Fig. 3: Shattering analysis for neural goal and uncertainty embeddings.

a Hypothetical neural embeddings for goal and uncertainty and corresponding linear separabilities. While there are three specific goals (red, blue, yellow), only two are depicted for simplicity. The full version of the class labeling and classification types for all dichotomies are in Supplementary Fig. 6. b SD for the different types of dichotomies. We trained separate SVM classifiers for each event (S1-fix') and averaged their decoding accuracies to obtain a single SD value. Since SD represents average binary classification accuracy, the chance level is 0.5. Statistically significant SDs were represented as color bars (Supplementary Fig. 7). Data are presented as mean ± SEM from n = 20 participants. Asterisks denote statistical significance (paired t-test, *: p < 0.05, **: p < 0.01, ***: p < 0.001, ****: p < 0.0001). All statistical tests were two-sided. c Correlations between the neural goal SD and the behavioral measures. Each point represents an individual participant. Solid lines present linear regression slope and dotted lines show 95% confidence bounds of the fitted line where there are statistically significant correlations. (d) Correlation coefficients between the neural SD and the behavioral measures. Only statistically significant correlations are represented with filled bars (Pearson’s correlation, *: p < 0.05, **: p < 0.01, ***: p < 0.001; two-sided test). For all the exploratory correlation analyses, multiple comparison corrections were applied with a false-discovery rate (Benjamini-Hochberg procedure) for the number of ROIs with q = 0.05. See Supplementary Table 3 for full statistical information. Source data are provided as a Source Data file.