Design structured rubrics that make an LLM classify or grade consistently and return strict JSON. Use when building an LLM-as-judge or classifier, when a model's category outputs are inconsistent or ambiguous, or when you need reliable machine-parseable results from a prompt.
