Fixed early fits from two inputs had negative group mean scores for later 70–110 Hz signals in seven SEEG rest records

We tested later signal reconstruction from two nearby bipolar inputs in 511 targets from seven public SEEG rest records. The shared inputs contained target contacts. The disjoint inputs used separate contacts. Fits used 10–130 seconds and tests used 350–470 seconds. The main band was 70–110 Hz. Targets had equal weight within each person, and people had equal group weight. Mean R² was −1.825 for shared inputs and −1.212 for disjoint inputs. The paired gap was −0.613. Its descriptive 95% person bootstrap interval was −3.536 to 1.301. The exact sign-flip p value was 0.969. The fixed comparison did not establish a shared-input advantage. A few extreme errors drove negative means. A description added after the main result showed large input power changes with the worst errors. All targets were retained. The package includes a tool that checks which montage inputs survive contact loss. It also gives an independent-contact proof for a shared-input score of one half. Four figures, source code, small replay data, full source hashes, a retrieval recipe, and checks are preserved. A fresh full-source extraction reproduced the numerical results exactly. The test measures offline signal reconstruction in the same session. It does not establish neural information recovery, BCI task accuracy, or a stimulation effect.

stereo-EEG · SEEG · signal reconstruction · bipolar montage · brain-computer interface · ridge regression · gamma band · reproducibility