What drives the next state?
Schematic phase portrait: a moving point, blending a path dictated from outside with the population’s own dynamics. Drawn in neutral until you answer. Illustrative; the measured quantity is the readout.
Which system produced this?
Every number here is the real measured value. Brains: rat orbitofrontal cortex (Schiereck et al. 2026; 164 sessions, 31 rats) and primate entorhinal cortex (Neupane et al. 2024; 14 sessions, 2 monkeys). The commitment result on the poster comes from a third, separate recording set (Luo et al. 2025; 115 sessions, 12 rats). Models: five instruction-tuned AI models, 1.5B–9B parameters, three families.
Curves: the per-time-point mean ± SEM of the dominant autonomous eigenvalue |λ| across the 115 sessions of the rat frontal and striatal recordings (Luo et al. 2025; demix, Huang et al. 2026), as deposited with the preprint and drawn on the poster: flat near 0.90 before commitment, about 0.93 by +250 ms. Per session, the rise between windows centered at −150 and +200 ms is a median +0.016, p = 2.2×10⁻⁵, in 75 of 115 sessions and, by rat mean, 10 of 12 rats; animal-nested 95% interval for the mean rise +0.005 to +0.044, while a mixed-effects fit over rats gives p = 0.098. Aligned to the stimulus (zero at 100 ms after onset, as on the poster), a smaller and more gradual rise: +0.002 in the primary windows (n.s.), up to +0.010 with the window spacing matched. Specificity to commitment is supported, not established. The naive-DMD view is an illustrative sketch of the paper’s description of that estimator’s read-out on the same data (flat and noisy); its curve is not deposited.