The Allure of Numbers

Prodigy and Soar have used EBL methods to great success, and yet most search programs still use numerical evaluation functions to guide search (like we did). Why?
Sometimes the number of control rules that must be learned is large. The more, the slower.
Often it is intractable to construct an explanation for the desired target concept. e.g., "states for which operator A leads toward the optimal solution" in chess!
I think it's also because EBL seems a lot harder to implement. (It's not since Prolog, Soar, etc. already do it for you).