I feel as though I remember reading that it's OK to have advance spec definitions (ie to define specs whose s/def refers to a spec defined later in the same ns. Can anyone confirm or deny that?

yes (although there are some known cases where that does not yet work right)