Transport of ridge penalties kept MIC fits and frozen-filter imaginary coherency equal across two equivalent zero-sum bases in 88 ds004504 EEG records

This study extends an earlier metric-ridge audit of a global EEG score to two-group maximized imaginary coherency (MIC). It tests fitted physical filters and their ordinary imaginary coherency on separate windows in all 88 processed ds004504 version 1.0.9 records. Each group had four full zero-sum contrast dimensions. At 10 Hz and ridge fraction 0.1, a fresh trace-scaled identity penalty in each equivalent basis changed all 88 fit scores. The mean absolute change was 0.027010 (pointwise 95% person-bootstrap interval 0.023389 to 0.030802); 232 of 3,828 person pairs reversed order. Frozen-filter test scores also changed in all records. Transport of the same physical penalty kept fit scores equal within 1.17e-15 and test scores within 1.85e-15. A proof gives the positive-definite power and invertible real basis conditions; filter equivalence also requires a separate leading singular value, which all tested fits had. All nine planned settings are reported. Independent calculations checked 2,376 fitted scores. A full repeat matched eight result files byte for byte. Fresh retrieval of all 88 complete source files reproduced 6,377 windows and 1,817,445 Fourier values exactly. The work concerns a fixed measurement rule; it does not show neural accuracy or disease prediction.

EEG · imaginary coherency · ridge regularization · basis invariance · sensor-space contrasts · spectral connectivity · reproducibility