In our study, whether the eyes were fixated to the left or right, the sample item was always at the same central spatiotopic (real-world) location. Thus, if PFC encoded locations in a spatiotopic reference frame (Figure S6B, bottom), its activity should be relatively invariant to the location of the sample object relative to the eyes (Figure S6C, bottom row). Arguing against this possibility is the differentiation between contralateral and ipsilateral sample locations in PFC activity (Figure 2). This suggests a retinotopic reference frame (Figure S6C, top row). On the other hand, if WM, at the cognitive level, also maintained locations in a retinotopic reference frame, remembered locations would be anchored to the location of gaze and would simply shift around with a saccade (Figure S6A, top). This makes predictions (Figure S6C, left column) identical to the stable trace model (Figures 2A and 2C) (i.e., that the memory trace remains in the original hemisphere). Instead, we found that neural signatures of laterality invert after the saccade (Figures 4 and 5), ruling out a retinotopic reference frame for WM (Figure S6C, left
