Fig. 7: Population-level encoding of task variables exhibits differential stability in RSC.

a Encoding of context (top row), motor choice (top-middle row), post-decision outcome (bottom-middle row) and post-trial outcome (bottom row) by pseudo-populations of RSC neurons pooled across all mice (the size of the corresponding subpopulations of neurons is indicated in Fig. 4c). Solid line, mean; shaded area, 95% CIs. Note the gradient in population encoding, where post-trial outcome and context are more stable than post-decision outcome and motor choice. b Heat maps of context (top), motor choice (top-middle), post-decision outcome (bottom-middle), and post-trial outcome (bottom) coding performance along trial duration and across all 5 days of experimentation. Note the stability of context encoding between trial start and trial end. Also, note the stability of post-trial outcome encoding after trial end. By contrast, the encoding of motor choice and post-decision outcome around the decision point decays more rapidly after day 2. c Normalized decoding performance integrated over the indicated 1.5 s windows relevant for each task variable (mean ± bootstrapped s.e.m.; n = 100 iterations). Exponential decay functions were fit to estimate the decay in encoding stability across days (black lines denote fits to the average normalized performance). Note the faster decay in motor choice and post-decision outcome encoding, and the slower decay in context and post-trial outcome encoding. To obtain CIs for the decay constants, we also fit decay functions to each iteration of the decoder performance (95% CIs, context = 5.9–9.7 days, motor = 1.4–2.5 days, post-decision outcome = 1.9–2.5 days, post-trial outcome = 25.9–50.0 days). For all panels: s trial start, d decision point, e trial end. Source data for C are provided as a Source Data file.