Exam CCAR-F Topic 1 Question 107 Discussion
Actual exam question for Anthropic's CCAR-F exam
Question #: 107
Topic #: 1
Question #: 107
Topic #: 1
You are building a structured data extraction system using Claude. The system extracts information from unstructured documents, validates the output using JavaScript Object Notation (JSON) schemas, and maintains high accuracy. It must handle edge cases gracefully and integrate with downstream systems.
Testing reveals that when source documents are missing certain specifications, the model fabricates plausible- sounding values to satisfy your schema's required fields. For example, a document mentioning only dimensions receives a fabricated "weight: 2.3 kg" in the extraction output.
What schema design change most effectively addresses this hallucination behavior?
Testing reveals that when source documents are missing certain specifications, the model fabricates plausible- sounding values to satisfy your schema's required fields. For example, a document mentioning only dimensions receives a fabricated "weight: 2.3 kg" in the extraction output.
What schema design change most effectively addresses this hallucination behavior?
Suggested Answer: B Vote an answer
The schema is creating a structural incentive for fabrication. When a field is declared required, the output must contain a value even when the source document contains no corresponding evidence. Structured Outputs can guarantee that Claude's response conforms to a JSON Schema, but schema conformance does not establish that every generated value is factually supported. Anthropic's documentation shows that the required array determines which properties must be present; therefore, source-dependent properties that may legitimately be absent should not be included as required fields. ( https://platform.claude.com/docs/en/build- with-claude/structured-outputs ) Option B corrects the problem at the contract level. Claude can omit the unavailable property rather than inventing content merely to produce valid JSON. A nullable representation could also be used when downstream systems require a stable key set, but forcing an unsupported non-null value is architecturally unsound.
Option A still requires placeholder generation and does not resolve the mismatch between the schema and available evidence. Option C relies on model-generated confidence, which is not a substitute for grounding.
Option D is a useful secondary control, but it does not constitute the requested schema-design change.
Anthropic recommends allowing uncertainty and requiring factual claims to be grounded in the provided material. ( https://docs.anthropic.com/en/docs/test-and-evaluate/strengthen-guardrails/reduce-hallucinations ) Official references/topics: Structured Outputs-JSON Schema design; Reduce Hallucinations-allowing uncertainty and grounding claims.
Option A still requires placeholder generation and does not resolve the mismatch between the schema and available evidence. Option C relies on model-generated confidence, which is not a substitute for grounding.
Option D is a useful secondary control, but it does not constitute the requested schema-design change.
Anthropic recommends allowing uncertainty and requiring factual claims to be grounded in the provided material. ( https://docs.anthropic.com/en/docs/test-and-evaluate/strengthen-guardrails/reduce-hallucinations ) Official references/topics: Structured Outputs-JSON Schema design; Reduce Hallucinations-allowing uncertainty and grounding claims.
by Gene at Oct 29, 2026, 10:18 AM
0
0
0
10
Comments
Upvoting a comment with a selected answer will also increase the vote count towards that answer by one. So if you see a comment that you already agree with, you can upvote it instead of posting a new comment.
Report Comment
Commenting
You can sign-up / login (it's free).