I believe the true answer to your question is that our observation of space fits a mathematical model for a 3-dimensional geometry. My wording is backward compared to historical development, which is why it's hard for people to decouple these things. Humans learn about the degrees of freedom in the space around them as an infant, and possibly even sooner. The fact remains, however, that concepts like $\sqrt{2}$ have a consistent formalization within the field of real numbers and a corresponding geometric interpretation (like the diagonal of a square). There is room for philosophy about what that interpretation means, but it is not an impediment to physics either way.
We've had several occasions to propose revisions to our concept of space and time. General relativity's conclusions are accepted beyond all reasonable doubt. Our mathematical models expanded in correspondence with our beliefs about the universe. String theory also offers revisions that involve much more difficult math, but these are all knowable and calculable. This doesn't change anything about the basic fact that we fit a mathematical system to our universe.
The only useful thought experiment I can imagine within this philosophy is if the mathematics of geometry would be discovered by mathematicians living in a world without our type of geometry. That begs the question of whether we have discovered some other mathematical system that could function as an alternative to geometry as we know it.
We have formal systems to work with here. Humans have developed many formal systems. With computers, we have been able to study the nature of a wide variety of formal systems due to the low cost of running experiments. All of what we do with computers fits into a certain class of mathematical system, with is Turing computing. Within these systems we are able to model real physical systems, which is reasonable, because as I said, geometry itself is a mathematical system, and other physical laws are added onto it as needed.
Turing computing, however, still can't emulate geometry in finite time. This is a deep philosophical question. For classical and simple physics, I can plainly state that the Galilean invariance rules out the most logical approaches to a cellular automaton at the root of space. You can go further into the details of Lorentz invariance to find even stronger statements, but it's not important. Even aether theories from the turn of the 18th-19th centuries reflected a logical inroad to pixelization of space. If an aether theory had been true, then a finite Turing machine could have validly been at the root of space.
Of course, those theories are not true. Reality has been adamant about preserving the laws of physics in all reference frames, which spells doom for theories predicting that space only approximately fits the mathematical system, and the result of some finite number of constituents. Our finite computing machines have a number of quirky errors in simulating reality, such as energy drift. In the real universe, truly conserved quantities like energy appear to be perfectly conserved. This relates to Noether's theorem. Basically, we observe that:
- Conserved quantities are conserved perfectly
- Relativity between reference frames holes perfectly
These two are two sides to the same coin. So when we observe that space and time are perfectly consistent in our universe, we can say that it's related to the fact that net energy and net momentum are perfectly consistent scalar and vector quantities.
In short, physics is about writing down the rules of a mathematical system that fits the universe. You can say that geometry (whichever geometry this is) can also be defined with a set of rules. Following those rules is impossibly difficult with finite formal machines. So space could be "false", meaning that it's really just an apparent system that exists on top of a more fundamental system underneath, but that more fundamental system would not be any more computable.