Adaptive covariance packets certified all fixed reader decisions in 1,348 SEEG windows from eight recordings for each of two invertible contact maps

Adaptive covariance packets certified all 1,348 later fixed reader decisions from eight public SEEG recordings for each of two invertible contact maps. Each reader used log power in three bands and was fixed from early trials. For the main map, packets used a mean of 84.69 bytes, with framing and scales. The mean within-person size reduction was 72.40% against float64 covariance and 47.05% against float32 covariance. No packet needed more than 16 bits per covariance value. Fixed 8 bit packets certified 1060 of 1,348 decisions. Without the check, 10 labels changed and 51 windows had invalid powers. The certificates kept 393 wrong task decisions. Mean person task accuracy remained 70.69%. The proof applies to the exact real reader defined by stored binary64 covariance and model values. It assumes matched contacts, an invertible map, included quantization cells, and positive power bounds. All tested native floating point labels also matched. This is an exploratory reanalysis of a reused cohort. The package has code, actual packets, exact source windows, figures, source checks, and replay instructions.

stereoelectroencephalography · covariance quantization · decision certification · signal compression · logistic reader