At a fixed rest threshold, 200 ms confirmation reduced background commands and response hits in eight offline SEEG tests

We tested a causal command rule with public SEEG data from OpenNeuro ds004473 v1.0.2. Eight patients had clinical electrodes and performed a key task. Models and a 95th-percentile early-rest threshold were fixed before the later test. Three consecutive positive scores spanned 200.118 ms. The rule reduced the equal-patient mean background command rate from 15.29 to 4.28 per eligible minute. The paired change was -11.01 (95% patient interval -13.24 to -8.82; exact sign-flip p = 0.0078125). Pre-keypress hits fell from 34.64% to 16.05%. Both measures fell in all eight patients. There were 816 test responses and 72.109 eligible background minutes. A pure delay control retained more response hits. That control was added after the primary results and is exploratory. This work supplies a reproducible test of a known confirmation method. It does not establish an online operating setting. Background commands are a proxy for unwanted commands because continuous intent labels are absent. The package contains features, fixed models, command arrays, four figures, source versions, retrieval code, and independent checks. All 26 numerical output files matched a separate execution. Fresh raw checks covered one window per patient and all continuous windows for two patients. No new-day or stimulation test was done.

stereoelectroencephalography · brain-computer interfaces · command confirmation · response detection · causal decoding · offline replay