Am I correct that DCP-o-Matic only does a nearest-neighbour chroma upsampling for chroma subsampled sources? Particularly deep reds exhibit some blockiness in the final DCP when such files are imported directly to DoM, and of course, such artifacts are more visible on the big screen.
I find that I'm actually getting better results if I upscale any 4:2:0 (and to a much lesser extent, 4:2:2) files to ProRes 4444 using something like ffmpeg's default bicubic upsampling before importing to DoM.
Pretty much all software video players (such as VLC) seem to be using at least bilinear or bicubic chroma upsampling, and it feels a bit counter-intuitive to me that their playback quality should actually look less blocky compared to DoM's DCP output...
I can provide some samples later if necessary!