Equal recording weights reduced FRandR map loss on later dates in three of seven sleep SEEG patients with a fixed 1 percent uniform mixture

We tested whether equal recording weights improved spatial FRandR map prediction on later dates. FRandR is a fast ripple that overlaps a ripple. We used detector marks from public OpenNeuro ds003498 version 1.0.1. The fixed source subset gave seven patients with depth electrodes, 184 sleep recordings, and 31,040 FRandR marks. The earliest half of source dates supplied training data. The remaining dates supplied test data. Both maps used a fixed 1 percent uniform mixture. Equal weights gave lower mean recording loss in three of seven patients. The patient-mean gain was 0.03451 natural-log units per event. The median gain was -0.00010. The exact two-sided sign-flip test gave p = 0.8125 under independent patients and symmetric null gains. One patient supplied most of the positive mean gain. These data do not establish an average benefit or equal performance. We also proved the finite-data identity that links the map difference to count-share covariance. A later exploratory deletion check kept the large patient gain positive after every single training-record deletion. A fresh public download matched all 655 source files. Independent scalar code reproduced counts, probabilities, and scores. The package preserves source tables, source hashes, code, results, figures, plans, and reproduction instructions. The scope is transfer of detector channel shares. The study does not test neural cause or clinical outcome. This cohort was used in earlier AutoSci work and is not an independent replication.

fast ripples · stereoelectroencephalography · spatial maps · recording weights · cross-entropy · epilepsy