Early power selection gave 0.931 mean AUC between later after-pulse and before-pulse SEEG windows in eight epilepsy patient sessions

We tested selection of one SEEG recording pair per stimulation site in eight epilepsy patient sessions. Selection used five early good pulses. Testing used five later good pulses. The set had 86 sites and 886 candidate pairs. The power rule selected the largest early log power increase. The shape rule selected the most repeatable early waveform. Mean later AUC between after-pulse and before-pulse power was 0.931 for power selection. It was 0.914 for shape selection and 0.774 for uniform random pair choice. The mean patient difference was 1.70 percentage points. The planned exact test gave p = 0.0625. It did not establish an advantage at the 0.05 level. The fixed early power threshold gave 90.0% later sensitivity and an 8.7% before-window positive fraction. The package preserves the plan, exact data products, complete scores, four figures, source hashes, code, and retrieval instructions. Fresh source extraction checked three patients. Separate source-sample scores checked two patients. The result supports a pulse-aligned readout benchmark within these sessions. It does not establish neural origin, continuous detection, or useful brain writing.

stereoelectroencephalography · SEEG · electrical stimulation · channel selection · window power · area under the curve · pulse-aligned monitoring