Extended Data Fig. 1: Model fit to a mouse with a larger number of sessions.

Using the plotting conventions of Fig. 2, this depicts the states and corresponding psychometric functions identified in a mouse that required 36 sessions to learn. This illustrates the counterintuitive phenomenon that long training trajectories were sometimes fitted by the model using a small number of states. Some of these states, particularly state 1 in this example, underwent substantial changes through the slow process, spanning the range of uninformed to proficient behavior (note that the type labels to the right of the PMFs are determined by the highest type reached by a state; thus, state 1 is labeled as type 3 even though it began as type 1).