I agree with you here-- science is about testability, and testability is about predictions. This does not mean that science is only about making predictions, because we do hope that some insight into the "nature of reality", as Robert Tulip puts it, will come from it-- we should not ignore the important philosophical component to the exercise, there's a reason science was born of philosophy. But the proper flow of knowledge is from what passes the tests implied by its predictions, into what can be called insight into reality-- never the other way around.

Now, it's true that Einstein explored the problem in the reverse direction-- he conceptualized about the nature of reality, and used that to design a predictive theory, but he knew about some of the kinds of predictions his theory would need to produce (a constant speed of light being the most important). Quantum mechanics also had some elements like that, it was a mixture between a mathematical formalism that made some kind of sense, and a set of preknown observations that would need to come out a certain way, even if counterintuitive. Both theories were quite counterintuitive, and neither was designed in the absence of surprising observational results. It was always those results that gave the theories their teeth-- they are why we accepted the theories. No scientific theories are based in logic, they are all based in predictive success.