I agree that if it is not achievable, it’s not interesting to pursue it. But if it is, then it may be worth doing e.g. better quality (even if more complex) mappings at integration points, in order to get better quality (i.e. more computable) data into the EHR. Just copying in crap creates a garbage dump, which might be ok for some uses, but it really doesn’t advance us much.
Being ‘right’ in these cases doesn’t equate to some idealistic notion of perfection, it’s simply whether the result will be computable or not. If it’s not, the data are less useful (which is not to say useless, of course).
This is why people like Barry Smith, Stefan Schulz etc work so hard on this stuff - it’s about enabling machines to help a lot more than they do today.