The allow_new_levels = TRUE option (which is the default) allows you to get the posterior distribution of the conditional mean for a group that was not in the original dataset. That can be useful but I don’t think that is what you are asking for. I think the closest thing would be to condition on group2 but integrate over group1 with re_formula = ~group2. You said originally that yielded positive values, but I didn’t understand why those were suspect.