It is similar to the difference between a designer and a mechanical engineer working on a merry-go-round. The designer (or physicist) is concerned ultimately with some observable phenomena and general behavior of the entire system. The mechanical engineer (or mathematician) is concerned with gaining complete understanding of the components, their precise limitations, and how they work together.
The mechanical engineer may spend a long time studying the threading of one of the bolts. After that study, they are incredibly confident of how it will behave in certain circumstances. Analysis of all such components being used in the merry-go-round could then be used in principle to rigorously analyze the behavior of the entire system, although it may be computationally infeasible.
The designer, on the other hand, may be asking more broad-scoped questions like, "Will the kids riding it be flung off at this speed?". You can answer such questions without modeling the bolt threading, maybe just simplify the whole system to a spinning disk and do a rough calculation. Maybe later they wonder if the animals need to be attached at both the bottom and top, or if they could be supported just from below. They may then move to a more refined model with spherical animals, but again not on the level of needing to know about the threading of the bolts. Part of their expertise is in making unfaithful simplifications of the problem without changing the answer too much.
What's important to stress here is that actually neither of these activities are necessarily scientific in themselves. Such analyses can be done even if there don't exist any merry-go-rounds at all. Many people do not consider mathematics to be a science at all (although it may be used as part of a scientific theory). But even more unexpectedly much of theoretical physics also is not science. For example, there was an immense effort to describe orbits of planets (including "retrograde motion") using nested epicycles. It turns out that by general mathematical principles (later formalized by Fourier) any motion observed could have been modeled using epicycles. So there is actually zero scientific content in the statement "planets move according to nested epicycles": it does not make any testable (falsifiable) predictions about the motions of the planets. If the prediction with two nested epicycles is wrong, you can add a third, or a fourth, or a fifth, etc. to accommodate any empirical challenge to that claim.
There is somewhat of an epistemic difference, though. A physicist might gain confidence in a certain novel inferential technique by creating predictions that rely upon it, and experimentally validating those predictions. This allows them to combine deductive and inductive knowledge in a somewhat unique way (compared to, say, an epidemiologist). What is unique about it is that physicists will gain confidence in inferences that are very high level (nested deep in the theory itself, not directly connected to the phenomenon being observed). Think for example about how separated CPT (Charge-Parity-Time) symmetry is, as a principle, from specific predictions about a particular physical system. This makes it difficult to say what the ontology of a physical theory is: are the epicycles "real" / "physical", or just theoretical artifact? Is the "wavefunction" real or just a mathematical convenience? Is the universe "really just a hologram"?
It is probably more accurate to view theoretical physics as a non-scientific quasi-mathematical discipline which bases its rules of inference on a combination of analytical knowledge and empirical knowledge, rather than as a scientific field in itself. After all, most of the "models" considered in theoretical physics are not even intended to model the real world.
In mathematics, we typically permit only analytical rules of inference, and again it is not really scientific because it makes no predictions. The ontology of mathematics incredibly liberal, making even the wildest of mereologists blush. The way mathematics is is used in science is more similar to the way English is used in science: it is used to express a scientific theory, but is not science in itself.