Setting negative SEEG power estimates to zero left a mean bias of 2.16% of calibration power under random pulse signs in eight sessions

We tested whether a before-pulse subtraction removes bias from setting negative shared SEEG waveform power to zero. We used OpenNeuro ds004696 version 1.0.1, with eight sessions, 86 stimulation sites, and 886 readout pairs. We compared the first five and final five good pulses at each site. All 1024 independent pulse-sign assignments gave an exact zero mean for the signed estimate. Setting negative products to zero in each window left a mean adjusted bias of 2.16% of first-group single-pulse power. The descriptive 95% person interval was 1.49% to 2.89%. All eight person estimates were positive. The proof shows that subtraction cancels this bias only when both windows have equal expected absolute products. Pooling readouts before the zero limit reduced each separate floor, but the adjusted bias remained 1.88%. On the recorded signs, zero limits lowered the adjusted mean from 32.09% to 31.33%. The control is constructed. It does not estimate brain noise or clinical error. The package preserves exact numerical inputs, all scores, three figures, source hashes, and retrieval instructions. Fresh extraction matched all eight source files. A full replay reproduced every numerical result file byte for byte. GPL analysis and extraction scripts are omitted to meet SCI code terms, so the public package cannot run that replay directly.

SEEG · intracranial stimulation · shared waveform power · sign control · zero limit · measurement bias · response mapping