Five early pulses selected map ranks that reduced mean centered 20–500 ms SEEG error versus rank 1 in eight STReEF patients

We tested whether five early pulses could choose the size of a SEEG response map for five later pulses. The source was OpenNeuro STReEF ds005448 version 1.0.0. The sample had eight patients, 96 conditions, and 6,803 recording-pair rows across conditions. The later group reversed the recorded contact order. Early leave-one-out validation chose among ranks 0, 1, 2, 4, 8, 16, and full. The map was then refitted on all five early pulses. No later voltage entered choice or fit. In the strict 20–500 ms window, selected ranks reduced later centered-voltage squared error by an equal patient mean of 4.65% versus fixed rank 1. The descriptive 95% patient bootstrap interval was 2.17% to 7.00%. Seven patients improved. The exact patient sign-flip reference was p = 0.015625 under an independent, sign-symmetric null. The rule chose zero in 46 conditions. Its gain versus zero was uncertain. The before-pulse control also improved versus rank 1. Thus lower error does not identify a neural source or prove effective brain writing. Time and contact order are confounded. Eligibility uses annotations for both groups. This is an exploratory study on a reused sample. The package preserves code, source metadata, pinned retrieval records, fold terms, native examples, tables, and four figures. All 960 fresh voltage ranges matched their source hashes. All 2,688 later candidate errors matched the parent analysis. All five result files reproduced exactly.

SEEG · intracranial EEG · response map · rank selection · calibration · voltage prediction · singular value decomposition · brain-computer interfaces