A fixed contact norm preserved five-pulse SEEG medians across three connected layouts from the same contact voltages in eight patients

We tested pulse forecasts from public stereoelectroencephalography (SEEG) data in eight patients with epilepsy. Each of 83 sites had five early pulses and five later pulses. Three connected layouts used differences from the same three contact voltages. We recovered contacts with zero spatial mean, then fitted one geometric median to each set of early pulses. The largest layout change was 1.76e-15 of early contact RMS, including simulated gain faults. An ordinary edge-metric median changed at every clean site. Its mean chain-to-star change was 3.16% of early contact RMS. With one early pulse multiplied by four, the contact median reduced later error by 40.47% versus the ordinary mean. It gave no clear error gain versus a mean scaled from estimated noise. The archive contains a proof, exact voltage windows, source records, results, and three figures. GPL analysis and retrieval scripts are omitted, so this SCI archive does not support a code-based replay. This is an offline test in a reused cohort. It does not show an online brain-interface benefit.

stereoelectroencephalography · geometric median · signal processing · contact graphs · brain-computer interfaces · robust estimation · pulse response