In 29 records marked as SEEG, calibration prefixes of 20 seconds increased pair overlap but did not establish a mean power gain

We tested eight-pair selection from 70–110 Hz power in public interictal records from 29 subjects. The released channel tables mark the contacts as SEEG. Calibration used the first 2, 5, 10, or 20 seconds. Test power came from seconds 40–60. Mean retained test power rose from 88.00% after two seconds to 89.45% after 20 seconds. The difference was +1.45 percentage points, with a 95% subject bootstrap interval of -3.92 to +6.04. This primary result did not establish a mean power gain. Mean overlap with the eight strongest test pairs rose from 5.76 to 6.41. The difference was +0.66 pairs, with an interval of +0.10 to +1.17. A check with both calibration windows ending at 20 seconds gave an uncertain overlap difference of +0.31 pairs (-0.21 to +0.76). Thus the prefix result does not isolate a duration effect. A planned median-power check gave higher retained median test power, but it used a different test target. The package preserves 168,960 pair-second power values, selected lists, three figures, source versions and hashes, code, and retrieval instructions. An independent EDF reader checked all values for one saved raw prefix. A fresh numerical replay passed. This is a recorded-power benchmark; task decoding and stimulation were not tested.

stereo-EEG · intracranial EEG · channel selection · high-gamma power · calibration · OpenNeuro