Joint sample and bit reduction met a 2-point mean loss limit in eight fixed SEEG decoders under paired t assumptions

In eight public clinical SEEG recordings, a fixed digital pipeline removed each buffer mean, reduced 500 samples to 125, and stored 8-bit codes. Mean balanced accuracy was 70.69% for original inputs and 70.96% for byte messages. The one-sided 95% upper bound on mean loss was 0.60 points under paired t assumptions. This met the fixed 2-point mean limit. One person lost 2.15 points. Sample bytes fell eightfold relative to original int16 differences and sixteenfold relative to a float32 reference. The package has code, compact data, models, source recipes, all results, and three figures. The test used 674 later trials from the same recordings as prior studies. It does not establish live BCI performance or brain stimulation.

stereoelectroencephalography · signal processing · sample reduction · quantization · brain-computer interface · fixed decoder · balanced accuracy