A fixed monitor warned in eight SEEG replays with output above 5%; its warning bound requires a conditional rate limit

We tested repeated checks of candidate output in eight public SEEG records from OpenNeuro ds004473 version 1.0.2. A fixed mixture monitor warned in all eight streams with window thresholds and in none with interval thresholds. The first warnings occurred after 4 to 41 eligible intervals. These records and output rates come from prior work. The new contribution is a SEEG audit tool and an exact control benchmark. The proof uses an established probability method. With independent output at rate 5%, repeated binomial tests warned by 146 checks with 19.81% probability. The mixture warned with 2.29% probability. Its bound of at most 5% for any warning requires a conditional output rate of at most 5% after each past history. A correlated control with marginal rate 5% violated that assumption and gave 29.41% warning probability. At independent output rate 10%, the mixture warned by 146 checks with only 47.72% probability. No warning therefore does not prove acceptable output. Background labels use later keypress records and do not prove absent intent. The study contains no stimulation test. The package preserves source features, models, pinned retrieval records, new code, all monitor paths, exact controls, figures, and reproduction instructions. Fresh checks matched 16 public raw ranges in eight people.

sequential monitoring · brain-computer interface · stereoelectroencephalography · conditional error bound · exact probability · candidate output