A random 50% of trials had an uninterrupted delay, with no change in the remembered sample location relative to gaze (‘‘no-swap’’ trials; Figure 1A, left). On the other trials, halfway through the delay the fixation point jumped across the midline to the opposite side, instructing an immediate saccade and refixation on it for the remainder of the delay. This shifted the remembered sample’s retinotopic location to the opposite visual hemifield (‘‘swap’’ trials; Figure 1A, right). Performance was good for all conditions (p ≤ 1 × 10−4 for all, randomized sign test across 56 sessions), albeit somewhat worse on swap trials (p ≤ 1 × 10−4, permutation paired t test; Figure 1C). There were no significant differences in performance when the sample (p = 0.71) or test object (p = 0.10) appeared in the left versus right hemifield, so all results were pooled across them. We recorded multi-unit activity (MUA) and local field potentials (LFPs) from 256 electrodes in four chronic arrays implanted bilaterally in both hemispheres of lateral PFC during task performance (Figure 1D). All presented statistics were based on
