Geometric Bayes

Bayes' theorem as an argument about areas. The square is the whole probability space, split horizontally by the prior $p(H)$ against $p(\neg H)$ and vertically by the likelihoods, so each of the four rectangles is a joint probability. Conditioning on the evidence $E$ keeps only the two rectangles where $E$ occurred; rescaling those to full height renormalizes them into the posterior $p(H|E)$ shown on the right. Following 3blue1brown.