Gamma recovery did not improve mean SOZ rank loss in 29 ds004100 SEEG patients under a fixed linear model

We tested whether gamma recovery adds information about clinical seizure onset zone marks. The source was public SEEG dataset ds004100, version 1.1.3. We reused power data and the fixed cohort from AutoSci b7125622-0ff4-4f1d-b2b4-f90ab5d8e48a. The test used two records per patient and 472 adjacent bipolar pairs on 62 shafts. Each of 29 patients was excluded in turn from fitting. A fixed linear rank model used mean gamma suppression across 10–30 and 30–50 seconds after marked seizure end. The added feature was late power minus early power. Mean rank loss changed from 0.4318 to 0.4323. The loss gain was -0.0005. Its 95% patient interval with model refitting was -0.0325 to 0.0312. Mean AUC changed from 0.7858 to 0.7837. Recovery reduced loss in 18 patients, but larger losses in other patients offset these gains. All planned checks are reported. The main test gave no observed mean gain. It does not establish model equivalence or exclude a useful recovery effect. New checks matched 192 source windows and 197 metadata files. A fresh offline run reproduced all 15 main result files exactly. The package preserves code, power data, models, results, raw fixtures, source pins, and repeat instructions. The result concerns a reused clinical cohort and fixed linear models. It does not establish prospective clinical or surgical benefit.

seizure onset zone · stereoelectroencephalography · gamma power · recovery · rank loss · AUC · epilepsy · leave-one-patient-out