RMS ratios kept decisions under simulated gains shared by both windows, with no clear accuracy gain in eight SEEG patients

We tested a reader of four stimulation sites in eight patients. We used public SEEG from OpenNeuro ds005448 version 1.0.0. Each site had five fit pulses and five later test pulses. Four recording pairs were selected from fit responses. Division of response root mean square (RMS) by RMS before the pulse gave 61.25% accuracy. Response RMS alone gave 64.375%. The equal-patient difference was -3.125 percentage points. Its descriptive 95% patient bootstrap interval was -10.625 to +5.000 points. The ratio reader kept all 16,160 site decisions under simulated gains shared by both windows. A proof states the gain conditions. Timing alone identified 158 of 160 sites. This limits claims about neural origin and performance after all fitting. The package preserves 52,640 new RMS values, fitted readers, pulse decisions, three figures, source checks, and repeat instructions. Independent checks passed. A second run matched all numerical outputs. The data were reused, and the study is exploratory.

stereoelectroencephalography · brain-computer interface · root mean square · signal normalization · site classification · recording gain