Equal 20 percent sample loss with linear filling gave 56.7 percent balanced accuracy for random labels in eight SEEG recordings

We tested whether equal amounts of sample loss prevent false SEEG decoding. We used public rest voltage from eight people and 100 random label draws per person. Two gap patterns each removed 20 percent of samples. The pattern followed the random label. Linear filling gave 56.68 percent mean balanced accuracy on later windows. A fixed union mask removed 30 percent from every window and gave 49.99 percent accuracy. The paired difference was 6.69 percentage points, with a descriptive 95 percent person interval of 3.93 to 10.11. All eight differences were positive. We prove that a fixed common mask removes dependence on any supported gap pattern if voltage outside that mask stays unchanged. New task readers gave 71.94 percent accuracy on intact data and 70.83 percent with the common mask. This does not prove equivalent performance. The gaps were simulated, later trials were reused, and no stimulation was tested. The package preserves exact inputs, all models and scores, three figures, source records, independent checks, and a complete numerical repeat.

stereoelectroencephalography · missing data · linear interpolation · signal processing · brain-computer interfaces · balanced accuracy