A preprint finds SCROLL-MT tracked changing forecast variance in Lorenz tests, while a Beijing PM2.5 series favored a simpler covariance model.