CFALD is deriving a binned, but full resolution luminance from each frame, prior to debayering. This preserves the SNR ruined by the debayering process. Also, because each sub is binned, the SNR is strongly increased. Stacking these trades back some SNR for some smoothing and perhaps some drizzle reconstruction.
Image 1: Using CFALD as Luminance
Image 2: Typical RGB Workflow