I parsed some logs of UAC controlling for very many variables (freshness of creature, room, attacker/defender conditions, etc.). For the character I tested, I have data on rabid squirrels and thyrils.
For rabid squirrels, the MM value varies by 1d37; that is, there are 37 unique MM values for rabid squirrels. The MM values I see are the integers in the interval [-36,0]. The frequency distribution is uniform. The random element of MM values does not seem to depend on stance -- changing stance adds or subtracts to a base MM value, but the MM value for any given attacker stance still varies by 1d37.
For thyrils, I see a similar story, except the MM value only varies by 1d2. The MM values I see are the integers in the interval [-3,-2].
My hypothesis is that a defender's stats (e.g. Evade/Block/Parry) affect its defensive MM randomly. I'm betting that someone with 100 ranks in Dodging would see a randomized advantage to MM like [-50,0], whereas someone with 10 ranks in Dodging might only see a randomized advantage to MM like [-10,0]. Obviously I made up those numbers.
A little more testing on players with controlled statistics would likely reveal a lot of information. Note that my tests only use one attacker, so statistics on an attacker may influence the random distribution. If we determine that MM only varies based on defender attributes, we could add MM values for each creature in the creature database, although those values would not be static. Rabid Squirrels might have a listed advantage of 1d37, whereas Thyrils might have a listed advantage of 1d2. Equipment likely also matters, though I have not tested this.
THIKKET 20:54, 22 August 2012 (EDT)