I can’t see anything controversial about that. 
3 Likes
Someone who has 10 years experience playing in an orchestra, after having concluded a conservatoire education, is without a doubt a professional musician
conservatoire: a music academy
Thanks for increasing my vocabulary :D It’s obviously related to conservatory, but I’m not sure in what exact sense…
1 Like
I think they’re synonyms, but “conservatory” also has the meaning of an archive, I believe. And I didn’t want to call it “music school”.
4 Likes
From -30 dan to +10 dan 
And 0 dan of course is equivalent to AGA/EGF 1 kyu.
2 Likes
The OnlyDans™ system is of course not that dissimilar from, say, the Elo ratings in chess, just rescaled so that difference in rating equals difference in handicaps. So it’s not that groundbreaking or controversial (except for calling them dan ranks 
).
Some may think a rating like 17.45 is more ugly than 1745, but I think the extra work of writing out the decimal point is made up for by the work saved in all the situations where you want less precision. So for instance we can talk about “players rated above 17”, instead of “players rated above 1700”.
For go in particular this system is great because handicaps are close to linearly additive for small rank differences. One interesting consequences of following this idea through is that there would be much fewer ranks on smaller boards. This is true already for Elo ratings (shorter games have narrower Elo widths), but the effect is more extreme for handicap stones (one handicap on 9x9 is commonly approximated as four handicaps on 19x19).
Also related is the fact that handicap stones become worth more the stronger you get, so this system would have fewer ranks on top even on 19x19. But this is where it’s nice to be able to add extra precision at will. Even if most pros theoretically ended up between the same integer ranks, we can distinguish them using their fractional ranks.
All in all, this could be a really nice system. (although a traditional system based on winrates in even games certainly has advantages too, perhaps most notably being applicable to all games, and thus familiar to everyone).
I think a really interesting point to discuss is where would be the most natural place to anchor such a system. Would it be practically possible to make the 0-point completely random play? Then there are theoretically some negative ranks, but all humans should be positive. The question is if low ranks are well-behaved enough to make that system somewhat stable.
2 Likes
All “reasonable” humans should be positive, but I guess reasonable is a bit subjective.
Is zero Elo (or any other point) anchored to anything in chess? The system just floats, right?
I think the very low ranks are a tricky issue for any rating system. Very weak players are often novices that are still learning and improving very quickly, so maybe their rank is not stable for long enough to measure reliably.
1 Like
Yeah I’m not aware of any human rating pool which is anchored at random play, it’s probably a bad idea in practice (despite its aesthetic appeal).
A compromise could be to decide on some rank more in the middle as the “real” anchor (could use some bot to avoid having to pick human anchors as KGS did (does?)), but then estimate the distance to random play empirically, and offset the ranks such that 0 approximately lines up with it.
It would certainly be possible to artificially deflate your rank below 0, there could even develop a subcommunity of players who are actually playing “misère go”, trying to get their rating as low as possible.
(although to actually be able to have the two groups coexist in the same rating pool, you would have to modify resignations and time losses such that when a player resigns or times out, the other player can freely choose if they prefer to win or lose the game)
(hmm, matching up players that want to lose with players that want to win gets a bit weird… so (to the surprise of no-one) having different games in the same rating pool is probably a bad idea after all…)
1 Like
Isn’t there used to only have Dan level system in Edo period Japan?
Although I am not sure when did the Kyu level system introduced in history. From the record I can find, a list of Taiwanese Go player in 1935 only has range from 15 kyu to 5 dan. And there weren’t any professional ranking back then.
1 Like
Only the FIDE Master and Candidate Master titles are based on solely on Elo in the general sense. IM and GM also require two or more norms over 27 games, where a norm (for the GM-to-be) is a performance of ≥ 2600 (not 2500) against opponents with average rating ≥ 2380 with the appropriate proportion of GM opponents, opponents from other countries, etc.
A 2300 player can also become a GM by winning the Women’s World Championship, the World Junior Championship, the World Senior Championship, one of the Continental championships (PanAm, Euro, Asian or African) or reaching the final 16 of the FIDE World Cup.
The relevant regulations can be found in the FIDE Handbook.
3 Likes
I don’t think you read my original post.
In its unabbreviated form, it said:
Of course, it also has to be noted that Elo inflates; and since the modern FIDE titles are basedly soley on Elo (since performance norms also derive from Elo) they also inflate.
Conrad trimmed my sentence down in his quote.
My claim was sloppy, though, since there are those exceptions you mentioned.
1 Like
Sorry, wasn’t intentional. I guess I didn’t fully grasp it. I still think wins in high-level tournaments is a better system.
1 Like
the system for both Elo and Glicko are only “anchored” to the entry points, aside from that it just floats relative to the population of new entrants.
Possibly… but it would require a couple things: 1) a random play player that is anchored to the 0 point, 2) that this random player would have to make a notable effect on the ratings pool (by playing against other players/bots), and 3) that the entry points are properly spaced away from this 0 point
Here’s another suggestion for OGS ranks: Only Kyus
The only 1 dan in this system would be a hypothetical perfect player. It would have an Elo rating of 0. All other players (humans and AI) would have kyu ranks and negative ratings.
I estimate that top AI like KataGo would be about 1k (able to win about 50% of games against a perfect player with 2 stones handicap and reverse komi).
Everybody else can measure their rank by determining the handicap they need to win about 50% of games against top AI that are good at giving handicap, like KataGo. So if you need 8-9 stones handicap against KataGo, your rank will be 9k.
6 Likes
Against a perfect player there is no win percentage: either the perfect player wins all games with certain settings, or none of them.
That is, once you win one game against the perfect player, you can just play the same moves, and the perfect player will lose again, since the fact that the perfect player lost, means that your moves form a perfect play.
Although, one could argue that the perfect player is playing like someone who knows they lost, but keeps playing on hoping for their opponent to make a blunder.
2 Likes
Well, I assume that the AI opponent of the perfect player is not fully deterministic, so that will be a source of some variation.
Also, you could determine at wich komi handicap the AI winrate against the perfect player crosses 50%. 1 komi point corresponds to about 1/13 of a rank.
When using integer komi, you could even find a komi handicap where the AI tends to get a jigo against the perfect player.
1 Like
If definition of “Perfect” Go player is that it plays towards as big territory as possible(with 100% chance) in current situation, then after few human mistakes, it will be possible to end game with bigger territory than it planned in the beginning, then after few more human mistakes, it even may win against handicap.
And, yes, there may be 2 different moves that gives perfect player equal territory, which it can choose at random. Then human will react differently.
2 Likes
I would define a perfect player as an entity for which there is mathematical proof that it never plays a game losing move in games with perfect (integer) komi. By consequence, perfect self play games will always end in a jigo.
But I have to admit that this says nothing about its ability in handicap games (stones or komi). So maybe there need to be additional proofs about optimal play in handicap games, where the perfect player has a disadvantage.
To maximize the handicap that a perfect player can give to any opponent, I think it needs the ability to perfectly predict all of its opponent’s moves. So it needs the ability to perfectly simulate any opponent’s thought process (all humans, AIs, all other potential entities in all of the multiverse’s lifetimes and ofcourse its own). There may be an issue of running into some of Gödel’s incompleteness theorems, so perhaps it cannot be proven mathematically that such a strong perfect player can exist.
4 Likes
A bit of a circular defn isn’t it? How do you define perfect komi?
1 Like
Perfect komi is the komi where neither color has a forced win.
And I think it’s not circular, but it is self-referential/recursive (as the minimax algorithm is).
2 Likes