Time pooling reduced SEEG prediction band width by a mean of 61.7% and mean four-pulse coverage fell to 86.3% in five patients

We tested prediction bands for completed 20–100 ms stimulation responses in public SEEG data from OpenNeuro ds004457 version 1.0.2. Each site used four fit pulses and four later pulses. Earlier sites set a maximum-error threshold for all pulses, recording pairs, and time bins. Later sites tested it. Pooling variance across time reduced normalized band width by a mean of 61.7% across five patients. Mean complete-batch coverage fell from 93.2% to 86.3%. The two methods covered 52 and 48 of 56 tests. Pointwise Student t limits covered 90.4% of coordinates but only 8.8% of complete batches on the equal-patient mean. The result shows a width and coverage tradeoff in this selected cohort. It does not establish equal coverage, a future coverage guarantee, or clinical benefit. The package preserves source bins, pinned retrieval code, all results, four figures, and replay instructions. A fresh raw-data rebuild exactly matched all 30 input arrays. A clean replay reproduced all scientific tables and 95 prediction arrays.

stereoelectroencephalography · electrical stimulation · prediction bands · conformal prediction · coverage · brain-computer interface