Upon rereading it, it seems the real insurmountable challenge is interrupts! It always seems like at that level of detail with this stuff, it becomes a decision to go slower in order to simulate NES hardware accurately. But my question is: how much code depends on that accuracy, and how much can you compromise for speed? A further question I have is can you do a cross game analysis of the whole NES library and find common patterns, reused functions, that you decompile into a kind of high level conversion that would hopefully or gloss over the need for specific functions to have that low level accuracy.