Fig. 3. Computational modeling reveals distinct value computations for wait time and trial initiation.

a Model schematic. b Example wait time model (orange) performance for mixed blocks only in held-out test data (maroon). c Rats’ (left; sample sizes in methods) and model-predicted (right) wait times in mixed blocks as a function of catch probability (lighter gray = lower catch probability). d Example opportunity cost and wait time dynamics from low (blue) or high (red) blocks to mixed block from inferential model. e Inferential model fit (orange) to rat data (maroon) can capture wait time behavior in held-out test data. f Inferential model fit to data predicts that wait times for 20 μL in mixed blocks are not sensitive to previous rewards  (N = 291 fits). g Example opportunity cost and wait time dynamics from low (blue) or high (red) blocks to mixed block from retrospective model. h Retrospective model can qualitatively capture trial initiation time behavior in low (blue) and high (red) blocks. i Retrospective model captures conditional trial initiation time trend across rats (N = 291 fits). j Model comparison using Δ BIC prefers inferential model compared to retrospective model when fit to wait time data (p = 1.07 × 10⁻³⁷, two-tailed Wilcoxon Signed-rank test, N = 291). k Schematic for sub-optimal inference model. l Transitions from mixed to low (blue) or high (red) blocks for wait time (left) or trial initiation time (right) separated by quality of inference for λ < 20th (light colors) or > 80th percentile (dark colors). *p < 0.05, one-tailed non-parametric shuffle test comparing logistic fit parameters, N = 116 (Wait time mixed to low slope p = 0.02, mixed to high right asymptote p = 0.003, Trial initiation time mixed to low slope p = 0.15, mixed to high right asymptote p = 0.21). All error bars are mean ± S.E.M. Source data are provided as a Source Data file.