Rank limits gave no site labels at a 90 percent coverage target across five reading times in five offline 6 mA SEEG sessions

We tested whether candidate ranks gave more correct stimulation-site labels than score margins. We used public SEEG data from OpenNeuro ds004457 version 1.0.2. The same reader used two fit and two calibration pulses per site. Both rules used a nominal coverage target of 90 percent across five reading times from 100 to 500 ms. The primary test used 138 sites and 552 pulses from five people at 6 mA. Ranks gave no labels. Margins gave 17 correct and 3 wrong labels. The equal-person change in correct labels was -2.78 percentage points. A set-size bound proved that rank labels were impossible for two people under their fitted limits. The wider cohort also gave no rank labels at this target. All voltage bins rebuilt exactly from 110 hash-checked public source files. The package preserves code, compact data, model states, pulse decisions, figures, and a raw retrieval recipe. This is an exploratory offline comparison in a reused epilepsy cohort. It does not test a live controller or a new session.

SEEG · brain-computer interface · conformal prediction · candidate ranking · selective classification · abstention · intracranial EEG