Anthropic CCDV-F (Claude Certified Developer - Foundations) Practice Test 32

Question 310

A prompt works on ordinary billing questions but fails on deliberate requests to waive approval rules. Which new evaluation set best tests that weakness?

  1. Increase temperature to see whether risky requests sometimes get more creative replies.
  2. Repeat the successful billing examples and count how often wording stays identical.
  3. Judge only whether the final answer sounds reassuring to a reviewer.
  4. Add adversarial and boundary-case prompts that try to override approval constraints, and score the resulting actions.

Explanation

Correct Answer

D:

A regression suite made only of happy-path examples cannot reveal policy bypass attempts. Include realistic role-spoofing, ambiguous authority and high-impact action cases, and assert both responses and tool-side effects.

Incorrect Answers

A:

Changing sampling may vary responses but does not establish that the policy is enforced. Security assurance requires representative hostile cases and explicit pass/fail criteria.

B:

Stylistic agreement on easy inputs provides little evidence about authorization robustness. The failure occurs on adversarial intent, so the test set must exercise that boundary.

C:

A fluent refusal or approval statement can diverge from actual tool activity. Evaluations must inspect authoritative execution behavior for high-impact operations.

Question 311

A data extractor often omits optional fields rather than emitting nulls. What clarification is useful?

  1. At the application layer, delimit quoted email bodies as source content under the actual user instruction.
  2. Employ diverse examples that cover distinct edge conditions and failure modes.
  3. Send the intended reader, purpose, evidence standard, and constraints.
  4. Specify how absent values must be represented in the required schema.

Explanation

Correct Answer

D:

Distinguishing missing data from an omitted key supports consistent downstream parsing and auditability. Define whether unknown values are represented by null, an explicit status, or an omitted key, then test that contract.

Incorrect Answers

A:

Clear containment of quoted text prevents the model from assuming the email sender can issue instructions to the assistant; the missing-field representation criterion differs from nested instruction boundaries; the gap persists.

B:

Variety communicates generalizable rules; repeated trivial cases add cost without clarifying boundaries; this may help multi-criterion integration but cannot by itself deliver missing-field representation.

C:

Context about the report's use guides which details matter and avoids generic content that misses the business goal; for this request, missing-field representation is the missing piece, not task audience specification.

Question 312

An application needs Claude to distinguish quoted emails from new user requests. What is the strongest prompt assembly choice?

  1. Delimit quoted email bodies as source content under the actual user instruction.
  2. For this workflow, explicitly request only the schema-defined fields and pair this with runtime enforcement.
  3. Review each revision against a fixed regression suite and documented failures.
  4. Classify the page as untrusted evidence, with instructions provided separately.

Explanation

Correct Answer

A:

Clear containment of quoted text prevents the model from assuming the email sender can issue instructions to the assistant.

Incorrect Answers

B:

Format instructions improve compliance, but production parsers still need structured-output controls or validation; when the goal is nested instruction boundaries, optimizing output-only prompt guidance misses that operational boundary.

C:

Prompt improvements should be measured on representative examples rather than judged by a single demonstration; changing iterative prompt regression testing does not establish nested instruction boundaries in this application.

D:

Clear data boundaries reduce instruction confusion; actual tool authorization still requires enforcement outside the prompt; testing nested instruction boundaries would still fail after improving untrusted content labeling.

Question 313

A prompt writer assumes saying 'think carefully' guarantees correctness on arithmetic. What should replace that assurance?

  1. Supply explicit evaluation criteria and a comparison output structure.
  2. Test each revision against a fixed regression suite and documented failures.
  3. For this workflow, render and validate template variables before making the API call.
  4. Delegate exact arithmetic to code and verify the result independently.

Explanation

Correct Answer

D:

Prompt wording can affect reasoning but cannot replace deterministic computation or validation for precise totals.

Incorrect Answers

A:

A criterion-by-criterion assessment reduces arbitrary preference and makes the recommendation auditable; when the goal is correct arithmetic enforcement, optimizing rubric-guided comparison misses that operational boundary.

B:

Prompt improvements should be measured on representative examples rather than judged by a single demonstration; this implements iterative prompt regression testing rather than the demanded correct arithmetic enforcement.

C:

Unresolved placeholders or broken interpolation are input-construction defects, not model reasoning failures; even if prompt variable interpolation succeeds, correct arithmetic enforcement still needs separate verification.

Question 314

A support bot must respond differently when no applicable policy is retrieved. What should its prompt and application specify?

  1. Give the intended reader, purpose, evidence standard, and constraints.
  2. At the application layer, render and validate template variables before making the API call.
  3. Include a clear no-evidence abstention path and test that branch.
  4. Develop a parameterized, versioned prompt template.

Explanation

Correct Answer

C:

Without an explicit failure outcome, the model may fill gaps with plausible text; the client must support an unanswered status.

Incorrect Answers

A:

Context about the report's use guides which details matter and avoids generic content that misses the business goal; because task audience specification is its concern, explicit insufficient evidence path remains outside its effect here.

B:

Unresolved placeholders or broken interpolation are input-construction defects, not model reasoning failures; where explicit insufficient evidence path matters, a control for prompt variable interpolation leaves that test unmet.

D:

Template variables isolate runtime inputs from shared instructions, making edits testable and consistent across use cases; testing explicit insufficient evidence path would still fail after improving reusable prompt template.

Question 315

A Claude response is syntactically valid JSON but the customer_id does not exist. Which validation is still required?

  1. Verify identifiers against an authoritative service before applying the result.
  2. Reduce the schema to the supported constraint subset and retest.
  3. Regard the output as incomplete and retry or continue with sufficient budget.
  4. Identify the refusal state before applying the normal schema consumer.

Explanation

Correct Answer

A:

Structure guarantees do not establish truth or referential integrity; lookup validation prevents invented IDs from driving actions.

Incorrect Answers

B:

Constrained generation has compilation and feature limits; retrying the identical invalid schema will not resolve it; it improves schema compilation limits only, leaving semantic identifier validation unproven.

C:

A generation limit can cut a structured document off before closure; a schema request is not proof of completion; for semantic identifier validation, a control centered on truncated structured response is not the appropriate substitute.

D:

Refusal is a valid model outcome and may override structured output shape; it needs a separate application branch; the semantic identifier validation criterion differs from refusal-aware parsing; the gap persists.

Question 316

An extraction pipeline needs reliably parseable JSON objects rather than prose surrounded by code fences. Which capability is preferable?

  1. Process all applicable content blocks according to their types and order.
  2. Confirm identifiers against an authoritative service before applying the result.
  3. Use supported JSON-schema structured outputs for constrained generation.
  4. Associate the output to an allowed enum and enforce rule-based notification eligibility.

Explanation

Correct Answer

C:

Grammar-constrained generation enforces response shape more reliably than request wording alone, subject to completion and refusal exceptions.

Incorrect Answers

A:

Responses can contain multiple segments; a single-block assumption silently drops information or ignores tool requests; this targets complete content block processing, while the system must demonstrate schema constrained JSON generation.

B:

Structure guarantees do not establish truth or referential integrity; lookup validation prevents invented IDs from driving actions; because semantic identifier validation is its concern, schema constrained JSON generation remains outside its effect here.

D:

Structured classification reduces parser ambiguity, but business authorization must still gate consequential actions; it improves validated output to workflow boundary only, leaving schema constrained JSON generation unproven.

Question 317

A service obtains a JSON response matching a declared schema, but its parser silently ignores unexpected extra fields before forwarding data to a payment service. What contract rule should it add?

  1. Disable all parsing errors and rely on the payment provider to guess missing meanings.
  2. Append the unknown properties to the system prompt for the next call.
  3. Reject or explicitly handle unknown properties at the trusted parser boundary before downstream use.
  4. Treat any parseable JSON object as authorized because Claude produced it.

Explanation

Correct Answer

C:

Schema-shaped output is not sufficient when consumers silently accept fields they do not understand. An explicit additional-properties policy and validated field mapping prevent accidental interpretation or propagation of unreviewed values.

Incorrect Answers

A:

Lenient error handling hides upstream incompatibilities. Payments require an unambiguous accepted schema, not implicit interpretation by an unrelated downstream component.

B:

Copying unexpected fields into higher-priority context is not structured validation. It risks turning arbitrary data into influential instructions without repairing the consumer contract.

D:

Syntactic validity and model origin do not establish a safe payment contract. The service must validate both allowed keys and business rules before creating side effects.

Question 318

An output schema contains unsupported or overly complex JSON Schema constructs, causing a 400 error. What is the remedy?

  1. Simplify the schema to the supported constraint subset and retest.
  2. Employ supported JSON-schema structured outputs for constrained generation.
  3. As an alternative, generate validated structured rows and serialize CSV with a standard library.
  4. Handle all applicable content blocks according to their types and order.

Explanation

Correct Answer

A:

Constrained generation has compilation and feature limits; retrying the identical invalid schema will not resolve it.

Incorrect Answers

B:

Grammar-constrained generation enforces response shape more reliably than request wording alone, subject to completion and refusal exceptions; the effect concerns schema constrained JSON generation; the success condition is schema compilation limits.

C:

A dedicated serializer correctly handles quoting and delimiters, avoiding brittle manual text formatting; the consequence is improved safe CSV serialization, not verified schema compilation limits.

D:

Responses can contain multiple segments; a single-block assumption silently drops information or ignores tool requests; when the goal is schema compilation limits, optimizing complete content block processing misses that operational boundary.

Question 319

A structured response stops at max_tokens and fails JSON parsing. What should the caller infer?

  1. Treat the output as incomplete and retry or continue with sufficient budget.
  2. Verify domain-specific numerical constraints before accepting or committing the amount.
  3. Define precise tool input schemas and validate arguments at runtime.
  4. Adopt supported JSON-schema structured outputs for constrained generation.

Explanation

Correct Answer

A:

A generation limit can cut a structured document off before closure; a schema request is not proof of completion.

Incorrect Answers

B:

Type validation cannot establish the permitted business range; deterministic invariants are required; it follows numeric business constraints constraints but leaves truncated structured response unspecified.

C:

Structured tool inputs reduce ambiguity, while executor validation protects external systems from unsupported values; for this request, truncated structured response is the missing piece, not tool argument schema checks.

D:

Grammar-constrained generation enforces response shape more reliably than request wording alone, subject to completion and refusal exceptions; because schema constrained JSON generation is its concern, truncated structured response remains outside its effect here.

Question 320

A safety refusal returns a successful HTTP response but not the requested JSON object. What must the parser support?

  1. Verify the received tool response contract and handle incompatible types explicitly.
  2. Check identifiers against an authoritative service before applying the result.
  3. Trim the schema to the supported constraint subset and retest.
  4. Detect the refusal state before applying the normal schema consumer.

Explanation

Correct Answer

D:

Refusal is a valid model outcome and may override structured output shape; it needs a separate application branch.

Incorrect Answers

A:

Tool outputs originate outside the model and can drift independently; blindly casting values risks silent failures; testing refusal-aware parsing would still fail after improving external result type validation.

B:

Structure guarantees do not establish truth or referential integrity; lookup validation prevents invented IDs from driving actions; the refusal-aware parsing criterion differs from semantic identifier validation; the gap persists.

C:

Constrained generation has compilation and feature limits; retrying the identical invalid schema will not resolve it; when the goal is refusal-aware parsing, optimizing schema compilation limits misses that operational boundary.

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