Bounded mean reference weights give finite posterior EEG ratio limits in all 88 ds004504 v1.0.9 records

We computed posterior EEG theta-to-alpha ratio limits for bounded changes in the 19-channel mean reference. We used all 88 processed ds004504 v1.0.9 records and 6,377 preserved windows. A positive spatial alpha term kept the denominator above zero. The method used scalar root search and global quadratic ball checks. It required no full-rank alpha contrast matrix. All 88 records gave finite limits, including 52 that failed the earlier eigenvalue safeguard. At Euclidean radius 0.01, the median upper-to-lower factor was 1.02098. Independent record weights could reverse 25 of 3,828 strict pairs. The fixed-cohort Alzheimer-control AUC ranged from 0.878352 to 0.881226. Each weight vector stayed fixed across both bands and all windows of a record. The package contains a proof, reusable code, matrices, source hashes, endpoint weights, results, and repeat instructions. A separate optimizer, constructed singular cases, random weights, three source replays, and a clean repeat checked the results. These are sensitivity limits for a fixed cohort. They do not give population or clinical performance limits.

electroencephalography · EEG · reference weights · theta-to-alpha ratio · sensitivity analysis · trust-region optimization · OpenNeuro ds004504