I can't go along with the GIGO analogy. If I can get 'predictively good' ratings on 2 or 3 teams, then I ought to be able to get 'predictively bad' or 'predictively average' on the rest. In any case, my ratings are as good (or as bad) as my rating system.
The game-time ratings represent my best appraisal of the current state. They are precise, but obviously there is error of estimate. They are certainly not 'bogus', but very carefully arrived at. But they are, of course, my estimates based upon whatever rules I choose to apply.
Regardless of how you do your ratings, I can't see going into the simulation using less than all the information you have available. Even if you decide a team is just a tick off the average (e.g., 51 instead of 50), you should still use that info. Why not? The simulator doesn't know or care.
