The test of a good designer, is how they fix their failures.
Getting a wrong answer from an initial bandgap design,
is as common as seeing detail-inaccuracies in a SPICE model.
Which are to be expected even more in "digital" foundry
PDKs where nobody ever thought the parasitic PNP would be
a valued analog element, for example.
Mismatch will not prevent you from finding a working
example, only from finding many the same. You have to
understand how to drive down or drive out mismatch.
But perhaps the issues are really asking you to pull your own
device data and verify that you're not simply the victim
of half-truths and outright lies, in the model files. And then
the fun of redoing them right, where it counts.