Feynman sums up science
What he doesn’t explain here, however, is that there are varying degrees of wrong. Some kinds of wrong mean you abandon the theory. Other kinds of wrong mean you adjust the theory. Once can look at the scrap heap of discarded theories for examples of the former — phlogiston, for example, predicted that mass had to decrease in the process of combustion, because it was a substance released during the process. After that was shown to be wrong, the theory was discarded. There was no way to fix the problems, since it wasn’t a matter of refinement.
An example of refining a theory is found in laser cooling. When it was first proposed, the mechanism was Doppler cooling, stemming from a simple model of photon recoil in a two-level system, and was experimentally confirmed. But eventually experimenters discovered conditions where their laser-cooled atoms were colder than the Doppler limit. Atoms are not two-level systems, and there are conditions in which their structure can be exploited to cool the atoms further: polarization gradients in the laser light, giving rise to “Sisyphus” cooling, in which the energy levels of the atoms are shifted depending on their position, and it is possible to have some atoms continuing to lose energy as they scatter photons, somewhat like the mythological Sisyphus, who was cursed to continually push a rock uphill. It was this discovery and explanation that won the 1997 Nobel prize for Bill Phillips, Steven Chu and Claude Cohen-Tannoudji.
There are more of these examples, with the hallmark being that the original theory is seen to no longer have universal applicability, but is still used under the conditions in which it applies. Relativity comes immediately to mind; we still use Newton’s law of gravitation, still use the classical equations for e.g. kinetic energy and do Galilean transforms when using them doesn’t introduce appreciable error.