A scaled pulse triplet kept five-pulse SEEG forecast size bounded under two replacements and increased error in eight unchanged sessions

We tested a short calibration rule for recorded responses to brain stimulation. The rule selects the closest three of five SEEG pulse waveforms. It then scales their mean by a value from zero to one. We proved that its forecast size stays bounded when any two inputs change and the other three stay fixed. This is an application of established minimum diameter averaging. In public OpenNeuro ds004696 data, we tested eight patient sessions, 86 stimulation sites, and 886 site-readout pairs. The first five good pulses fitted the rule. The final five pulses tested it. Under two specified artificial common faults at gain +10, the mean patient error reduction was 84.02% against the existing scaled mean of five. The 95% patient bootstrap interval was 82.25% to 85.77%. On unchanged calibration data, the mean error increase was 5.63%. Moderate artificial faults could also increase error. The proof limits forecast size. It does not guarantee low error or safe stimulation. The package contains exact trial products, source hashes, figures, and archived results. The GPL source code remains in the local archive and is omitted from the SCI package.

stereoelectroencephalography · robust aggregation · minimum diameter averaging · pulse response · brain-computer interface · fault tolerance · forecast error