Fig. 1: Choice- and reward-history exert opposing effects on animals decisions.

A Discrete version of the VI task illustrating trial structure and behavioral epochs. The upper panel shows task-relevant events. The lower panel shows the event timeline; dashed lines mark variable intervals and solid lines fixed intervals. B Top: performance of animal D004 in one session. Black vertical lines indicate block transitions and the change in set reward probabilities. Choices were convolved with a Gaussian filter (length = 10 trials, s.d. = 5). Right choices were coded as 1 and left choices as − 1. Rewards were convolved identically. Bottom: left panel shows experienced reward rates across block transitions; right panel shows filtered choices across the same transitions. C Relationship between reward ratio and choice ratio across sessions. A significant positive association was observed (two-sided linear regression: slope = 0.65, 95% CI = 0.578–0.724, t(df = 127) = 17.69, r = 0.89, P = 2.25 × 10⁻²⁹). No adjustment for multiple comparison tests (NAMCT) were applied. D Mean ± s.e.m. reward-harvesting efficiency for artificial agents and animals. Agents include Random, Alternation, Rew_Prob, and Optimal_Baiting. Performance from animal D004 (left) and all animals combined (right; n = 4 animals, 76 sessions) is shown. Animals were compared to agents using two-sided Mann–Whitney U tests; *** indicates P < 0.001. Artificial agents were evaluated on the same reward schedules as animals. E Logistic regression for past right rewards, past left rewards, and past choices predicting current choice (mean ± s.e.m.; n = 82 sessions). F Alternation rate (mean ± s.e.m.) as a function of the difference in set reward probabilities (low = 0; medium = 0.3; high = 0.5–0.6). Differences across the three conditions were assessed using a linear mixed-effects model (LMM) with alternation rate as the dependent variable, reward-probability difference as a fixed effect, and animal identity as a random intercept. Two-sided P-values from the LMM were FDR-corrected using the Benjamini–Hochberg procedure; ** indicates P < 0.01 (P = 0.003 and 0.009). Data were collected from n = 2 animals across 45 sessions. Source data are provided as a Source Data file.