et al." for co-authors when the full list exists only in an external document not in the input
  2. The model extracts citation counts as locale-formatted strings ("1234") when the schema requires integers
  3. The model extracts dates as ISO 8601 datetime strings ("2003-03-15T00:00:00Z") when the schema requires only the date portion (YYYY-MM-DD)
  4. The model extracts keywords as a nested object organized by category when the schema requires a flat array of strings

Answer(s): A

Explanation:

A: The model extracts "et al." for co-authors when the full list exists only in an external document not in the input Correct.
Retries won’t help because the required information is not present in the input context . The model cannot recover missing data through repeated attempts.
B: The model extracts citation counts as locale-formatted strings ("1234") when the schema requires integers Incorrect.
This is a formatting issue that can be corrected through retries with validation feedback.
C: The model extracts dates as ISO 8601 datetime strings ("2003-03-15T00:00:00Z") when the schema requires only the date portion (YYYY-MM-DD) Incorrect.
Also a format mismatch , which retries can fix easily.
D: The model extracts keywords as a nested object organized by category when the schema requires a flat array of strings Incorrect.
This is a structural mismatch that can typically be corrected with retry feedback.



Your invoice extraction uses tool use with strict JSON schemas. JSON syntax errors never occur, but 12% of extractions fail semantic validation--for example, line Item amounts don't extracted total, or vendor IDs don't match valid formats. These failures currently route to manual review.
What's the most effective approach to reduce manual review volume while m accuracy?

  1. Retry the extraction up to 3 times when validation fallis, accepting the first result that passes validation.
  2. Implement post-processing logic that automatically corrects common amors, such as recalculating totais from line items when sums don't match.
  3. When validation falls, make a follow-up request with the document, extraction, and validation errors for model correction.
  4. Add stricter schema constraints with detailed field descriptions to prevent the model from generating invalid values initially.

Answer(s): C

Explanation:

A: Retry the extraction up to 3 times when validation fails, accepting the first result that passes validation. Incorrect.
Retries without targeted feedback often repeat the same mistakes and don’t reliably fix semantic inconsistencies.
B: Implement post-processing logic that automatically corrects common errors, such as recalculating totals from line items when sums don't match. Incorrect.
While useful for specific cases, this is narrow and brittle, and doesn’t address broader validation failures like incorrect IDs.
C: When validation fails, make a follow-up request with the document, extraction, and validation errors for model correction. Correct.
This provides targeted feedback , enabling the model to fix specific issues, significantly reducing manual review while maintaining accuracy.
D: Add stricter schema constraints with detailed field descriptions to prevent the model from generating invalid values initially. Incorrect.
Schema improvements help upfront, but they cannot fully prevent semantic errors like mismatched totals.



Your team is extracting structured data from 50,000 legacy legal contracts under a two-week deadline. Initial testing with 500 sample documents shows 82% pass JSON schema first attempt, while the remaining 18% fall due to diverse issues—missing required fields, malformed dates, and incorrectly identified parties. Documents that fail typically need refinements targeting their specific failure modes before extraction succeeds.
Which batch processing strategy is the most cost-efficient while still meeting the deadline?

  1. Split documents into 10 sequential batches of 5,000 each, analysing results and refining prompts between batches to improve extraction quality progressively.
  2. Submit all 50,000 documents via batch API, then submit failed extractions in successive batches—refining prompts between each batch—until all documents pass validation.
  3. Use the real-time API for all 50,000 documents since the batch API's 24-hour processing window creates unacceptable deadline risk.
  4. Process 2,000 sample documents via real time API to identify failure patterns and refine prompts, then batch process all 50,000 with the optimized prompts.

Answer(s): B

Explanation:

A: Split documents into 10 sequential batches of 5,000 each, analysing results and refining prompts between batches to improve extraction quality progressively. Incorrect.
This introduces unnecessary sequential delays and reduces throughput, risking the deadline.
B: Submit all 50,000 documents via batch API, then submit failed extractions in successive batches— refining prompts between each batch—until all documents pass validation. Correct.
This maximizes throughput and parallelism upfront , ensuring the deadline is met. Then it uses targeted iterative refinement only on failures , making it cost-efficient while handling diverse failure modes effectively.
C: Use the real-time API for all 50,000 documents since the batch API's 24-hour processing window creates unacceptable deadline risk. Incorrect.
This is unnecessarily expensive and not required given batch processing capabilities.
D: Process 2,000 sample documents via real-time API to identify failure patterns and refine prompts, then batch process all 50,000 with the optimized prompts. Incorrect.
While proactive, this assumes failure patterns generalize well, which the scenario suggests they don’t—since failures are diverse and require case-specific refinements .



Your extraction pipeline processes contracts that frequently include amendments.
When a contract contains both original terms and later amendments (e.g., original clause specifies "30-day payment terms" while Amendment 1 changes this to "45 days"), the model inconsistently extracts one value or the other with no indication of which applies.
What's the most effective approach to improve extraction accuracy for documents with amendments?

  1. Preprocess documents with a classifier that identifies and removes superseded sections before the main extraction step.
  2. Implement post-extraction validation using pattern matching to detect amendments and flag those extractions for manual review.
  3. Redesign the schema so amended fields capture multiple values, each with source location and effective date.
  4. Add prompt instructions to always extract the most recent amendment value and ignore superseded original terms.

Answer(s): C

Explanation:

A: Preprocess documents with a classifier that identifies and removes superseded sections before the main extraction step. Incorrect.
This is brittle and risky—accurately identifying and removing superseded clauses is complex and may lead to loss of important context.
B: Implement post-extraction validation using pattern matching to detect amendments and flag those extractions for manual review. Incorrect.
This increases manual review but does not improve extraction accuracy or resolve ambiguity.
C: Redesign the schema so amended fields capture multiple values, each with source location and effective date. Correct.
This preserves both original and amended values with context , enabling accurate interpretation and avoiding ambiguity about which value applies.
D: Add prompt instructions to always extract the most recent amendment value and ignore superseded original terms. Incorrect.
This relies on model judgment, which is inconsistent, and loses traceability of how values changed over time.



Viewing page 8 of 31
Viewing questions 36 - 40 out of 149 questions


Post your Comments and Discuss Anthropic CCAR-F exam prep with other Community members:

AI Tutor AI Tutor 👋 I’m here to help!