"(A theorist) has much more to do than the generalization of observed regularities. Rather, he must construct a new system of concepts, some of which cannot at all, and others only partly, be traced back to observables; furthermore, he must devise a system of laws that contain these newly created concepts; and finally he must provide an interpretation of his system that can only partially be empirical, but that must still suffice for use as a theoretical system for observable processes. The concepts he uses may be quite abstract theoretical concepts; nevertheless, he is safe from the danger of slipping into speculative metaphysics as long as he can show that all these concepts can be used to make predictions."

Stegmüller (1989), p. 466 f