A joint score for 20 to 100 ms SEEG responses reduced mean before-pulse acceptance and increased mean wrong-site acceptance in five offline sessions

A fixed joint score changed the errors in an offline SEEG command check. The study used OpenNeuro ds004457 version 1.0.2, five people, 183 stimulation sites, and 732 test pulses. Three early pulses per site fitted a common response model. A fourth pulse set each score threshold near 80 percent early acceptance. The joint score compared the commanded site with an equal mixture of background and other known sites. Against the other-sites score, mean before-pulse acceptance fell from 25.67 to 23.33 percent. Mean wrong-site acceptance rose from 16.81 to 18.19 percent. True-site acceptance changed from 76.21 to 76.97 percent. The primary equal mixture of negative inputs improved by only 0.48 percentage points. Its person bootstrap interval was -2.66 to +1.70 points; the sign-flip reference p value was 0.75 under independent symmetric null person differences. Person changes were mixed. These are hypothetical command pairings and before-pulse proxies, not observed device faults. Most test pulses preceded full session calibration. The result applies to offline models, not live control. All 110 source files passed fresh hash checks. Raw extraction and analysis replay matched exactly. The package has figures, code, all binned inputs and scores, source receipts, and repeat instructions.

SEEG · brain-computer interface · response verification · likelihood ratio · intracranial stimulation · offline validation