I wrote this about finding the optimal discrete character model structure. I hope this will be useful for empiricists. The main point is that relying only on the package default models is bad and can have nasty consequences or parameter estimates and ASR.

www.biorxiv.org/content/10.1...