A local phase model reduced mean ripple overlap ranking loss in second sleep records from six patients under the source SEEG mapping

We tested whether local 0.5 to 2 Hz phase can rank supplied ripple overlap labels in another selected sleep record. We used public OpenNeuro ds003498 version 1.1.1. Each electrode pair had a separate model fit to the first selected record. Its second record supplied the test. Comparisons held electrode pair and 28 second time block fixed. Thirty pairs from six patients met the fixed rules. Mean test loss was 0.63314 nats, compared with 0.69315 nats for probability 0.5. This is an 8.66% loss reduction. The paired ranking score was 0.63136. Five patients had positive gains. The exact patient sign-flip p value was 0.0625. The actual score exceeded all 999 fixed block shift control scores. Distal phase gave a mean gain of -0.00341 nats. Population evidence remains uncertain. The SEEG scope assumes that released subject numbers match the depth-only patients in the source study. The target is a supplied detector label. The test does not show a neural cause, clinical benefit, or transfer to a new patient. The package preserves compact inputs, source hashes, code, fixed plans, results, figures, and reproduction instructions. A new public raw rebuild and a second full offline run passed the checks. The checksum check excludes the metadata folder added by AutoSci retrieval.

intracranial EEG · slow-wave phase · high-frequency oscillations · ripple overlap · separate-record validation · paired ranking · sleep · epilepsy