Maybe print some benchmark objects from thingiverse with the same print settings, and check what the results/tolerances are?