I'm not speaking about science as a whole or a scientific program, but a scientific claim. CERN is certainly not falsifiable, but it produces predictions which (often) turn out to be true. It does so by devising falsifiable claims and then testing those claims.
In other words, the method to create interesting predictions which turn out to be true is to create interesting predictions, then test those predictions, and update your understanding of the world based on them. Once your world-model is good enough, your predictions will often be true. And, perhaps, eventually your predictions will be so often true that they become uninteresting, so you must move on to other questions.
Fair enough, I think Kuhn was referring to things like the world-models and you're referring to finding out if the predictions of the model matches reality.
The heliocentric model of the solar system made less accurate predictions than the geocentric model for years, because the geocentric model was mature and had had lots of tweaks applied to it. In that time, you could have asked the heliocentric model to make a prediction, and shown that it was wrong compared to the geocentric model. You would have been wrong to conclude that heliocentrism was wrong though, it just hadn't matured as a theory enough yet.
In other words, the method to create interesting predictions which turn out to be true is to create interesting predictions, then test those predictions, and update your understanding of the world based on them. Once your world-model is good enough, your predictions will often be true. And, perhaps, eventually your predictions will be so often true that they become uninteresting, so you must move on to other questions.