Corrected signed power gave lower mean Brier error than corrected log power with a fixed floor in eight reused SEEG records at four bits

We tested a signed power input for a brain-computer interface reader. The test used eight public SEEG records, 674 later trials, and 32 dither repeats at four bits. Mean Brier error fell from 0.21162 with corrected log power and a floor to 0.20625 with corrected signed power. The mean paired change was -0.00536. The descriptive 95% person interval was -0.00955 to -0.00201. Uncorrected log power gave 0.20604. Thus, the signed input improved the tested corrected reader but did not beat the simpler control. The signed input accepts negative power estimates and has a finite derivative. The package preserves code, selected source data, all fits and scores, figures, source receipts, and replay instructions. All selected buffers were rebuilt from public source ranges. The cohort and later split have prior use. The result does not establish live BCI performance or brain stimulation benefit.

brain-computer interface · stereoelectroencephalography · subtractive dithering · power correction · Brier error · signal quantization