This skill should be used when the user asks to implement LLM-as-judge, compare model outputs, create evaluation rubrics, mitigate evaluation bias, or mentions direct scoring, pairwise comparison, position bias, evaluation pipelines, or automated quality assessment.