Hi, I was having this question, since it is really strange. In this blog post it says is ELO-based.
I think that the post is old or wrong. Please note that this is not a complain or anything just an observation, but with the statistics at hand, score distribution does not follow an ELO-like distribution. The score gained in the first 5 spots is just big compared to all the other scores.
Maybe there is a point bonus for the first spots?how much? And a point bonus for each problem?
edit: Although ACRush’s case is notorious, because 5 winning streaks and now he has almost twice the score as the second person, what really brought my attention is the last page in the rankings
Oh, I see, and I think I see the problem. I don’t have N, but let’s take it the most accepted problem.
The most accepted problem in DECEMBER 2010 (ASTRGAME) had 91 Accepted. Meanwhile DECEMBER 2012 (GRANAMA) has 1830.
I think here lies the problem. Maybe the bonus factor has too much weight nowadays compared to the regular factor. I think there should be some adjustments. Do you know if it was ever adjusted?
But if you see the first comment, rudradev commented to take k as a decreasing function , which I assume they must have implemented.But I agree , still there has to be changes made , otherwise top few contestants will become unreachable
My opinion is the long contest rating has not got too much information.(I think my rank is overestimated too…) I feel when I check somebody profiles and I see the rating graph, I couldn’t read too much infromation from this…Maybe I feel this because the y axis is changing every month. (I mean I feel every month (in the top 50-100… everybody get +200-500 points at least, so the y rating range is moving…) For example if I compare this with the topcoder there are not moving the y axis (significantly), and when I check somebody profile, and rating graph I feel it contains lot of information. So I think in the user profile instead of the long contest rating graph, the codechef should show the performance os rank graph (e.g. x -axis is the month/y -axis is the place in that month or the long contest rank).
I agree with you and I think it would be nice if profiles were more informative (like showing the position in the contest). Still, a more representative rating would also be nice.
No only top 50-100 get +200-500 but rarely people get negative score. It weird that people in top 20 do only 6 problems and their score don’t decrease.