Simulated contact loss increased beta repair error in 21 SEEG patients under a fixed model

A physical contact can serve two bipolar SEEG channels. This study tested simulated, known contact loss in 35 public records from 21 patients. A fixed spatial model was trained on an early beta window. It was tested on a separate later window. Contact loss removed both channels that used the contact. The control removed the target and one unrelated channel. It held channel count constant. Contact loss increased normalized repair error in all 21 patients. The patient mean increase was 0.0796. The 95% patient bootstrap interval was 0.0659 to 0.0943. The exact two-sided patient sign-flip probability was 0.000000954. The primary contact-loss error was 1.1289. A zero prediction has error 1. Thus, the result supports a physical-contact fault benchmark. It does not show useful mean repair by the primary model. Planned secondary checks gave the same contrast direction. The package preserves compact data, source hashes, source checks, model matrices, code, scores, and figures. All 355 source files passed fresh hash checks. A fresh extraction matched every packaged array. These short epilepsy records and simulated losses do not test a live BCI or safe brain writing.

stereoelectroencephalography · SEEG · contact loss · signal repair · brain-computer interfaces · fault benchmark