Early and delayed selection gave 42.5% and 45.6% mean delayed SEEG accuracy for eight remote sites per session in five 6 mA sessions

We tested whether delayed responses select better stimulation sites for delayed brain-interface readout. We used public SEEG data from OpenNeuro ds004457 v1.0.2 in five people with epilepsy. All retained stimulation sites used 6 mA and were outside the fixed recording shafts. Each candidate site supplied four calibration pulses and four separate later test pulses. The primary test selected eight sites per session from either 20–100 ms or 100–180 ms responses. Both sets used the same delayed-response decoder. Early selection gave 42.5% mean delayed accuracy. Delayed selection gave 45.6%. The mean gain was 3.1 percentage points, with a descriptive 95% patient bootstrap interval of −3.75 to +8.75 points. The exact sign-flip reference p value was 0.50. These sessions give no clear evidence of a benefit from delayed selection under this rule. Before-pulse signals showed background differences in one session. The data came from site blocks, so this test does not establish a stimulation-specific neural effect or online feedback utility. The package includes two figures, voltage inputs, source hashes, code, models, predictions, and replay instructions. Fresh raw extraction reproduced all inputs byte for byte. Numerical replay reproduced all 16 analysis files. The earlier source result was also reproduced exactly. All results use the same source cohort and are not independent patient replications.

stimulation site selection · stereoelectroencephalography · SEEG · brain-computer interfaces · response-window comparison · voltage decoding · calibration · epilepsy