If the floor does get lowered then it’s worth being mindful of the default matchmaking range of +/- 3k
1 Like
In practice it will seem so because the standard deviation at those ranks is in the order of 5-10 ranks. But from looking at 100’s of games of sub 25k players playing against my bots, I think there’s a real difference in the ranks:
- 26k: plays go, fails at making eyes, counting liberties, but does occasionally try
- 40k: plays go by building walls but doesn’t know that making eyes is a thing
- 50k and below: fills the board with geometrical patterns, ie. not playing recognisable go
This is of course just a personal observation and not accurate because this group barely plays ranked. But it feels right. 
Proposal:
Show the real ranks (down to 69k) to people who have a rank below 25k.
This will keep them motivated to see their rank move up when they win. It removes confusion about the ranking system.
Cap the real ranks at 25k for people who have a rank of 25k and above. (The current situation). This keeps everything as it is now for people who don’t need that extra detail.
6 Likes
In general, Elo gaps between go ranks shrink towards lower ranks. The Elo rank gap seems to be about 100 points around 1-2d EGF and about 25-30 points around 20k. At random play I assume the Elo rank gap is close to zero, because a couple of handicap stones would hardly benefit a random player. So perhaps we can interpolate that the Elo rank gap around 60k is somewhere around 5-10 points.
If 25k corresponds to an Elo rating of about 500 (as seems to be the case on OGS), and you extrapolate downward by subtracting 35 Elo rank gaps linearly diminishing from 25 to 5 points, you end up at an Elo rating of roughly 0 for 60k.
4 Likes
I really appreciate all your efforts, I really do! (This is also why I am a contributor, despite no longer playing online myself.)
Having said that, I believe you only tried to fix it and there still is significant room for improvement.
In fact, the situation is significantly worse than I initially thought!
I only learned after I made my initial statement, that differences in playing strength are simply ignored below 24 kyu.
If this is true (and I’m afraid it really is), then this is entirely outlandish in my eyes. (Sorry for the harsh words.)
I have worked with rating systems a lot (and I mean: A LOT), I try to convey the beauty of chess (for decades) and Go (for years now) to beginners and for me one of the most important aspects for someone new to the game is to not let them think they are stupid and to not let them play with/against someone who is significantly above their level.
I can not even begin to comprehend why a game server would lump everyone below a certain level together into one pool. It is like saying “Below a certain strength the server is happy to create pairings that will absolutely make no sense whatsoever and frustrate everyone involved.”
Again, sorry for my emotions but this is super frustrating for me.
Does chess.com make the same mistake? Of course not! They certainly don’t want to frustrate their (potential) customers!
I did search for a beginner on the server yesterday and found JoCatz.
He/she played 10 rated 9x9 games yesterday. I briefly looked at the games and I believe 9 of the 10 games didn’t make any sense before they even started, because the opponents were playing on an entirely different level and were just eating JoCatz alive.
How much fun did JoCatz have yesterday? Close to zero I would imagine.
Will he/she ever come back? Who knows.
Please fix this. Please respect the playing strength of beginners and pair them against players with similar level.
9 Likes
I simply don’t believe this*. Particularly when it’s 60k vs 30k like dokbohm and other perpetual losers. Sure the error bars on the ranks are larger, people change ranks more rapidly and handicap stones don’t really work, but there’s still information in those ranks that can improve pairing.
/* but if someone collects some stats of say 100 40k vs 30k games and they are 50-50 ish I’ll stand corrected.
6 Likes
I’m all for this part of it. I’ve thought for years that it’s unfair that people below 25k have to dig to even figure out what rank they are
I’ve heard this a lot. I thought it wasn’t the case, and that the problem was purely aesthetic (not that I don’t think even then it’s a big problem which should be addressed. People below 25k aren’t somehow “less worthy” than you or me to know what rank they are, and seeing their rank can help opponents know how to help them improve or whether to challenge them), but if it is indeed true that the matchmaking ignores differences below 24k, that is a whole new dimension of a similar scale to the problem
Yep. 100% agree. I know it can be hard to find matches if there legitimately aren’t many players as weak as oneself, but it should at least be minimally facilitated via honesty about ranks at all levels so that appropriate open challenges may be accepted when applicable, and matchmaking taking into account all ranks likewise
I agree. I trust that a difference in Glicko-2 means on average a difference in outcomes, and OGS ranks are a simple empirically derived formula conversion from Glicko-2, which entails that the same applies to a difference in ranks
I suspect that this whole “ranks below 25k don’t predict the winner” idea came from OGS failing to account for the half-stone inaccuracy of handicap names. Of course when the rating system is fed information that’s distorted by half a stone, the error will add up enough to overcome the actual predictive power of it in more extreme circumstances
3 Likes
If Arpad Elo would hear that a difference in rating would (on average) not have predictive power for the outcome, he would not stop spinning in his grave.
The entire point of Elo (and Glicko) is to calculate ratings based on performance and provide estimates for expected outcomes based on rating differences. And this is true for any level of play.
2 Likes
If the Elo rank gaps around 35k are as small as ~10 points per rank, the Elo gap between 30k and 40k would only be some 100 points.
So even though the 30k would have about 50% chance of winning when giving 9 stones handicap to the 40k (which would be the case by definition, with a rank gap around 9 between these players), the 30k would still only have about 65% chance of beating the 40k in an even game (the win probability for a 100 point Elo gap).
2 Likes
It may just be repeating what others are saying, but I feel so strongly that ignoring ranks below 25 kyu is grossly unreasonable that I cannot resist adding my voice and words, saying please fix this! I have been trying to help @dokbohm, and he had no idea that he was actually ranked 69 kyu (or where to see that), and he gets paired against 25 kyu players without handicap. That is absurd; a player like him should be given a considerable handicap, and would have far more chance to win the odd game. (Or “would have had”, since for some time he has only played to delay the capture of a given number of stones for some target number of moves.)
I am sure there is a dearth of information on what handicaps would be appropriate for these ranks — or rather of awareness, as there must be plenty of information in OGS. It should just be statistically evaluated, and applied to those ranks. Perhaps the ranks are also wrongly calculated because of the poor matchmaking. It certainly seems unlikely that a 25 kyu can give 40 stones to a “65 kyu”, so the linearity presumably breaks down somewhere. Perhaps the handicapping algorithm below 25 kyu could even be self-calibrating.
7 Likes
It’s even better than that (for the case in favor of allowing full rights to sub-24ks), since handicaps greater than 9 stones 19x19 equivalent are not ranked afaik. So while you are right that such an extreme handicap is likely to be inaccurate, that’s not an issue we need to worry about in order to give 25k and below full rights as OGS denizens
3 Likes
I believe that at these ranks – and with such rank differences – they shouldn‘t play 19x19 anyway but rather 9x9, and, believe it or not, when I play with total n00bs on 9x9 I sometimes give them 9 HC stones 
4 Likes
I don’t believe so. They should play whatever size they want and this won’t interfere in establishing their rank.
Besides I guess that 9 stones on 9x9 is not equivalent as on 19x19 for our ranking calculator (if I remember well some older thread). I hope so!
3 Likes
as one of those lowly players you are talking about – playing a 9x9 game is a completely different game of go to a 13x13 as it is to a 19x19 which i prefer by playing a 9x9 game one mistake maybe two and its over - but when one playes a 13 x13 game there are different levels to the game the start the middle and end – in 19x19 game– the start— the near middle– the middle - the near end - then the end—– so one can make a mistake in one of those areas but continue on and at least salvage soemthing from the game — not so in the 9x9 game its slam bang in and out one mistake if your playing a really bad player also— then maybe two mistakes then its done so i believe low level players should play larger game boards to fully enjoy the game rather than go to a 9x9 board and pla couple of moves then ite really over – my option form a low level over 60k player
Seems I didn’t express myself clearly enough … what I meant was …
… when playing with such huge rank differences it is better, IMO, to play on a smaller board – and thus with fewer Handicap stones, NOT 20 or 40, etc. – instead of on a 19x19 board with several dozens of HC stones.
But sure, everyone is, of course, entitled to do as they please 
2 Likes
Looking at profiles, I see that the lowest possible rank on OGS is 69k (100 Elo), while 25k is about 670 Elo. The difference, 570 Elo, is quite big, it is as big as between a 5k and a 16k.
7 Likes
Elo gives the weaker player a 3.6% chance in this case (Glicko would say roughly the same).
Not exactly a thrilling contest - unless you’re sandbagging.
Do this 10 times in a row to beginners and you won’t see 95% of them ever again.
2 Likes
losers??? yes we lose—- but losers – Guess I know where i stand here
Please don’t be quick to take offence where none was intended - especially in a multi-lingual environment.
These discussions are aimed at helping people with ranks below 25k - such as yourself - to find best pairings. It’s important to be able to describe the different skill levels, all the way to “people who lose almost all their games currently”.
It we succeed, you may actually win more, because you will be paired more “fairly” more often.
3 Likes
i stopped on this site, taking offence. i just quoted what the poster was lumping everyone into. that is at the bottom, newbies and old bad players alike as losers – not a nice feeling in any language to be labelled a loser but then again when the shoe fits i guess, to some its quite harsh to me just par for the course on this site ny skin is thivk enough – but thanks for the understanding explanation
I think in context “losers” did not mean the same thing that it means when used as an insult
4 Likes