Extended Data Fig. 7. Movement-corrected choice decoder.

(a) Using a model based on movement variables, we subtracted all movement-related activity from raw fluorescence data and applied the choice decoder analysis to the resulting residuals. (b) Removing movement-related activity reduced choice prediction accuracy, in particular during the delay and response period when most choice-related movements occur. In all PyN types, predictions remained above chance levels, suggesting that part of the choice-related activity is independent of observable movements. (c) Movement-corrected choice kernels revealed the same cortical patterns as seen in the regular choice decoder (Fig. 6c), demonstrating that choice signals are not solely driven by movements.