Even such a domain specific definition is borderline useless because there is no way to constrain the "utility" which rational agents are supposed to maximize -- which makes the concept of a rational agent tautological, and not at all predictive. ("People behave that way because they behave that way". I might encode my observations in a utility function but that doesn't suddenly help me generalize to other situations)

The other thing is that the limited situations which psychologists/economists measure people in are very artificial, and there's no reason for people to ace the experimenter's metric.

Responding to grandparent: It's not just that project are efficient and not quite correct -- they're probably more correct and scientists have no way to tell.

"If you judge a fish by it's ability to climb trees, it will live it's whole life thinking it is stupid."