I think it's best to stick with Popper's definition of a good scientific theory: it needs to make falsifiable predictions and have them not falsified by future experimental data.
General Relativity, e.g., passed this with flying colours (perihelion of Mercury,  atomic clocks on airplanes etc). From what I've heard, string theory, e.g., doesn't.

An interesting example of where it gets tricky is the neutrino. Iirc it was postulated to explain an energy deficit in a nuclear reaction. But then people were able to use it to predict other experimental results. Is dark matter like that? Cosmology is really tough from this point of view.