something i haven’t seen mentioned that might possibly be a way to recognise it would be the bitrate. an AI model is limited by the dataset it was trained on and i guarantee you it wasn’t trained on lossless datasets. as someone who is an audiophile with my setup it’s pretty easy to spot and pick out a terrible mix from a good mix with an R2R DAC that doesn’t make music sound more pleasant than it actually is the way most sigma DACs do