Technically speaking it doesn't. Theory is the highest degree of certainty you can give something in science. People seem to conflate "hypothesis" with "theory". Hypothesis is what you start with, theory is what you find out from that hypothesis.
Usually, you start with an hypothesis about some aspect of some theory, design an experiment that is informed by theory and interpret the results according to theory. The Scientific Method version of science is so outdated that even the AAAS and similar groups have been trying to change how it is taught and books by e.g., James B. Conant in the 40s and 50s were already trying to address the popular misconceptions while philosophers of science and scientists had found the heart of The Scientific Method had failed fantastically and were struggling to replace the naive positivism of the 19th century.
Most scientific research doesn't refer to theories except insofar as the theory concerned is more or less the field of research of which the study seeks to contribute to.
It's always possible to further refine it
Not always. The most spectacular failure of the "Hypothesis -> Experiment -> Theory" model (and more complicated versions) was modern physics. The debate over the nature of light was nothing new, Newton's corpuscular theory having been refuted first by Young's experiment and then by Maxwell and other work in classical electrodynamics. Light was a wave. That it was part of the electromagnetic spectrum was a further refinement of the kind I think you refer to. Einstein's solution to the photoelectric effect (that light is composed of "particles" of energy he called quanta) was not a refinement, but a contradiction. Physics of the time held that there was matter, and there were things like light and sound that propagated through matter (more precisely, through a medium) but did not have any existence apart from the effects on the medium. More importantly, particles were fundamentally different from waves, and something was either a wave or a particle. So on the one hand we had all this evidence that light was a wave, which meant it couldn't be a particle. On the other, we had experiments that could only be explained if light was a particle.
It turned out that the problem wasn't with experiments, and the solution wasn't refinements. It was recognizing that almost the entirety of physics was at its core WRONG. The framework itself had dictated that light prove to be a particle or wave and thus experiments sought to determine which. The only reason the physics community didn't spend decades split into "particle" vs. "wave" camp was the simplicity of the experiments and the obvious impossibility of reconciling them with physics (what we now call "classical physics").
The biomedical model of mental illness is another example in which it was not findings and research which changed the dominant theory but various other factors, and after 30+ years of research to demonstrate the basic postulates set down by the founders of diagnostic, biomedical psychiatry (particularly with the publication of the DSM III) we've found more counter-evidence than evidence. Other examples could be easily marshaled.
We know more about evolution than we do gravity. Evolution at least works on every level we've observed.
Most of evolutionary psychology is bunk. It's part of evolutionary theory. Even if we limit ourselves to evolutionary biology, though, a central component is under increasing debate, namely “fitness” :
"the fact is that there is a major problem in the foundations of evolutionary theory which remains unsolved, and which continues to give life to the debate. The definition of fitness remains in dispute, and the role of appeals to fitness in biologists’ explanations is a mystery."
Mills, S. K., & Beatty, J. H. (2006). The propensity interpretation of fitness. In E. Sober (Ed.)
Conceptual Issues in Evolutionary Biology (3rd Ed.) MIT Press.
"Out of all this, the principle of selection...has received major attention and the discussions in this area have led to many treatments abundant in the scientific literature. Beyond no doubts, this is for several reasons, such as the weaknesses in the definition of ‘‘who’’ is carrying out this selection, or exactly ‘‘what’’ causes it to come by? What is it exactly that is selected for?
And if fitness, then how is this fitness defined? Debates in this area have not ended and are not likely to be settled for years..."
Nielsen, S. N., & Emmeche, C. (2013). Ontic Openness as Key Factor in the Evolution of Biological Systems. In
Evolutionary Biology: Exobiology and Evolutionary Mechanisms (pp. 21-36). Springer.
and so on. Evolutionary theory is so broad that in one sense were there not serious debates about numerous issues we’d have reason for very serious concern. Evolutionary theory is not only the framework wherein which work in biology, astrobiology, chemistry, complexity & computational sciences, etc. is carried out but the basis for fields like evolutionary psychology and large parts of systems sciences.
Gravity, however, is much, much more constrained. In one sense it is far better understood than evolutionary theory, because it is so much smaller and the problems with it so well identified. In another sense, it is far less understood because in general relativity it really isn’t a “thing” at all (not a field, nor a force, but a series of equations modelling the way spacetime curvature “tells” matter how to move). Also, while Newtonian gravity continues to be useful across the sciences, it is simply wrong, and we know this. Evolutionary theory has grown, has become more and more refined, has spawned sub-disciplines in which hypotheses are tested and models created, but it has never received anything like the challenge Newtonian gravity did: all the changes were more or less the kind of “refinements” you refer to.
Gravity kinda' peters out past a threshold.
Newtonian gravity fails as we approach two thresholds, but it fails to be useful. It is always and everywhere wrong, and no successful unification between gravitation in GR and QM exists.