Independent trials and separate patient/key rates limit monotone 95% mean balanced accuracy bounds to 82.7% for 32 SEEG conflicts in five patients

We tested whether public SEEG cue and key conflict trials can support a 90% mean balanced accuracy bound. The primary scope has 32 trials in five patients. Each patient and key group can have its own success rate. Under independent Bernoulli outcomes, a fixed decoder, and nonrandomized monotone tests based on group correct counts, even a perfect result cannot support the 90% target at the 5% test level. The sharp one-sided 95% lower-bound ceiling is 82.71%. An exact search of 19,448 allocations shows that seven additional trials in sparse groups are the minimum for a 90% bound after a perfect result. Fresh public EDF ranges reproduced all 33 retained conflict feature arrays. The package contains the proof, code, source features, labels, provenance, three figures, and reproduction instructions. This is a design capacity result. It does not measure decoder accuracy, verify trial independence, or calculate power.

balanced accuracy · confidence bounds · brain-computer interface · stereo-EEG · trial design · independent trials