Five-pulse consistency selection had an uncertain 5.1-point mean scaled forecast gain over nearest SEEG pairs in eight patient sessions

We tested SEEG pair selection with five calibration pulses in eight patient sessions. We compared pulse consistency with selection of the nearest recording pair. OpenNeuro ds004696 version 1.0.1 supplied data for 84 choice sites. The source products came from earlier AutoSci work. We tested scaled forecasts against five separate later pulses over 20–500 ms. Mean skills were 56.40% and 51.32%, respectively. The mean gain was 5.08 percentage points. Its descriptive 95% patient bootstrap interval was −1.82 to 11.94 points. The exact two-sided patient sign-flip test gave p = 0.21094. It assumes independent patients and sign symmetry under the null. This did not establish a gain or demonstrate equivalence. Restricted choices in the nearest pair’s region or shaft had mean gains below one point. The package preserves code, exact inputs, scores, source hashes, two figures, and reproduction instructions. Fresh retrieval checked 52 metadata files and rebuilt all patient 05 inputs. The test reused later data and source quality labels. It concerns within-session voltage. It does not establish neural origin, transfer between sessions, or useful interface control.

SEEG · electrical stimulation · sensor selection · pulse consistency · forecast skill · patient bootstrap