We should use goquest-like ranking system where rank and rating are different things. All these discussions show that a lot of people want matchmaking (rating) behave differently than progression (rank). We already did it with humble rank.
Current system has rating and rank coupled together. And all the stuff we’re trying to add on top of it is only making the system more complicated and more obscure. That’s why ideally we want to have two different systems altogether: progression rank and matchmaking rating. This way we can list our requirements for progression and for rating separately and it would be easier to find solution that satisfies all requirements.
I agree. I don’t even understand how they can be meaningfully decoupled - obviously they can be, since someone else does it, but that would be a whole different proposal, and a totally new system rather than a relatively small change being proposed here.
I submitted a pull request to reduce min rank from 25k to 35k, but anoek has a bit on his plate right now so it might be a bit before it gets looked at.
This would be unavoidable even if we only lowered to 26k lol, the answer should just be “minimum rank was extended to more accurately reflect your rating points”
If anoek feels the same, it is ludicrously simple to make it any other number… 30 and 35 seemed like the two most likely places to step down to, I simply chose 35k to match @Kosh’s feature request
Rank is assigned to a user based on an algorithm to extract rank from rating points. This would theoretically extend infinitely in both directions, which is why when it was first implemented we had 11 dans and such. In response, artifcial caps were put on both ends to restrict ranks between 25k and 9d… so all that is needed at this time is to adjust the value of the cap, everything works the same thanks to how it was implemented
They just look at what rank was assigned to the user… they neither know nor care how the rank got there.
Note - I haven’t tested that aspect of things. If it’s to be adopted, this will likely go into beta before the main site where we can give that a decent going-over.
The only change needed should be to change lower limit of rating → rank conversion. This could be defined in one point, or at hundreds of different points in the code. In the first case it’s easy, in the second it’s harder to be sure all parts are using the same limits.
I’ve spoken with anoek, and he has clarified that the deciding factor of where the minimum rank falls is not total population of players but rather viability of the handicapping system.
I’m only summarising here, but basically he doesn’t want a minimum rank so low that handicap stones lose meaning, and it is around this point that we should focus our arguments. If we have good reason to believe that handicap below 25k has merit, we should posit those notions now.
regarding pairing, he has stated that the server takes rating points into consideration on the back end, not merely rank, and will try and pair you with the most appropriate player available
Have we good reason to believe that handicap stones have meaning at the moment anyway for any particular rank or range of ranks?
What fraction or percentage of ranked live/blitz/correspondence are handicap games and are at the correct handicap?
I feel if the summary is relying on
then the first thing I’d like to know is if handicap stones do have their intended meaning at other ranks.
I do believe the ratings work and make sense, x kyu probably beats y kyu if x<y and it could still be true for no komi (one stone) up to maybe 2-3 stones, but surely at that point if you don’t play a lot of handicap games it can get a bit sketchy (4-9 stones)?
Personally I doubt that handicap games at very low ranks make sense. If anything, I feel like there is a case to be made that handicap games should never be ranked…
Handicap games kind of only work for ranks around 15k to 5k.
Below 15k and a handicap stone doesn’t really do too much.
Above 5k and handicaps have a much bigger effect. At Dan level, a handicap stone is worth way more than a single rank. Tokumoto once explained that pro levels are 1/3 of a stone per rank.
I guess this means we are now looking at removing ranks below 20k and above 3d according to anoeks argument…
Also, keep in mind Kageyamas story about how a 1k with 6 stones could give a pro an even game, but lost to the same pro with 9 stones. Relying on handicap stones as an end all be all doesn’t work in this game. They are not a great equalizer.
My feeling is that we don’t play enough handicap games for handicap to meaningfully map to rank per stone anyhow (would be good to see some hard data on this if someone has any?).
Add to this the low likelihood (my guess) of TPKs chosing handicap vs each other, and it seems like a moot point.
It’s great news for our ranking system if we play more handicap than I realised.
It is actual play that keeps rank aligned to rating for handicap - if lots of handicap play is played then the only way to proceed up in rank is to be able to win with 1 stone over a person below you.
(If the actual formula is too far off, then this would cause rank instability under handicap play)
