A lot has already been written about the ranking system on OGS and other servers. I’d just like to add one thought that has been standing out to me lately. After spending several weeks playing almost exclusively on other servers, I’ve become especially aware of the negative effects of OGS’s ranking volatility.
For a significant portion of the player base, it’s practically impossible to estimate a player’s actual strength based on their rank. I regularly see opponents fluctuate by at least two ranks, and quite often by as many as three, in both directions. A good or bad streak can send players flying up and down the ladder so quickly that their peaks and slumps sometimes repeat in less than a month—and this keeps happening over and over again.
At least in the lower to mid SDK range, the skill differences between players with the same official rank can be surprisingly large. This doesn’t just affect the rank calculation itself. What concerns me more is how it negatively impacts the overall playing experience.
In general, it already takes quite a while to find a game on OGS. As a player, I only have one way to reduce the waiting time in automatch: I can allow a wider range of opponent ranks. But why would I, for example, allow opponents two ranks above or below me if, given OGS’s ranking volatility, that can easily translate into an actual skill difference of four ranks?
I understand that using a highly volatile ranking system is a deliberate design choice. I’m sure there are valid arguments in its favor. But in practice, this extreme volatility undermines what I think is one of the most basic expectations players have: finding a reasonably even game without having to wait too long. That sense of balanced matchmaking exists far less on OGS than it does on other servers.
I really wish OGS would seriously reconsider the way its ranking system is calculated. This isn’t just about whether rapid rank changes feel rewarding or frustrating. For me, the consequences go much deeper. The ranking volatility creates real problems for the overall playing experience, to the point where OGS becomes a less attractive server—even though, in terms of overall user experience, I still think it’s the best platform available.
I feel like the flexible and volatile system is great for new players, who are learning new things at very rapid pace, and who want to track their own progress via their rank.
But for users who have played for a long time, their skill level is usually quite stable, thus OGS’s rating system seems very random at times. Everyone has good days and bad days, but after many years and thousands of games, we don’t suddenly get 2 stones weaker within one single day.
So as a solution, i think we could slow down the rating changes after some limits. Maybe something like “rating impact of only 50% when the account is older than 1 year and has played more than 300 ranked games”? That would basically mean, that a 19x19 would have the same impact on the rating as 9x9 games have now.
The volatility is not that high that averaging the displayed rating from the last 10 games would be meaningfull for anything than a new account and in this case it is not what I would want to see.
I would encourage anyone trying to argue here to get familiar with the Glicko-System that is used for the calculation of ranks on this site. Glicko 2 has a concept of rating volatility that usually goes down with the number of played games. There are constants in the equasions that can be tuned. But the overwhelming success of this system shows that it should be applicable for OGS.
I hope I dont drop nonsense but I think you get matched by your overall rank. Meaning if youre 1k in live matches, 5k in correspondence and 3k overall you get matched with 3k opponents in any time section. Correct me if Im wrong.
On every server, ranks fluctuate by ±1 anyway, so if you allow automatch to select opponents who are 2 ranks stronger, whatever you do to the ranking system, you’ll get paired to opponents who are 3 ranks stronger in reality. Does it make a big difference if instead the opponent is 4 ranks stronger?
People’s playing strength is extremely volatile online though. There’s like no way to fix that. Lots of people don’t take online games very seriously or will play tilted leading them to play a couple stones below their rank.
On any other server, I think the playing strength is just as volatile; it’s just that their rank doesn’t adjust as quickly. I frequently find players my rank who’d play like trash for the game that I’m wondering how they ever got to their rank.
On Pandanet, the ranks adjust much slower. But that means while I’m improving, I’m literally crushing all of my opponents game after game until I’m bored because the ranks barely budge. Hence why I stopped playing on that server. And I swear, there’s also players who’s massively mis-ranked on that server because the ranks don’t budge.
I think your understanding of the used rank is correct. But I don’t think that exempt rare situations the playing strength of players differ that much. All this has been discussed vastly in this forum.
We know that volatility is part of the deliberate design of the system. Aside from that, the OP presents no evidence that the problems discussed are caused by the system volatility. Nowhere do I see any consideration given to the idea that those problems are caused, at least in part, by the huge amount of sandbagging and botting on OGS.
I always play even (not handicap) 19x19 games always with exactly same opening.
Apparently “huge amount of sandbaggers and botters” do not exist among my opponents.
So I doubt other sdk often encounter them too.
I think players with not stable graph just do something chaotic themselves.
(accidentally)
As you should know, such an argument is specious, because absence of evidence isn’t evidence of absence. Most people have never seen a green fireball or ball lightning, but that doesn’t mean those phenomena don’t exist.
I have seen the evidence regarding sandbagging, in the form of validated reports, but its nature precludes me from showing you, just as I can’t show you a green fireball. Moreover, publicly discussing sandbagging or botting in particular is unwise, as it can give trolls ideas.
I suppose it is safe to say that in the past two months, Community Moderators have dealt with at least 69 sandbaggers who throw games (I don’t know exactly how many because i am not awake 24 hours a day). Most of them have thrown more than five games.
Now, that may not be large as a percentage of players or games, but it is not insignificant. All I have asked is for that factor to be “considered,” because up to now in this thread it was ignored.
It’s hard to estimate the percentage of sandbaggers or botters, I don’t think they represent much more than 1%, but due to the small playing pool on OGS, you can get paired with the same player several times in a row.
Other factors that may cause rank fluctuations:
Playing with an unstable internet connection, and losing many games by timeout.
Playing mainly 9x9. On that board size, a mistake can be fatal, and it’s quite common to lose against a much weaker player.
I wouldn’t want to say that playing much stronger or weaker opponents isn’t fun. It’s a game of Go, of course it can be fun.
What it boils down to is: It is more difficult to find an even game on OGS compared to other servers. The volatility of ranks seems to be at least a contributing factor for this happening. Most of the time I am not looking for a game with very unbalanced skill levels. And still that’s what I get significantly more often on OGS than on other servers. That’s how I would summarize my experience comparing some of the servers.
Possibly, this is due to Westerners playing in a more volatile way. Due to the strong focus on theory, lots of Westerners are like Dan level at Josekis and common shape patterns. However, once it comes to fighting, the play would be extremely inconsistent. This is especially true at the lower sdk levels (which I was at not too long ago) where oftentimes have strong understanding of theory, but would occasionally make ddk level blunders while reading which makes it hard to climb.
This’ll manifest as people looking like a Dan level player when they got lucky and didn’t make serious mistakes, but looking like ddk players when they are. Also, need to factor in that after 1 serious mistake, people will tend to make more (since when people are losing, they’ll often play hyper aggressive to get back into the game).
On other servers like Fox, it’d be like very little knowledge of theory accompanied by mediocre reading at equivalent ranks so their play will look more consistent.
What if rank system of OGS is just more honest? Players actually play worse sometimes and OGS actually showing it fast?