Extended Data Fig. 3: Trajectory error encoding is present across trial types and is not explained by encoding of other task variables.

a, Violin plots of the percent change in the akaike information criterion (AIC) when each variable is removed from the model (removed variable indicated above each plot). Negative values indicate worse model fit when the variable is removed (see Methods for variable definitions). Reward history: t₆₃₀ = 4.53, p = 2.85 × 10⁻⁵; upcoming reward: t₆₃₀ = 6.64, p = 2.71 × 10⁻¹⁰; cue identity: t₆₃₀ = 3.80, p = 6.36 × 10⁻⁴; locomotion: t₆₃₀ = 5.42, p = 3.45 × 10⁻⁷, two-sided t-test on linear mixed effects model for main effect of model type on AIC, p-values bonferroni corrected for 4 comparisons (n = 316 sites, 10 mice). Each data point represents one site. b, TE coefficients from models with (grey) and without (coloured) each variable, averaged across sites and mice using a linear mixed effects model (df = 315, n = 316 sites, 10 mice). c, Violin plots of the peak TE coefficient magnitudes across recording sites in models with (grey) and without (coloured) each variable. TE coefficients were unaffected by exclusion of each variable (except cue identity). Reward history: t₆₃₀ = 0.74; p = 1.0; upcoming reward: t₆₃₀ = −0.65; p = 1.0; cue identity: t₆₃₀ = −2.69; p = 0.30; locomotion: t₆₃₀ = −1.02; p = 1.0; two-sided t-test on main effect of model type on TE coefficient magnitude in a linear mixed effects model (n = 316 sites, 10 mice), bonferroni corrected for 4 comparisons. d, Percent change in AIC comparing the full model to a model with the TE term replaced with a binary congruence only term (no scaling with angular speed), presented as in a. t₆₃₀ = 4.52, p = 7.47 × 10⁻⁶, n = 316 sites, 10 mice, two-sided t-test on effect of model type in a linear mixed effects model. e, Model coefficient t-statistics for TE (orange) and congruence only (grey) averaged across recording sides and mice using a linear mixed effects model (df = 315, n = 316 sites, 10 mice). f, Violin plot showing the maximum coefficient t-statistics for TE and congruence coefficients in the full model and congruence only model respectively across recording sites and mice. t₆₃₀ = −4.24, p = 2.59 × 10⁻⁵, n = 316 sites, 10 mice, two-sided t-test on effect of model type on coefficient t-stat in a linear mixed effects model. g, TE coefficients averaged across sites and mice using a linear mixed effects model for trials split by cue identity (left, df = 315, n = 316 sites, 10 mice), initial running direction (middle, df = 233, n = 234 sites, 7 mice, only mice with >30 trials in each direction were included), and congruence (right, df = 315, n = 316 sites, 10 mice) independently. h, Violin plots comparing TE coefficients computed on trials split by cue identity (left), initial rotational velocity direction (middle), and congruence (right). Cue identity: t₆₃₀ = −1.23, p = 0.66, n = 316 sites, 10 mice; rotational velocity direction: t₄₆₆ = −0.65, p = 1.00, n = 234 sites, 7 mice; congruence: t₆₃₀ = −0.24, p = 1.00, n = 316 sites, 10 mice, two-sided t-test on effect of trial type on maximum TE coefficient in a linear mixed effects model, followed by a Bonferroni correction for three comparisons. Shaded regions and error bars in all plots are 95% confidence intervals. Thin lines in violin plots and cue aligned averages represent averages for individual mice. For box plots in a,c,d,f,h, the centre point is the median, the lower and upper bounds are the first and third quartiles, and the whiskers extend from the box bounds to the most extreme value no further than 1.5 x interquartile range from the bounds.