Bipolar SEEG shape at 20 to 100 Hz gave 55.78 percent expected accuracy for shared fair random shaft labels in 33 patients after target contact exclusion

We tested whether bipolar SEEG shape can transfer shared random shaft labels after exclusion of both target contacts. We used OpenNeuro ds004100 version 1.1.3 and the first 60 seconds of two interictal files per patient. The primary sample had 33 patients, 409 named shafts, and 3,534 pairs. We removed spectral power level and linear log-frequency slope at 20 to 100 Hz, with 55 to 65 Hz excluded. Each shaft had one independent fair binary label shared across files. Labels did not affect source selection. We averaged exactly over labels and uniform source subsets. The source count matched the count after full target shaft exclusion. Expected nearest-source accuracy was 55.78 percent. Uniform source choice gave 53.35 percent. The equal-patient mean gain was 2.43 percentage points, with a 95 percent patient bootstrap interval of 1.49 to 3.54 points. Full shaft exclusion gave exactly 50 percent by proof under the label assumptions. The fixed disjoint-pair check gave a 2.31-point gain in 32 patients. All six fixed secondary checks are reported. The test identifies a possible route for arbitrary label transfer within familiar patients. It does not show clinical model leakage, a neural cause, or performance in new patients or independent sessions. The package preserves input spectra, source pins, source weights, code, results, figures, and reproduction instructions. Independent mathematical checks and a full repeat analysis passed. Fresh pinned public prefixes reproduced all archived spectra from all 72 source files.

stereoelectroencephalography · SEEG · spectral shape · random labels · shaft-level evaluation · source exclusion · patient bootstrap · model evaluation