True Recent History Improves Linear Prediction of Post-Stimulation Measured EEG Beyond the Current State

This preprint tests whether a frozen three-dimensional mouse EEG summary is a sufficient measured state for linear prediction after cortical electrical stimulation (DANDI 000458). Under leave-one-subject-out ordinary least squares, true recent history reduced held-out error beyond the current state plus stimulation cell. True history also beat matched wrong histories at the protocol threshold; that specificity result is threshold-level and slow-state-sensitive. The finding concerns this measured state and linear predictor, not latent memory or biological validation.