preference-guided calibration

A method that converts human pairwise comparisons or rating data into evaluation criteria. It adjusts the judging framework without changing model parameters, offering low cost and strong interpretability.