A check of the newest 11 SEEG samples accepts 500 ms windows with constant runs after stimulation in ds005558

We tested checks of SEEG input windows after nominal stimulation in OpenNeuro ds005558 version 1.0.0. We used all 11 bipolar SEEG recordings from seven people, with 754 stimulation events and 754 controls. The primary input had 500 samples at 1000 Hz. Each event had 101 completed windows after nominal train end. The first check accepted a channel when its newest 11 sample codes were not all equal. Some accepted windows still contained an earlier complete run of 11 equal codes. The mean conditional rate across people was 6.36%, with equal person weight. Two sessions from one person had rates of 37.07% and 52.44%. Controls had zero observed violations. A causal check of the full window excluded each complete qualifying run. For an isolated run with no later qualifying run, this check adds 489 ms under the primary settings. The measured median extra waits at a 90% channel threshold were 390 ms and 490 ms in those sessions. The package has a proof, stream code, sufficient run tables, 22 raw excerpts, four figures, source pins, and a rebuild recipe. Fresh source audits and an offline numerical replay passed. The large effect is from one person in a reused cohort. This work tests a repeated-code pattern and does not establish BCI accuracy.

SEEG · brain-computer interface · signal processing · sample repetition · streaming quality check · stimulation