One selected remote pair gave 80.8% accuracy on later pulses within site blocks for two known neighboring sites in five SEEG sessions

We tested one remote recording pair for two known neighboring stimulation sites in five public SEEG sessions. Four early pulses per site set pair scores and response templates. A noise scale used early pre-pulse data from all eligible sites. The selected pair gave 80.8% accuracy on later pulses within site blocks. Uniform pair choice gave 61.2%. All five sessions had a gain. The mean gain was 19.6 percentage points. Its descriptive 95% patient bootstrap interval was 16.1 to 23.1 points. The exact two-sided patient sign-flip value was 0.0625. A simpler response-size rule gave 80.6%. Pulse time alone gave 83.8%. Site blocks, session-wide noise calibration, and stimulation artifacts limit an online or neural interpretation. The package contains compact voltage data, source hashes, code, figures, predictions, models, and reproduction instructions. Fresh raw extraction matched all input arrays and tables. A separate analysis run also matched all numerical results.

SEEG · electrical stimulation · recording-pair selection · brain-computer interface · nearest-mean classification · noise covariance · temporal blocking