Five offline SEEG sessions gave 90.6 percent accuracy for eight sites at mean 60 ms cutoffs with 20 ms starts

We tested whether calibration could choose a shorter cutoff for stimulation-site identification. We used public SEEG recordings from five patients in OpenNeuro ds004457 version 1.0.2. This study extends prior work on the same cohort. Screening used 732 calibration pulses across 183 candidate sites. Each session scored 32 later pulses at eight selected sites. The voltage window started at 20 ms. Four calibration folds selected cutoffs of 80, 40, 60, 60, and 60 ms after the pulse. Mean cutoff time was 60 ms, compared with 100 ms. Chosen cutoffs identified 145 of 160 test pulses correctly, or 90.625 percent. The 100 ms comparison identified 142, or 88.750 percent. The mean paired gain was 1.875 percentage points. Its descriptive 95 percent patient bootstrap interval was -0.625 to +4.375 points. The exact sign-flip p value was 0.50. These results do not establish a population accuracy gain or equivalence. A planned 30 ms start lost 1.875 points against its 100 ms comparison. Matched before-pulse accuracy was 13.750 percent; balanced chance was 12.5 percent. The fitted reader uses only bins that end before its cutoff. We checked all 110 raw source hashes. A fresh raw rebuild matched all 30 input arrays and five event tables exactly. Independent checks matched 75 site fits and 4,000 predictions. A complete repeat matched all numerical outputs. The package preserves code, compact inputs, fitted models, pulse decisions, three figures, and retrieval instructions. This offline test does not establish neural origin, live feedback benefit, or successful writing to the brain.

stereoelectroencephalography · SEEG · electrical stimulation · site identification · calibration · brain-computer interface