Equal Fp1 noise powers change posterior theta-to-alpha ranks after mean reference in 88 ds004504 v1.0.9 records under zero cross-power

We tested equal noise band powers at Fp1 in 88 processed ds004504 v1.0.9 EEG records. The model assumes zero band cross-power with all source channels and keeps source artifact processing fixed. The 19-channel mean reference adds 1/361 of that noise power to each posterior signal. At 20 microvolts band RMS, noise with constant 4–13 Hz density reverses 54 of 3,828 person pairs. Spearman correlation is 0.99792. Alzheimer-control AUC changes from 0.88027 to 0.88123. Each pair has at most one positive crossing under this model. Exact rational crossings give the full paths. The 10 percent reversal threshold has an infimum of 121.78839 microvolts. The strict target requires power just above that crossing. Theta-only and alpha-only thresholds are 39.68 and 46.31 microvolts. The first-10-window constant-density path never reaches 10 percent. Bipolar P3-O1 and P4-O2 signals remain unchanged. These are conditional finite-cohort sensitivity results. They do not measure actual noise or clinical test accuracy. All 31 repeated numerical files matched their stored hashes. Three hash-checked source files passed matrix and independent spectral checks within numerical tolerance. The package preserves code, matrices, exact crossings, proof, source hashes, checks, and repeat instructions.

EEG · mean reference · noise covariance · theta-to-alpha ratio · rank sensitivity · spectral power