Variance correction gave a mean 20–500 ms SEEG voltage score gain of 0.178 with five calibration pulses in eight sessions

We tested whether a variance penalty improved the choice of one recording pair after stimulation. The test used public SEEG data from OpenNeuro ds004696 version 1.0.1. It had eight patients, 86 sites, and 886 candidate instances. Each site used five early pulses for calibration and five separate later pulses for testing. The corrected rule selected the largest mean waveform energy minus twice the estimated variance of the five-pulse mean. For 20–500 ms, it changed 25 of 86 choices. The equal-patient mean gain was 0.178 on a shared site calibration energy scale. The 95% patient bootstrap interval was 0.025 to 0.407. The two-sided patient sign-flip reference gave p = 0.0625. Five patients had a gain, one had a loss, and two were ties. The gain before stimulation was larger, at 0.516. Both rules had negative mean utility in that control window. The result concerns predictable measured voltage. It does not establish a neural source or live BCI benefit. The package includes four figures, all scores, a selection tool, a conditional fixed-candidate proof, source versions and hashes, and reproduction code. A new complete source download matched all 556 matrices in one session. A repeat analysis reproduced all eight result files with identical bytes. This is an exploratory extension on a cohort used in earlier sci studies.

stereoelectroencephalography · electrical stimulation · brain-computer interface · variance correction · recording-pair selection · predictive utility