I mean the huge issue is that anoek and Alex are taking handicap stones to be the absolute truth and only real measurement in go ranking, when handicap stones are not really representative of rank in a traditional or modern sense.
Pros do no play official games with handicap stones. A 4p doesn’t get a stone against a 7p. Further to this, handicap stones are not at all favoured in China.
From Kageyama:
With the traditional handicap placements, the only consistent strategy Black can follow depends on the use of influence. This is particularly true in the early stages of the middle-game fighting.
To further my point, read over Rank And Handicap at Sensei's Library. I want to draw your attention to “However, one should never lose sight of the purpose of handicaps, which is to make teaching games more interesting and instructive. Using a handicap system in tournaments may be questionable to begin with, so that the question of making it a fair system becomes irrelevant.”
Today I see pros offering reverse komi instead of handicaps, as handicaps change how you play, and if you don’t know how to play with handicaps then simply applying traditional even-game theories makes you play worse. So in effect, handicap stones measure how well you play with handicaps, not how well you play go.
Ah, come on, Kaworu. It’s easy criticize without offering anything yourself. Unless you offer anything better, handicap stones is the only reasonable way of defining rank spacing. Otherwise how do we define what the difference between 6k and 9k is?
If we make up some different rank spacing and 6k will have to give 9k 5 stones for it to be an equal game that would be a pretty bad system. I do think it’s a custom everywhere that the rank difference shows the number of handicap stones for the game to be interesting and equal. And how do you make sure that it’s indeed the case? You define the rank difference through the handicap stones, simple.
And taking pros as an example is silly because pro ranks measure achievement and not strength. An old 9p is still 9p despite their playing strength. If we take pros as an example, we should switch to goquest system, their ranks are more an achievement than strength indicator too.
Among amateurs (and especially online because you play strangers) ranks are needed for finding equal opponents and determining handicap stones. (Well, feeling good about yourself improving too).
Once I participated in IRL handicap tournament. 3 games ended by resignation but the other 3 ended with 7.5, 2.5 and 3.5 difference (2-4 handi stones). That’s a pretty damn good system under which I can meet a complete stranger, calculate handi from ranks and get these very close intense games.
Anyway
This is kind of an impossible standard. Finding meaning in handicap stones. What does that even mean. I could see that handicap can confuse beginners. But what would meaningful handicap entail. Hmmmmmmmmmmmmm. Is the point in that with handicap stones whoever has handicap stones wins in 25k+ games?
They’re using an ELO system where you just say I want 2100 to be 1 dan, and all other grades/ranks are spaced 100 points apart (it’s different for pro ranks)
On the face of it the definition has nothing to do with handicap stones. However since handicap games are accepted and counted in rating changes and all the tournaments I’ve been to use handicap McMahon system (just to make it enjoyable for people), then the correspondence between rating and rank is imposed that way.
You’d think I would’ve remembered the stats about the percentage of handicap games since I was also active in that thread, but I didn’t Thanks!
It does feel like handicap games are in the minority from that sample, with say 25% of reasonably ranked players playing handicap games at all, and 16% of games in the sample being handicap games.
It does kind of suggest that the current state of correspondence between ranking and number of handicap stones could be a bit loose.
I wouldn’t be sure for instance if I’ve played one of each type of handicap game on OGS ranked (1-9 stones), I’ve more than likely played games with 1-3 stones (I think some are ongoing although with Chinese placement), and I’ve definitely played standard 9 stone games before. It wouldn’t make me feel like I’ve a 50/50 chance of beating someone ranked lower than me though with the proper handicap
Maybe we should introduce a handicap ladder (Also maybe I should just expand my automatch settings to ± 9 although I am mainly playing correspondence at the moment so not using it that much)
The only reason we have ranks at all is to facilitate handicap. Apart from that, it would be much simpler to switch to only using glicko rating. Rank only exists on OGS as a way of estimating how many stones a player may need to give another in a handicap game. Now, people might assign other meanings to it such as identity or progression, but those are not its purpose.
I suppose that’s a completely reasonable point; we could just get by with rating alone.
I guess I do just prefer having ranks which are a discretised way of measuring progress. It’s also a more natural way to communicate strength to other plays in person - I’m an x kyu - although if we didn’t have the rankings, maybe people would get used to saying I’m 1500 or 1000 etc. It would probably be more natural to round to the nearest hundred just because decimal.
I’m quite surpised (to put it midly) that some people in this thread consider ranks and handicap unrelated. As stated by others, rank differences are defined by handicap and handicap is derived from rank difference. That’s how ranks have been used for generations in real life go clubs (at least in Europe).
If players on OGS rarely play handicap games, then ranks are pretty meaningless, or at least an unneccesary complication. If you only play even games and you’re just looking for opponents near your level, you can just use Elo ratings and forget about ranks.
The EGF system is not an Elo system. An Elo system has 100 points difference for 2:1 odds in an even game. The relationship between even game odds and rank difference is not a linear mapping of 100 Elo points per rank. It varies across the rank scale.
Statistics collected from historical EGF game results indicate roughly this relationship:
rank width (=100 EGF points)
Around 10k EGF: 50 Elo points = 57% winrate in even game with 1 rank difference
Around 1d EGF: 100 Elo points = 64% winrate in even game with 1 rank difference
Around 7d EGF: 200 Elo points = 76% winrate in even game with 1 rank difference
The ranks exist because that is the tradition in Go and also because of the connotation with handicap.
For this reason, the sensible cap for ranks is 20k - 9d.
Lower ranks are a fiction created by false humility, bad personal experiences from skewed rank assignments or other circumstances that temporarily seem important to the affected person.
There are no 42k players, and there are no 33k who can fairly give 9 stones to them.
According to rating, they clearly have.
Glicko is the rating system on OGS and so that is what we should be looking at.
That being said, I see another big issue at work here and that is the absolute chaos in the lower ranks on this server.
I have been following the progress of a fresh beginner here for two months and my impression is that all ranks beyond 15k are mostly a roll of the dice, even on accounts with over 100 ranked games on them.
I don’t know why that is, but I know it won’t be fixed by a lower rank cap.
Instead, I believe OGS should work to increase the confidence of its users in the ranking system.
We get all of these questions on Glicko that should be really easy to answer.
Establish a FAQ page on the specific OGS ranking system, not just link to the Glicko article on Wikipedia.
Provide more Glicko-relevant information:
the user’s volatility value, a part of the Glicko system
the boundaries of rating periods and which games form a batch
Open-Source the Glicko implementation to independent review.
I’m listing the third point because I vaguely remember from past threads that it was implemented “in-house”. Also someone found it weird that even with frequent games, the variance never seems to go below ±50 - an interval that still spans multiple ranks above and below the mean!
Unfortunately, such threads are quickly shut down with hand-waving “Glicko is complicated and works in batches, just trust that it works out”. Even if true, this does not inspire confidence.
20k children in my childrens club can routinely give 4 stones handicap on 9x9 to the novice children in the same club.
I can give 4 stones handicap on 9x9 to those 20k players and 8 stones to the novices.
Ranking all these players in one big 20k bucket feels very wrong to me.
Why not re-rank them between 15k and 20k?
Their skill difference will be well-defined and you don’t have those ridiculous numbers.
Also, you’re talking about 9x9, where one handicap stone will have a much greater impact.
How do they fare against each other on 19x19?
What is their Glicko2 rating?
Without these answers, I’m afraid that this anecdote has no bearing on the OGS system.
Because in real life tournaments their club ranks seem already inflated as is. They cannot beat adult 20k-15k players.
Calling something ridiculous is not actually a proper argument.
I know, I use one handicap stone per 6 ranks.
I only let them play on 19x19 from 20k and up. Between 35k and 20k they play on 13x13 (with handicap of 1 stone per 2 ranks difference). In my club, a 20k can give 8 stones handicap on 13x13 to a 35k.
They play real life games only. Only the strongest have played a few online games in EGF youth events.
According to the EGF system a pro rank is 1/3 stone or 30 EGF points wide (derived from the handicaps between pro ranks used from classical times to the Oteai system that the Nihon Kiin used to rank pros in the 2nd half of the 20th century).
In the EGF system 1k = 2000 EGF and 7d = 1p = 2700 EGF, so a 1k should get 8 stones and give komi.
9p = 2940 EGF, so 1p should get 3 stones handicap without komi (as it was in the Edo period and also in the Oteai system, I suppose).
I was not using Elo ratings but EGF ratings which is defined as 100 points per rank (full handicap stone), not as 2:1 odds like Elo.
If I remove the EGF ratings from my post, it becomes like this:
1p = 7d EGF (by definition of the EGF), so there is a difference of 7 full stones between 1k EGF and 1p, corresponding to a fair handicap of 8 stones with black giving komi.
Between 1p and 9p is 8 pro ranks difference. With 1/3 full stone per pro rank, this is a difference of about 2½ full stones, corresponding to a fair handicap of 3 stones without komi.
Thanks!
.