Firstly, unexpected results actually lower sigma more than expected ones, as we see from [url=http://research.microsoft.com/en-us/projects/trueskill/skilldia_5.jpg]top right here.[/url] For the relation between teammate sigma and mu change, [url=http://research.microsoft.com/en-us/projects/trueskill/equations.jpg]this[/url] clearly tells that lower teammate sigma = higher mu gain (the sigmas in c = team sum sigmas). A large amount of lost games acts as overwhelming evidence for the player being bad. And due to quitting, the player's performance in each game is incredibly consistent; the exact same every time, actually. And yes, a legit ten should level a new account up as fast as someone who quit to ten. If they had the same sigmas, that is. But thanks to the consistency of the quitter, he may have a lower sigma, and thanks to that also lower mu, than a standard legit player at the same level.