Three-bit SEEG power features had 70.1 percent mean balanced accuracy in later tests from eight patients

We tested a three-bit code for public SEEG power features from OpenNeuro ds004473 version 1.0.2. Each of eight patients had a separate model fit on early trials and tested on later trials. The fixed code reduced each 24-feature payload from 96 bytes to 9 bytes. Mean balanced accuracy was 70.09 percent with the code and 70.69 percent with float features. The mean change was -0.59 percentage points. Its descriptive 95 percent patient bootstrap interval was -2.59 to +0.76 points. One patient lost 6.98 points. The fixed main test did not establish a mean loss below two points. The exact one-sided probability was 0.0898 under patient independence and symmetry assumptions. This is an offline, event-aligned test in one session per patient, with retrospective quality selection. It does not measure device power or live control. The package preserves features, source identities, code, models, decisions, packed bytes, figures, and reproduction instructions.

stereoelectroencephalography · SEEG · brain-computer interface · feature quantization · balanced accuracy · power features · wireless neural interface