From another perspective, what you describe as mathematical truth is actually just a proof of a self-consistent mathematical model. Mathematical truth usually doesn't extend further than the page it's written on.
We can show that our mathematical models are internally self-consistent and match observed reality. These models describe the reality we observe, but not are inherently linked to the true nature of reality. Most of the time, new observations invalidate models we previously considered "true".
I think you're thinking of far more complexity than I am. I'm thinking of common applied math - accounting, weights and measures, carpentry. These things still have the reality correspondence problem, but truth isn't necessarily absolute. If I measure a cup of milk, I can safely treat it as a cup of milk regardless of the number of molecules in it. That's a kind of truth. Utility/ matching observed reality ought to be enough truth for anyone. As we get better and better at observing, the models will change, and that's good.
We can show that our mathematical models are internally self-consistent and match observed reality. These models describe the reality we observe, but not are inherently linked to the true nature of reality. Most of the time, new observations invalidate models we previously considered "true".