That’s a good question. Let me try to give you some idea.
When we are doing the calibration calculation, we are trying to find a solution for coordinates of all corner points such that all the measured belt lengths would converge at the same point (center of the router), for each of the measurements. In an ideal world, all those ends of those belts would meet at the same point.
In real world, however, we find the best solution, where all those points come the closest to each other. There’s always some distance between those ends, and we calculate the average of the average distances between those end points. Fitness is 1/that average distance. So if it’s >1, that means that that distance is roughly <1mm. If it’s 0.5, that’s ~2mm and so on.
However, it’s important to note that this is the best solution for the real measured belt lengths. If, say, all the measurements of belt lengths are off by some amount, we can sometimes still find a good solution with high fitness, however, the frame dimensions would be different from the real ones.
So in short, high fitness might not guarantee good results, if there is some systematic problem with measuring, but low fitness most certainly guarantees that the measurements are no good.