The problem is simply the lack of validation -- I detailed it a bit below in my top-level comment.

The only exploits that are inherently applicable in the case of a simultaneous deterministic simulation model are those of the information-revealing class (e.g, "maphack"). This is where a standard AH system comes in.