A fixed SEEG reader showed excess contact overlap in wrong site calls at 20 to 100 ms in five patients

We tested whether wrong stimulation site calls stayed near the true site in public SEEG data. We used OpenNeuro ds004457 version 1.0.2: five patients, 138 sites, and 15 fixed bipolar recording pairs. Four early pulses per site fitted a fixed covariance reader. Four later pulses tested it. At 20 to 100 ms, mean site accuracy was 33.71% and shaft accuracy was 60.39%. Among wrong calls, the share on the true shaft exceeded a uniform wrong-site reference by 22.99 percentage points after the pulse and 3.60 points before it. The paired increase was 19.40 points, with positive values in all five patients. A descriptive patient bootstrap interval was 8.37 to 30.06 points; the exact two-sided sign-flip reference was 0.0625. Shared stimulation contacts supplied 97.1% of this observed group increase. Separate pairs on the same shaft added only 0.56 points. The result supports a spatial error audit for stimulation feedback readers. It does not establish neural cause or live interface performance. The split was retrospective, site order was blocked, and these sessions had prior use. The package preserves every call, all fitted models, three figures, source code, data, and provenance. A fresh raw rebuild matched all 2,453,000 voltage bins and all event fields. A separate numerical replay matched all results exactly.

stereoelectroencephalography · SEEG · electrical stimulation · brain-computer interface · signal classification · spatial error analysis