Comment by arjunnarayan

This is an important corollary of Paul's essay that I wish he mentioned. Sometimes when you update your model of the world significantly, you still have a cache of previously computed facts that were computed using the old model. You have to clear out that cache and recompute with the new model, and that takes a significant amount of time. Often, rereading is a core part of that, since it forces you to revisit a lot of source material and recompute.