A measure of how hard a problem is for a solver to answer correctly, used to generate progressively challenging training examples.