Uniform graph fits reached the smallest bound for one ratio error among unit flows in 105 of 154 connected SEEG pairs across six sessions

We tested linear response-map forecasts for 167 pairs in six epilepsy sessions. The source was public SEEG from OpenNeuro ds004696, version 1.0.1. Donors had to finish before test windows. All graph weights preserved node differences. Uniform fits reached the exact minimum influence of one ratio error in 105 of 154 pairs with a past path. A graph cut rule reduced influence in the other 49 pairs. The primary error permits one ratio change of at most a factor of two for each forecast. The reverse base and recorded target are fixed. At 20 to 500 ms, the mean of separate squared log error bounds fell from 2.0411 to 2.0323. This was a 0.43% reduction with equal session weights. Recorded error fell by 0.38%. A later check with one shared ratio error per session also gave a 0.38% gain. Only three sessions changed, and the descriptive sign-flip result was p = 0.25. The study does not establish a population benefit. The package retains pulse products, flow certificates, figures, forecasts, and verification results. It omits GPL-licensed analysis and extraction scripts, so the exact local code is not available for public replay. A fresh sub-06 derivative rebuild matched all 65 directed entries exactly. The other compact input files matched their archived hashes.

SEEG · stereoelectroencephalography · brain stimulation · response map · graph flow · calibration · error bound · bounded influence