Surviving contrasts raised mean 20–100 ms condition accuracy from 71.24% to 73.15% after known simulated contact loss in eight SEEG patients with unchanged surviving signals

We tested whether a difference between surviving recording contacts can improve a fixed SEEG reader after one known simulated contact loss. We used eight patients from public STReEF data, OpenNeuro ds005448 version 1.0.0. Each patient had four known stimulation conditions, 20 calibration pulses, and 20 separate test pulses. Calibration selected four adjacent bipolar pairs. The reader used eight time coefficients per pair from 20 to 100 ms. All surviving voltages stayed unchanged. Each selected physical contact was made unavailable in turn. Mean accuracy was 71.24% after deletion of all affected pairs and 73.15% when the reader retained a contrast across a lost internal contact. The equal patient gain was 1.92 percentage points. Six of 58 contact cases permitted a bypass; all six improved. The descriptive 95% patient bootstrap interval was 0.54 to 3.79 points. The exact two-sided sign-flip value was 0.0625. This small test does not establish a population benefit. This result extends an earlier bypass study to another public dataset and a reader with calibration-selected pairs. The voltage identity and projection rule are prior mathematics. All 320 fresh raw range hashes matched. All 421,120 feature values were rebuilt exactly. Independent distances and a separate replay passed. The package has code, a direct contact reader, models, decisions, three figures, source records, and retrieval instructions. The split is retrospective. The test does not establish live interface performance.

SEEG · intracranial EEG · neural signal processing · contact loss · bipolar recording · fault tolerance · stimulation response · condition classification · epilepsy · calibration transfer