Using expert human judgment to train or validate an AI evaluator so it aligns with professional standards.