Ten-pulse response-label AUC exceeded mean five-pulse subset AUC at 10 to 90 ms in UMCU21 SEEG with agreed binary labels fixed

We tested pulse-count comparisons in one public SEEG record, UMCU21, from OpenNeuro ds004774 v1.0.0. The test used surface stimulation, six depth readout contacts, and 188 cells with two expert labels. With agreed binary labels fixed, 38 disputes permitted 2^38 label vectors. For a fixed absolute voltage score at 10 to 90 ms, the planned ten-versus-first-five AUC difference had exact limits of -0.010253 and +0.064570. The planned mean-subset check averaged all 252 five-pulse AUCs. Its limits were +0.008105 and +0.041217 under a common label vector. Thus, ten pulses exceeded this average for every allowed label choice. Only 64 of 252 subsets had positive lower label limits. The mean-subset result retained a positive lower limit at 20 to 90 ms, but could reverse for N1 alone. These are finite-data sensitivity limits, not confidence intervals or evidence of online interface benefit. The package includes three figures, source epochs and labels, all scores and attained-limit labels, an exact-bound proof, source hashes, code, and replay instructions. All 330 raw epoch ranges and nine source files passed fresh download checks.

evoked potentials · stimulation · AUC · label uncertainty · SEEG