In Figure 3G we evaluated the time course for two orthogonal representations of second-step state which occurred pre- and post- trial outcome. We defined unit projection vectors from the regression weights for second-step state at a time point mid-way between choice and outcome and 250ms after outcome. We then projected the regression weights for second-step state at each time point onto these two vectors to obtain time-courses for each representation. To avoid selection bias distorting the time-courses, we divided the data into odd and even trials and used the odd trials to define projection vectors that weights from the even trials were projected onto, and vice versa.
