A two-pair shaft limit did not establish higher minimum accuracy with mean replacement in eight simulated later-trial SEEG shaft-loss tests

We tested whether a limit of two bipolar pairs per electrode shaft improves an eight-pair SEEG decoder under complete loss of one shaft. We used 2,014 retained trials from eight public OpenNeuro ds004473 recordings. Models learned from earlier trials and were tested on later trials. We removed each candidate shaft in turn and assumed known loss with training-mean log-power replacement. Mean minimum balanced accuracy was 65.44% with standard selection and 65.77% with the limit. The paired gain was +0.33 percentage points (95% patient bootstrap interval -1.09 to 1.74; exact two-sided p=0.750). Three patients improved, three worsened, and two had equal scores. The evidence did not establish a gain or equivalence. The pair-count limit also did not set a 25% limit on absolute coefficient share. The package preserves code, features, source identities, all models and predictions, 123 shaft-loss cases per rule across patients, independent validation, and three figures. The result is conditional on offline trial exclusions, cued task-state labels, the replacement policy, and one split in each clinical recording. It does not test online fault detection or stimulation.

SEEG · brain-computer interface · shaft loss · fault tolerance · balanced accuracy · intracranial EEG · logistic regression · robustness