Ten SEEG readers showed no clear contact training gain over matched controls (known simulated loss; 50% probability; 400 ms delay)

We tested whether contact loss training lowers later sound amplitude error in ten people with SEEG recordings. The public data were OpenNeuro ds004703, version 1.1.0. The test used 95 passages, 400 ms added delay, and a 50% simulated loss probability. Each fault removed one known physical contact and all selected pairs that used it. An exact regression solver fitted each loss rule from early passages. The control matched loss counts and every pair's omission probability. Contact training reduced mean error by 0.116 baseline percentage points. The descriptive 95% person bootstrap interval was −0.071 to +0.424 points. Six people had lower error, and four had higher error. One person supplied most of the mean decrease. The evidence does not establish a benefit or equivalence. A proof identifies the mask moments that change when pairs share contacts. The package preserves source features, retrieval recipes, models, predictions, uncertainty draws, checks, and three figures. All 58 primary files reproduced in a second analysis. Fresh source ranges reproduced one passage for each person. This exploratory reanalysis does not test intended speech, real hardware faults, a live device, or stimulation.

stereoelectroencephalography · brain-computer interfaces · speech amplitude · robust regression · contact loss · mask regression · neural decoding