Unusual tournament Manthan 2011 showed the weaknesses of the current rating formulas. After the end of contest the winner of the competition tourist has expected rating change equal to "-1":)
This has arisen due to reduction of the calculated changes in such a way, that total sum is equal to 0. That is, Gennady was in some pros before such changes, but as the sum of all the changes turned out to be positive, then all the ratings were a little reduced. As a quick dirty hack in the round I turned off this feature, leading to a one-time inflation. Even a small mistake: I noticed that participants in the second division get too many points after a good performance on "Div. 2 Only" rounds.
In short, the formulas were a little changed so that the problem with round Manthan 2011 would not have arisen. On "Div. 2 Only" good place will cause not +200 points, but about +100 (of course, there are more dependent on the participant expected place). Actually, now get points a bit more complicated, although I do not think it will be very noticeable. On non-"Div.2 Only" rounds both divisions participant ratings are recalculated in common.
P.S. At the last round the division separation was not switched off, which led to the fact that members of the second division got less score than first division members with similar results. Of course, this defies logic. Thank you, ErzhanDS, for your observation. Changes in ratings after the last contest has been recalculated.
New formulas validated using past rounds and statistics from TopCoder. It seems they are good. I think that our formulas will continue to evolve a bit. As they say, c'est la vie.
Here is the link to the post about rating.