Chomsky kept sliding around as the evidence (largely published by his students or others with rather less credit given) necessitated. This sounds like flip-flopping; it also sounds like revising your hypothesis in the light of additional data.

It's like "before" or "earlier." Descriptively it's an adverb, I'd say, with a required implicit argument being "the time referred to in the speech act." That kind of thing is also messy.

Note that principles and parameters is not an exceptionless set of rules. A word may be lexically specified and escape from the general parameter setting. There are a lot of exceptions. It works best with simple propositions.

P&P fails to account for the facts (i.e., requires that a lot of work be done by the lexicon) in cases of language change. Languages can flip from head initial to head final, and it's never a neat change. This makes P&P a rather unpleasant framework.

Later there'd have been all kinds of required movement to let the parameter setting work well--then you need to specify feature checking presumably in spec position, whatever the status of spec is these days. (It's been a while since I checked out of generative syntax; last I was around we were working with Move alpha in a Minimalist framework, and that was a bit more than a decade back.) A lot of that checking was, again, required by lexical specification, IIRC.

Give your framework in a bit more detail maybe I can update it or give a better answer. PM me, since I don't often get back to check on my posts.