A distance rule gave no clear mean error gain and lower site accuracy in five offline SEEG tests at 20 to 100 ms

We tested whether a distance rule improves the readout of a known stimulation site from public SEEG pulse responses. We used five patient sessions from OpenNeuro ds004457 version 1.0.2. Four early pulses per site fit the reader. Four later pulses gave 724 tests at 181 sites. Both decisions used the same fitted site probabilities. At 20 to 100 ms, the distance rule gave 21.11 mm mean error and the most-likely-site rule gave 22.55 mm. Each patient had equal weight. The 1.44 mm decrease had a descriptive 95% patient interval of -0.98 to 4.20 mm. Exact site accuracy fell from 41.19% to 32.30%. Errors above 20 mm rose from 37.53% to 42.95%. A proof gives the same tradeoff with exact probabilities. A geometry-only rule explains the apparent gain in the before-pulse control. The package includes compact data, code, fitted models, pulse results, source hashes, a raw retrieval recipe, and three figures. All 110 raw source files passed hash checks. A raw rebuild and an independent analysis matched the saved inputs and results. This is an offline study in a reused clinical cohort. The time split is later within each site, rather than after the full fit period. Physical distance does not measure function or harm. These tests do not show a general gain for an interface in use.

stereoelectroencephalography · site decoding · brain-computer interface · distance loss · site accuracy · human neuroscience · offline evaluation