A joint SEEG check with five early pulses accepted 71.0 percent of later 20 to 500 ms maps after contact order reversal in eight people

We tested a published joint full conformal response check in public STReEF SEEG data. This cohort differs from the cohort used to develop the check. We used 96 fixed conditions in eight people and all 6,803 readout-condition pairs. Each condition had five early pulses and five later pulses with reversed recorded stimulation contact order. At 20 < t < 500 ms, the unchanged check accepted 341 of 480 later maps, or 71.0 percent. The descriptive 95 percent person bootstrap interval was 60.6 to 80.2 percent. The nominal one-candidate level is 83.3 percent only under complete-pulse exchangeability. The data do not establish that assumption. The accepted set also contained a complete zero map in 76 of 96 conditions, or 79.2 percent. Requiring every separate readout check to pass accepted only 1.7 percent of later maps. Joint-set nesting explains this gain. Contact order and time changed together. The zero and voltage-gain checks were hypothetical, not measured device faults. These results show a limit of transferring this rule to these larger maps. They also show why acceptance alone cannot certify an evoked response. The package preserves the plan, exact pulse products, source metadata, code, ranks, figures, and retrieval instructions. A fresh rebuild checked all 960 pinned public raw ranges. All inputs and numerical results were reproduced exactly.

SEEG · conformal prediction · brain-computer interface · electrical stimulation · response monitoring · epilepsy