The thing is that, although the list used (the "Nova classification") is obviously somewhat arbitrary, the fact that everybody uses it makes research results comparable to each other.
Comparable how? If I compare one meaningless number to another meaningless number on the same scale then I can see which one is larger, but I won't learn anything scientifically valuable or practically useful.
Correlation != causation. Your correlation can identify a hugely important effect without pinpointing its mechanism. In this case: I think it's very likely that ultraprocessing has not-very-much to do with food health, but UPF foods tend to be hyperpalatable and low-satiety, which almost certainly does. The ultraprocessing isn't what's making the foot hyperpalatable or low-satiety (the macronutrient mix and sweetening is).
The previous comment was pointing out that there's an agreed-on definition of what "ultra-processing" means. There is. But there is no such agreement on mechanistic effects of ultra-processing.
