LLM-as-Judge

Using a model to score another model's output. Useful for volume; unsafe as the only adjudicator on irreversible actions.