The raw Glicko-2 scores are better than the kyu/dan conversion for players that predominantly play non-handicap games, which is the majority. There already is a switch in the user page that allows one to see the rating matrix as either. Make this work with the rating shown next to usernames and the rating-limiter when creating a game.
But… why?
3 Likes
Because:
- Glicko-2 scores are tightly related to the chance of winning/losing.
- The kyu/dan system is based on a non-lineal transform of the Glicko-2 score. Non-linear means that is uneven.
- The conversion to the kyu/dan system is subject to change as experimental evidence changes.
In what way will these facts improve the experience of players?
2 Likes
By providing a meaningful rating, comparable to other games (Elo is standard and Glicko-2 ratings are on the same scale) and closely related to win probability.
One kyu/dan difference represents 14 points, this is very concrete even if you don’t play handicap games. It’s easier to think in terms of point loss than in terms of winning probability.
2 Likes
That idea didn’t get much appraisal in the kung-fu federation.
2 Likes
Ok I see your point, although I don’t see a demand (apart from you of course).
If ranks are equal, then win probability is around 50%. That is nice to know.
If ranks aren’t equal, then players know that the lower ranked player has a smaller chance of winning. I don’t think many players want to know that their win probability is for example 9%, with a std.deviation of 5%, rather than “very small”. But I could be wrong.
3 Likes
That’s just a framing issue. If we assume win likelihood is ground truth, then yes, kyu-dan is nonlinear relative to that. But while that is ground truth in chess, it is not in Go: handicap stones are. And it is not kyu-dan that is nonlinear in light of that, but rather win likelihood
5 Likes
It is for non-handicap Go.
I could just as reasonably say that ground truth for chess is only win likelihood in even games, ground truth for handicap chess games is actually whom who can give an f-pawn handicap to. But that’s not what chess uses even in handicap games, just like win likelihood is not what Go uses, even in even games
1 Like
False. All current rating systems (in Go or otherwise) compute ratings based on win rates.
You are describing the technical means by which we arrive at kyu-dan rankings. The ground truth by which we judge whether or not that is successful, is still the kyu-dan rankings themselves
2 Likes