You are doing the right thing in not just believe what the model shows. If you want a bg ref to work you need to understand what is not making work ideally.

Generally the problem is current density being too large so it no longer behaves as an ideal device -- usually from the IR drop in the base resistance. You should go to the model and zero out the non-ideal components (Rb, knee current, beta, etc). Also check the equations that your model uses. There are lots of temp dependencies in there: XTI, EG, temp dependence of EG, etc... make sure you hand calc includes these.