Is the "poll" officially open right now in this thread?
If so, I'd have to vote for individual game behaviour modified appropriately by situational behaviour, as both are equally essential. The total stats should follow real-life stats reasonably well, if CPU's starters/bullpen usage resembles real-life managers closer at all. (Note the British spelling of behaviour for emphasis.....)
For example, one of the reasons why CGs are high in game I think is that CPU manager tends to leave his starter in regardless of pitch count, when the team is up by 4 runs (or 5... am not exactly sure of the exact threshold) and the starter is allowing few runs... so some starters end up with pitch count of 140 or even 150 in those blow out games. Quite often I see those starters left in too long get exhausted (zero energy), start walking or getting hit hard in the 9th inning, only to end up creating a save situation and eventually relieved anyways (8.1 & 8.2 innings pitched....). This is from CPU vs. CPU games, and not simmed games though (although I hope manager AI isn't too different in those game modes...). There may be other criteria being used by CPU to decide whether to pull starter or not, but these days pitch count is rather strictly enforced (except in special circumstances... no hitters, etc.), which CPU manager should consider more...
Thank you Brian for great attention to detail as always! I hope the "two weeks" timetable is for a patch and not for the next year's game though...