An 80 percent band gave 82.1 percent mean coverage of test SEEG pulse waveforms in five patients at 50 to 250 ms

We tested uncertainty for one later SEEG pulse in public OpenNeuro ds004457 version 1.0.2. The test used five patients and 1,081 site and receiver pairs. Each ten-pulse group had five fit pulses, four calibration pulses, and one later test pulse. We scored 103 bins from 50 to less than 250 ms after stimulation. At an 80 percent target, the scaled whole-waveform band covered 82.11 percent of complete test waveforms, with equal patient weights. Mean full width was 861.85 microvolts. Pointwise calibration covered 80.27 percent of bins and 15.22 percent of complete waveforms. A constant-width band covered 80.18 percent of complete waveforms at 480.02 microvolts. The Bonferroni Student t band covered 90.32 percent at 1,321.64 microvolts. Thus, scaling did not give smaller mean widths in this test. The rank guarantee requires exchangeable calibration and future waveforms, conditional on fit data. Four calibration pulses cannot give a finite 95 percent band under the tested nonrandom split rule. We did not establish exchangeability in patients. Reverse order gave 78.59 percent coverage. Simulations showed loss of coverage when test noise increased. The package contains source locks, pulse windows, saved bands, code, tables, and repeat instructions. All 3,477,760 saved voltage bins matched native source samples. Independent numerical checks passed. A repeat matched all 18 analysis files byte for byte. This is a secondary study of the same patients used in the cited pulse-window work.

split conformal prediction · prediction bands · stereoelectroencephalography · SEEG · waveform coverage · calibration · pulse response · OpenNeuro