CCA-F: Claude Certified Architect — Foundations

By Anthropic · Question Mode

QuestionQ114

Context Management & Reliability

You are developing a structured data-extraction system with 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.

Following deployment, you discover that 12% of extractions have semantic errors that still pass JSON schema validation (for example, a duration such as "30 minutes" is incorrectly entered in an ingredient-quantity field). Human reviewers can examine only 20% of extractions.

Which approach most effectively directs reviewer attention?

Explanation

JSON schema validation can verify an output's structure and permitted data types, but it does not establish that a syntactically valid value is semantically appropriate for its field. Calibrating model-generated field-level confidence scores against a labeled validation set makes those scores meaningful for risk ranking, so review thresholds can prioritize the extractions most likely to contain semantic errors while respecting the 20% review capacity. Anthropic's evaluation guidance supports using test cases and quality assessment to evaluate outputs and improve prompts.

Learn more

Community Discussion

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