Slow key presses do not lower held-out high-gamma press detection in 8 SEEG subjects; reaction time correlates weakly and positively with LDA move score (ds004473, response-locked windows)

We tested if a slow reaction time (RT) makes a high-gamma (70-150 Hz) SEEG press detector less accurate. Data: OpenNeuro ds004473 (cued key press, 8 subjects, 132-298 trials, 100-112 bipolar channels), archived features from sci 04683a2a, and RT from the public events.tsv files. A shrinkage LDA with 5 time-ordered folds scored held-out move windows (key press -0.25 to +0.25 s) against rest windows (before fixation). Result: the RT-to-score Spearman rho was positive in 7 of 8 subjects (mean +0.082, exact sign-flip p = 0.039; after trial-index detrending +0.099, p = 0.023). The AUC was 0.797 for the fastest third of trials and 0.824 for the slowest third (difference p = 0.070). Thus slow presses did not lower detection, and the effect is small (|rho| <= 0.19). Limits: only 8 subjects; response-locked windows can include cue-related activity on fast trials; there is no hand label. The package contains the code (numpy only), the features, the events files with checksums, the results JSON, and a figure.

SEEG · high-gamma · reaction time · motor decoding · shrinkage LDA · held-out evaluation · OpenNeuro ds004473