Adding 70–110 Hz power raised mean balanced accuracy from 65.7% to 68.9% for windows before responses versus rest in later trials from eight SEEG records

We tested whether 70–110 Hz power increased task-state accuracy more after a key response than before it. We used public SEEG data from eight people in OpenNeuro ds004473 version 1.0.2. Each reader used the same eight bipolar pairs for both response periods. Low-band differences in early trials set the pair ranks. Early trials trained separate logistic readers. A total of 1,000 later trials tested them. Adding 70–110 Hz power to 8–12 and 13–30 Hz power raised mean balanced accuracy before the response from 65.75% to 68.90%. All eight people improved. The mean gain was 3.15 percentage points before the response and 1.16 points after it. The planned after-minus-before gain was -2.00 points. Its exact two-sided sign-flip reference was p = 0.08594. A descriptive person bootstrap interval was -3.75 to -0.10 points. These summaries do not establish a general difference between periods. With an earlier rest baseline, the before-response gain was 1.61 points. A rest-position control scored about 49%. All 121 reused file hashes passed. Independent source checks passed for eight stored ranges and 16 new ranges. A full numerical repeat matched all eight output files. The package preserves features, source records, code, readers, predictions, checks, tables, and three figures. The result applies to selected task windows in a small reused clinical cohort. It does not establish live intention decoding or a brain stimulation benefit.

stereoelectroencephalography · brain-computer interface · balanced accuracy · high gamma · task-state decoding · logistic regression · signal features