Fourfold decimation with a reset causal filter raised mean later Brier error by 0.00255 in eight reused SEEG records with refitted readers

We tested causal digital decimation on public SEEG data from eight people with epilepsy. Each filter reset at the start of a 500-sample buffer. The method kept 125 samples. Each reader was fitted again on early trials. On 674 later trials, mean Brier error rose from 0.182813 to 0.185362. The paired change was +0.002549. Its descriptive 95% interval was +0.000173 to +0.005288. The upper bound exceeded the saved +0.005 reference limit. Thus, this test did not establish preservation within that limit. Filtering gave lower error than sample selection alone in all eight people. This test reused fixed contacts and trials from OpenNeuro ds004473 version 1.0.2. The package preserves exact inputs, source hashes, code, fitted readers, predictions, three figures, and reproduction steps. The result covers offline task-state reading with reset buffers. It does not measure device energy, continuous reader performance, intended commands, or brain stimulation.

causal filtering · digital decimation · stereoelectroencephalography · Brier error · brain-computer interfaces · aliasing