A frequency derivative projection left a 0.294 mean Brier error increase under added line signals with expected RMS equal to training RMS and 0.5 Hz fundamental shifts in eight SEEG records

We tested line removal on public SEEG buffers from eight people in OpenNeuro ds004473 version 1.0.2. Each buffer had 500 samples. The reader classified before-fixation and before-response states on later trials. All methods used the same fixed trial and contact choices. A projection removed 60 and 120 Hz sine signals and their first frequency derivatives. Its original mean balanced accuracy was 72.10%. We added two line signals with combined expected RMS equal to each pair's training RMS. The fundamental shift was -0.5 or +0.5 Hz. The second harmonic had twice that shift. The planned mean Brier error increase was 0.29388 for derivative removal and 0.30896 for fixed sine removal. The paired contrast was -0.01508. Its descriptive 95% person bootstrap interval was -0.03352 to -0.00128. Derivative removal still gave only 50.05% balanced accuracy. At 0.1 Hz shifts, its error stayed close to the original error. The planned secondary method omitted 114 to 126 Hz from high frequency power. Under the larger added shifts, its error increase was 0.001108 and its balanced accuracy was 71.26%. A proof gives exact invariance at nominal frequencies and a local residual bound. The bound does not ensure reader accuracy. Independent software checks and a source voltage audit passed. This is an offline test with simulated interference. It does not identify natural line noise or test brain stimulation. The package contains voltage, code, fitted readers, results, three figures, source pins, and reproduction instructions.

stereoelectroencephalography · line interference · signal processing · frequency derivative · Brier error · short buffers · brain state classification · SEEG