Now that I know some of the basic facts of the Go world, I understand that levels of player are judged in a way that 10kyu player can give 9kyu player one handicap stone (with 0.5 komi) and play evenly; 10kyu player can give 8kyu two handicap stones, so on and so forth. Obviously this should be based on long time accumulated playing and when the strength of player stabilized.
That said, I am wondering why amateur level starts at 30kyu and tops out at 7dan? That’s a very wide range and that means the top level amateur can give an absolute beginner 37-stone handicap game but not 40 stones? Also I realized the kyu and dan system were there before the internet and various go servers. So back in the day, basically one would play against a certain level player (say 1dan) and decide his/her skill level is about 3 stones behind so he/she is roughly 3kyu?
Nowadays, thanks to internet and OGS, I can play against people all over the world. My essential question is how should I judge my “real world Go skill” based on my ratings and rankings from thousands of games played on OGS? If I eventually settled on 3kyu on OGS, does that I am 3kyu level in traditional amateur Go world sense or if there’s a difference how much better or worse am I?
Rankings are a way to decide on standing within a particular population.
Every server has a population that is largely (albeit not completely) independent from that of other servers, i.e. different people frequent, say, mainly IGS or mainly OGS, or only Fox, etc.
Everytime you enter a new population of players, you have to work out your relative standing, because ranks are by definition relational. There is nothing intrinsically “3 kyu” about being 3k on any given server.
Long story short: you will have to play other people with a more or less established rank among the population in question to find out where you fit in.
And as you said the more you play the more accurate your rank actually is, since we all have good days and bad days, and sometimes we win/lose because of pure luck rather than skill. Especially at the lowest ranks where players still are unfamiliar with basic shapes or common sequences.
A long time ago I visited a go club where I was ranked at around ~1kyu after a few games, but I remember playing back then in KGS and I believe my rank was more like 5kyu or something like that. So it’s really relative to the people or ‘community’ you are playing with.
I want to add that generally the highest handicap is a 9 stone, after that you can start giving reverse komi in addition to the handicap. Or you can let the weaker player do free placement instead of following the star points. A 37 stone handicap would be a little outrageous.
If Alice plays evenly (wins about half of the games) against Bob when giving him X stones, and Bob plays evenly against Charlie when giving Y stones, does that mean that Alice should play evenly against Charlie when giving X+Y stones?
I think it becomes tricky when taking these conjectures to very large extremes. It would be difficult to say how much handicap is necessary for the 30 kyu and 7 dan player to play evenly together. A 30 kyu player is still very much a beginner and possibly still rapidly learning and improving. Maybe in losing a 37 handicap game to the 7-dan player, they would have learned some things to then easily win a 30 handicap rematch game.
Thanks a lot for all the kind replies. I was actually wondering how different “rating” is from “ranking”, now I get some sense.
I asked the question because in all the books I am reading, the author refers to the problem level or knowledge level using the kyu system. Some books were written in the 70s and 80s so I think it was based on the traditional ranking system. I am totally spoiled by the convenience of OGS and able to play as many games as I want without even walking out of my door. But I wanted to see how online kyu is working. I went through “Graded Go Problems For Beginners Vol 1 and 2” by Yoshinori fairly quickly but now stuck really hard in the middle of Vol 3. Also the “Lessons In The Fundamentals Of Go” by Kageyama listed some “beginner problem” which I found to be very hard to comprehend. I feel I’m totally novice level despite I can sometimes win against much higher ranking player.
Anyway, keep learning and keep improving is worth more than anything else
Interesting, when we reduce your argument to its core, you not only question the transitivity of handicap stones as a function of rank difference, but simply setting X=0, Y=0 we can see that you question the very meaning of rank.
But I think the whole thing works in the opposite direction: The ranking (rating, actually) system is not designed to test whether you are in fact n [grade] and could give m stones to Charlie, it is designed to make the most accurate predictions of the outcome of future (even) matches.
In that sense, ranks (here) are just a convention based on even-game winning probabilities. I doubt Glickman even considered the possibility of non-even games; mostly because rating (i.e. “skill”) is supposed to account for all differences in winning probability. By adding handicaps into the mix, you’re messing with the system from two angles. One is observed player skill, the other is a change in the game itself.
In theory, we all play handicap games. If we did, then 1 kyu of rank would strongly correspond to 1 stone, because if I am 10k and I can’t beat a 11k with 1 stone, then I will quickly become 11k myself.
That is how the stone to rank correlation started: by definition of the ranking system being based on the handicap stones.
At OGS so little handicap is played that it’s only indirect “forces” that keep the correlation in place.
These forces are (1) the initial calibration of ranks to what people think is reasonable, which in turn is based on handicap-based systems to a reasonable degree (2) the care in place to make sure that OGS ranks don’t “drift” (3) ongoing informal calibration of our ranks with other sites where more handicap is played
Yes, for X=Y=0, it becomes a question of whether rank itself is transitive, which is a fundamental assumption made when measuring a player’s ability with a one-dimensional rank/rating.
To ask the question a different way: is it possible to find three players, where Alice usually beats (wins a significant majority of matches over) Bob, Bob usually beats Charlie, and Charlie usually beats Alice? Perhaps such a situation could arise due how their individual strengths, weaknesses, and playing styles match up.
For the game of go, maybe ranks and handicap stones (if not taken to extremes) are mostly transitive, at least in a broader, population-level, statistical sense. However, I believe the intransitivity of skill is an observable phenomenon in many other games/sports.
is it possible to find three players, where Alice usually beats (wins a significant majority of matches over) Bob, Bob usually beats Charlie, and Charlie usually beats Alice?
I’m quite certain, actually. Then again, the rating changes would balance each other out so their displayed rating would not change, provided they play each other equally often. I’ve thought about adding a “concurrent validity” / “versatility” element to rating, indicating the degree to which the rating is informed by different sources… doesn’t have to affect the actual rating used for matchmaking, though.
Case in point, there’s a guy on KGS who has been about the same rank as I for the longest time (I started playing him when we were both 10-8 kyu, though he’s been stuck at 1d whereas I reached 3d a few years back), and to my knowledge my winrate is somewhere around 90%.
Using the example of Alice, Bob and Charlie, another thing to consider is the style of play. Maybe Alice knows a lot of josekis while Bob is really agressive and Charlie is really good at invading… or whatever. My point is that the stronger you get, the more you are ‘expected’ to know more joseki or handle different situations, have better reading, etc. That’s another factor why as you get stronger, you move on to the next rank slower.
For instance Tygem is really agressive and people get away with overplays all the time which is why a 2d Tygem is probably 1d or even weaker in other servers, where they may be used to see other styles of play often.
Exactly. As anything below a 7d player, we can readily make the assumption that your knowledge of the game is incomplete in many ways, and therefore there exists a high degree of variance in your playing strength from move to move. The change in your playing strength over time is by no means a continuous function that can neatly fit on a 2D graph, but more akin to a many-dimensional field with variable topography that either coincides with or diverges from observable reality at critical points related to concepts fundamental to the game. When your game happens to traverse an area of your field (which represents your playing strength) that diverges from reality, you make mistakes that change the binary result of your game, which is what your rating is tied to. It’s the reason why weak players, particularly kyus, with similar ratings will often experience results with extreme point differences, instead of always tending towards some mean or normal point difference. It’s also the reason why a player who excels at a couple of fundamental concepts can sometimes achieve and maintain an artificially high rating that doesn’t accurately reflect how strong he or she actually is (a good example of where this can readily be found is in players who only play against weak bots, ie. below 5d).
Ratings are, at best, statistical approximations that only very vaguely and poorly track the position functions of multiple players on a “win-loss ratio” dimension relative to each other. What ratings really represent are the probability distribution, between two players playing a game, of the game tending towards an area of the fields that represent their playing strengths where one player’s field will diverge further from reality than the other player’s and thus lead to a change in the binary result of the game within some confidence interval.
In reality, some player A who has the same rating as some player B, while their relative ratings would suggest a 50% chance of either player winning, may actually have a real probability, based on each player’s understanding of the game, of winning against player B 95% of the time. Maybe games between players A and B nearly always lead to some position that player A has a far better understanding of than player B, while their games simultaneously very rarely approach positions that player B has an advantage in over player A. The rating system has absolutely no way to be predictive of this case.
What’s the take away? Don’t pay too much attention to ranks or ratings, either your own or your opponents, or put too much stock in handicap stones as an indication of a difference in strength. I’m pretty sure that a 30k will beat a 7d nearly 100% of the time with a 37 stone handicap placed in any meaningful arrangement.
This is the only point of your argument I disagree with. A 7 Dan would slaughter a 30 kyu no matter how many stones he had unless litterally every point was illegal.
if you have 37 handicap, you can draw 2 solid lines almost completely across the board on the 4th line for guaranteed 100+ points and easy fights. You can enclose the all of the corners solidly and make multiple ponnukis in the center, or just make a box in the center. I mean, you’re really underestimating how extreme a 37 stone handicap actually is, even in the hands of a complete beginner.
Even when you do a traditional placement for the handicaps, the 7d certainly has a good chance to confuse the 30k and kill multiple things on the board, but the 30k will still very likely win every game, because there’s just too many stones placed that can bail him out of his mistakes and limit white’s ability to make points.
and I won 
