A 20-trial delay in correct feedback changed mean absolute coverage error by 0.23 percentage points in eight SEEG replays at a 90% target and a 0.05 step

We tested delayed feedback for a fixed SEEG reader. Feedback labels were correct. The test used eight recordings from OpenNeuro ds004473, version 1.0.2. It used 674 later trials. Each trial had rest and pre-response windows. The whole-trial coverage target was 90%. The update step was 0.05. A 20-trial delay changed mean absolute coverage error by +0.23 percentage points relative to no added delay. The 95% paired person bootstrap interval was -0.12 to +0.62 points. This primary result was uncertain. Mean single-label output fell from 34.31% to 28.67% of windows. Mean error among these outputs was 17.59% and 19.60%, respectively. The package gives a delayed feedback benchmark, exact state arithmetic, and an audit of pending feedback. It includes code, feature arrays, fixed reader inputs, every trial decision, four figures, source hashes, and retrieval instructions. A fresh source rebuild reproduced all retained features exactly. The cohort and reader had prior use. Response alignment and later eligibility make this a retrospective test. The result does not establish live BCI performance or a brain stimulation effect.

stereoelectroencephalography · adaptive conformal inference · delayed feedback · brain-computer interface · coverage · retrospective replay