That is awesome and should be the minimum standard, but even just getting the entire infrastructure set up is something almost all scientists are not willing to

That is awesome and should be the minimum standard, but even just getting the entire infrastructure set up is something almost all scientists are not willing to do. In my experience, one should be happy if code is version controlled in a proper way. Too often, it isn't. In ML this might be a little different but in my field at least (electrical engineering) this is not the case at all.