I think that somewhere between needing a perfect model for any results to be valid and extrapolating forever from an imperfect model lies the useful truth. Key is "useful".

An analogy. Driving from here to there, from, say, home to your office.

I doubt if very often your are headed directly toward your office. Over any reasonable length of time you are getting closer, though.

Over any reasonable distance, the trip is worthwhile. You are getting closer. Travelling is worthwhile.

If you stay on any one heading too long, you'll miss your office by a long ways.

And if you wait until a road is built straight from home to your desk you will never get to work.

Arf