This is a rather detailed theory of how the cerebellum works. It closely tracks with work in deep reinforcement learning on the topic of Guided Policy Search, but fills in many details about a hypothesized neural implementation that are, so far as we can tell, wholly novel.