6.3 LLM-as-JudgeSelf-evaluation, cross-model verification, and scoring rubrics for LLM output quality.