A process where humans and agents iteratively resolve disagreements to align on shared evaluation standards.