I'm also not sure why you take my example of such a criterion (qualified properly and for that very reason with among other things) as the extent of all possible criteria and construct from that a 'very narrow definition'.
Because it simply gets you into some kind of madness. I have a book here on my nightstand, a B-58 pilot manual. As some kind of alternative fact, it has a few million multi-dimensional plots of its behavior included.
Would I turn this into some kind of test data, a B-58 simulation would be approximately a 2000 dimensional entity. 2000 dimensions, all of them independent. How realistic can a simulation of it be on a PC? For some combinations of the dimensions and flight conditions, I don't even have test data to compare, I would need to fly a few B-58 to destruction for this. And this does not even include structural simulations, I just assume the structure to fail as whole, like the manual does.
And the Space Shuttle, that you should know pretty well, there are far more dimensions. And far more configurations that add dimensions to it. How can you verify it to be "realistic".
For SSU, we are aware, that realistic is an approximation. That we can't simulate subsystem level with full accuracy, but need to use pre-calculated data and simplified models (Like how the data bus behaves). We can't create a FVM of the hydraulic circuits and solve it in real time, so that the flow changes include turbulences and local voids properly. We can't even create a simplified two dimensional flow model of it to solve in real time and expect it to be playable.
And don't get me started on simulating a SSME properly during startup. I am happy the way it is right now, because no number cruncher simulation exists right now that can reproduce the measured data of the SSME during startup. Simulating three-phase flows with the necessary accuracy and complex combustion processes at the same time is still impossible. Despite NASA sure trying to get one. The German DLR measured the Russian RD-0120 around 2000 very accurately on their test stand for one day having a simulation of it - but there is still no simulation model.
Even simulating a fuel dump properly under varying conditions of the Space shuttle would increase the CPU load a lot. CPU load, that we can use elsewhere better.