Known task position gave 99.96% format accuracy without SEEG in 13 Numbers patients within source table groups

Other patients' task labels predicted 2,617 of 2,618 stimulus formats correctly without a brain signal in 13 public Numbers patients. Accuracy was 99.9618 percent. The reader required the source table group and original task position. It left each patient out and used only other patients in the same group. It also predicted 2,491 of 2,492 presented numbers and all response-trial labels correctly. Two observed task orders described all 2,618 source rows. On the same 211 later trials in an existing 11-patient SEEG test, task-position balanced accuracy was 100 percent. Fixed SEEG power and voltage readers gave 57.0 percent after onset and 55.2 percent before onset. Their training information differed from the task-position reader. A one-trial position error reduced mean format macro recall to about 20 percent. These results show that known task position can supply labels in this dataset. They do not show that a SEEG model used the schedule or that brain signals lack label information. The study provides a control method for BCI reading tests. It does not test intended communication or brain stimulation. The package preserves source code, labels, features, 22 raw signal examples, 33 models, predictions, figures, source records, and reproduction instructions. Fresh retrieval checked all 616 analyzed signal ranges. An offline repeat gave byte-identical numerical records.

brain-computer interface · stereoelectroencephalography · task order · label leakage · format decoding · cross-validation · neural decoding · reproducibility