CCA-F: Claude Certified Architect — Foundations

By Anthropic · Question Mode

QuestionQ97

Prompt Engineering & Structured Output

You are developing a structured data-extraction system using Claude. The system pulls information from unstructured documents, validates its output with JavaScript Object Notation (JSON) schemas, and preserves high accuracy. It must gracefully handle edge cases and integrate with downstream systems.

Your system extracts event metadata (date, location, organizer, attendee_count) from news articles by using a JSON schema in which every field is nullable. During evaluation, you find that the model often produces plausible but incorrect values for fields that are not mentioned in the article—for example, returning "500" for attendee_count when the source has no attendance information.

What is the most effective approach to reduce these false extractions?

Explanation

An explicit instruction to return null whenever the source does not directly state a field value establishes an abstention rule for missing evidence and uses the nullable schema as intended. This reduces unsupported extractions without forcing a fabricated non-null value. Anthropic recommends restricting Claude to information contained in the provided documents and allowing it to express uncertainty to reduce hallucinations.

Learn more

Community Discussion

No comments yet. Be the first to start the discussion!