Three consecutive SEEG scores detected 14 of 320 later key responses before keypress after separate early calibration in eight recordings

We tested whether three consecutive scores improve a fixed SEEG response detector after separate early threshold calibration. We used all eight patients in OpenNeuro ds004473 version 1.0.2. Each rule used its own threshold with at most one observed false output per non-target calibration minute. The final 40 valid trials per patient supplied 320 test responses. Three scores detected 14 responses before keypress; one score detected 31. Patient mean detection was 4.38% versus 9.69%. Mean false outputs were 1.89 versus 1.66 per non-target minute. The mean detection change was -5.31 percentage points. An exact patient sign-flip test gave p = 0.125, under independent patients and sign symmetry. It did not establish a detection difference or equivalence. Adjacent score inputs overlap by 80%. The study concerns this fixed decoder in short, cued clinical recordings. It does not test stimulation. The package preserves four figures, all outputs, candidate thresholds, event matches, code, source versions, and repeat instructions. Independent rule checks passed. New raw retrieval reproduced all 34,268 continuous scores and 769 range hashes.

stereoelectroencephalography · brain-computer interface · neural decoding · continuous replay · threshold calibration · false outputs · response detection