A fixed SEEG reader showed no clear format gain after onset on unseen even numbers in 11 patients

This study tested whether a small SEEG reader could identify seen digits versus spoken number words on later, unseen numbers. Each patient model selected eight bipolar pairs from early odd-number trials. Testing used later even numbers. In 11 patients, 234 training trials and 118 test trials, mean balanced accuracy was 53.1% after onset and 52.9% before onset. The mean gain was 0.15 percentage points. Its descriptive 95% patient interval was -12.01 to 11.59 points. The primary conditional p value was 0.1603 under format-label exchangeability within exact number and test-time block. The fixed test did not show a clear gain. It does not establish that the true gain is zero. This is an exploratory extension of reused clinical data from OpenNeuro ds005415 version 1.0.0. The package preserves features, trial maps, 44 models, all 472 decisions, null draws, source records, code, and three figures. A fresh source-range repeat reproduced all 649 available feature trials exactly. Independent raw and model checks passed. A clean repeat gave identical numerical results. The task measures perception. It does not test intended communication or brain stimulation.

stereoelectroencephalography · brain-computer interfaces · sensory format · number identity · held-out testing · ridge reader