A t-test of unique pulse pairs exceeded 5 percent rejection under independent pulse signs in SEEG groups of ten pulses from five patients

We tested whether a one-sided t-test of unique pulse pairs kept a 5 percent rejection rate under independent whole-pulse signs. We used public SEEG windows from five patients in OpenNeuro ds004457 version 1.0.2. We kept 1,080 links, each with ten pulses, and computed all 1,024 sign assignments through 512 equal-weight classes. The equal patient mean rejection rate was 7.248 percent before a pulse and 7.419 percent after it. Every patient mean exceeded 5 percent. An exact sign-rank test gave 4.883 percent under the same sign model. Ten equal pulse shapes gave an exact counterexample with 10.9375 percent t-test rejection. A planned duration-search check gave 14.286 and 13.482 percent rejection; the exact test repeated selection and kept its bound. An added exploratory check held the available interval fixed and found higher rejection with duration selection. These results describe a constructed null model and selected signal shapes. They do not estimate clinical false connection rates or prove that recorded noise follows the sign model. The package includes a tested Python tool, proofs, all compact data, source hashes, complete primary sign statistics, and reproduction instructions. A fresh raw decode matched every saved bin. A second run reproduced all ten numeric outputs byte for byte.

SEEG · statistical calibration · whole-pulse signs · exact test · unique pulse pairs · Canonical Response Parameterization