a new autoevaluation method that is provably better (at least no worse) than evaluation irrespective of the quality of the autoevaluator