Is it possible that everything that is difficult to understand could actually be converted to single inferential steps? If so, this is related to the idea of using Using spaced repetition for learning.