Is the video tracking solve error a precise or approximate value? By this, I mean: is a 0.60 solve error always objectively better than a 0.61 solve error on the same video (with different markers)?
Tracking solve errors are in most matchmoving softwares calculated as root mean square of solved position difference over all solved trackers. The position difference of a tracker is the difference between tracked 2d position and position of solved 3d location in screen space. Ideally it should be the same, but usually is not. These differences are squared, summed and the square root of this product is the final error value. The unit of error value is horizontal width of pixel.
Now, due to the nature of this average error, different track configurations can give the same error value. A solve with all tracks having rms error of 0.6 and a solve with mostly very good and a few bad tracks can also have the same rms error. So no, you can’t deduce anything supermeaningful from rms error itself, but it gives you an overall idea if your solve is good or not. To dig deeper you have to check the rms errors of all trackers and fix or eliminate the ones with higher error. And then refine the solve.
Thanks for your detailed reply! I think I get how it works now.