Extended Data Fig. 4: Stimulus neurons recapitulate first-picture activity during second-picture presentations in all question contexts, indicating content-based rather than feature-specific encoding.

Session-wise decoding accuracies of first stimuli during second-picture presentations in all five question contexts using SVM decoders trained with stimulus neuron activity ~ 600 ms after first-picture onset (400–800 ms bin). Solid lines and shaded areas depict means and standard errors, respectively. Data are compared to chance (25%, dashed horizontal black line). Solid lines in red below indicate significant differences (cluster permutation test, two-sided, p < 0.05). During second-picture presentations, first-picture decoding accuracies exceeded chance beginning ~ 600–1000 ms after second-picture onsets in all question contexts, arguing against a mere recollection of stimulus features.