Going by the FAQ, a difference in kyu is 50 points. I just finished two games in quick succession with Aganju, and the rating shifts are bizarre.
In the first game, Aganju won. His rating was 13.4k and mine 12.4k, which translates to 50 points, so the expected change would be 18.29 points, or 0.37kyu. Except after the game, we only changed 0.1k, so his new rating was only 13.3k and mine 12.5k. If we go by the ranks when we started the game, this is even more off, since his was 14.7k and I was 96k.
Next game, finishing just moments later, his rating was 13.3k and mine 12.5k, or a difference of under 50 points. I won this one, and my expected gain should be under 13.71 points, or 0.27k. But each of our kyu ratings changed by 0.5 each, putting him at 13.8k and me at 12.0k. At the start of the game he was at 15.0k and I was at 9.8k.
What the heck? How can a weaker player beating a stronger player make for a 0.1k shift, and a stronger player beating a weaker player make for a 0.5k shift?
The only way this makes sense is if it's going by the rating at the start of the game, and it's doing the adjustment *backwards*. Putting us at 5k difference at the start of the game is 250 points, or an expected difference of 25 points if the *weaker* player wins, which is half a kyu. Since kyu counts down, is the software messed up?
Besides which, shouldn't handicap games count as EVEN games? Isn't that the point of the handicap?
What I would have expected from the software is that the handicap is counted in, so with a starting difference of 5k - 5handicap = 0, the first game would be consider within 'equals' (which is the idea of handicap).
Maybe it uses the rating at the end of the game (much easier to get), so we were 1k off - 5 handicap = -4k. So I should have won easily (4k ~ 20:1), and therefore I got only 0.1k for it. Second game, the diff was .8k - 6handicap = 5.2, so I should have won even easier, like 50:1, and my loss results in -0.5k.
I didn't do the exact math, but it might work out, at least it is nearer to what happened than any other theory yet.
Not that I think it is good or correct if it is so. Just saying this theory might explain it.
I tried to come up with a better solution (if only for the record), but it is not that simple...
Hmm, that makes some sense. But I dunno, I checked another game where I started the game with a ~250 point lead and gave 5 stones, but when the game was over I only had a ~125 point lead. I lost the game, and the rating adjustment made sense for it based on my rating at the end of the game, whereas with your idea, it should have been based on my opponent starting the game as if he had a 125 lead, if I follow your logic?
Little Golem