For reference, we replot (from Figure 4B) classification accuracy when both training and cross-validated testing were performed on constant contralateral trials (Figure 6E, desaturated green). As a control, we computed cross-classification accuracy when the same no-swap contralateral trained classifiers were tested on no-swap ipsilateral trials (Figure 6E, desaturated brown). This reflects baseline cross-classification generalization due solely to any bilaterality in the prefrontal neural code. This control analysis resulted in poor accuracy, further evidence for independence between the contralateral and ipsilateral hemifields. For most of the post-saccade delay, cross-classification of the contralateral-shift trials (Figure 6E, orange) was also poor and not significantly different from the control (p ≥ 0.01, paired t test). Here, classifier training and testing were performed at the same time points relative to the end of the delay. Similar results were obtained with training and testing at all possible relative times (Figure S3), indicating results are not dependent on the specific timing scheme used.
