Derived means modified to work for ranks (assumed to be determined to a large degree by handicap in informal club games against stronger club members, at least for kyu ranks). It’s no longer a true Elo system which is purely based on even game odds.
I suppose that’s true, but it’s much easier to quote the things it’s closest to than to have to call it something else 
I guess the usual ELO system doesn’t also have this epsilon parameter, where the sum of the expected outcomes is 1-epsilon? Some of these things could just to be down to a practical implementation of the system vs the theoretical system too I imagine.
Anyway, it’s kind of off topic I suppose, I didn’t realise how many posts I’d missed!
In summary - we don’t necessarily need to ranks to just be able to rate players one above the other - glicko does that.
We’d like the ranks because
- They’re useful to dictate the number of stones that should give a roughly equal game between differently rated players.
- other go servers have them, egf AGA, books etc all talk about ranks.
- (me at least) ranks are on a more manageable scale to talk about, somehow it’s harder to comprehend the skill difference between a 1540 and 1720 rating player, while it’s a bit easier to understand the difference between an x and y kyu player (I believe)
But then why can’t we include 25+ kyu ranks again? Handicap stones might not work for them? [I really would like to organise a sitewide tournament to test this :)]
The 25 kyu cutoff felt arbitrary not knowing the evidence for other than the quote from anoek
4 Likes
I also don’t agree with this at all.
If it were true, why do we bother correlating our ranks with other servers?
If it were true, why do we put rank in the most prominent place next to our name when most people don’t even care about handicap?
In fact, if it were true, we should express it in stones-delta.
I should see your “rank” as the number of stones between us. The absolute value would be meaningless.
But we all know that is not true, and no-one would go for a suggestion like that.
This is not consistent with the prominent and careful way we treat it.
Actually, a key the purpose is for credibility of the server and players. People want to know where they fit in the world of skill and ability.
That’s not to say that all servers and ranking systems do align well - but the interesting thing is that we know this is the case because people care about that, and comment on it.
If the purpose of rank was only to determine handicap, it would be buried in the matchmaking system, and we would not be having this discussion.
9 Likes
Well there can be a difference between the basic purpose of something and the meaning it has for people. Otherwise nobody in their right mind would ever buy a Ferrari.
4 Likes
I do think that ranks have a significance beyond determining the handicap that challenges both opponents. By it’s relation to handicap, it is a natural universal scale to measure progress in go. It’s not perfect, because everybody uses a different system (unlike chess), but it’s mostly okay.
2 Likes
How many non-tournament games (on OGS) use handicap?
2 Likes
In small go communities like real life go clubs, you don’t have many opponents of exactly your level. So IME most of those informal games use handicap to make challenging games for both opponents. Sometimes with reduced handicap for teaching games.
1 Like
Sorry, I’m only talking about OGS. This entire discussion is about what OGS can do with its large datasets of games, glicko2, etc. At no point have I ever been talking about real life go clubs, only OGS. OGS is very capable of doing more than any real life go club can do.
I want to know what the limits of what we can do with ranks, probabilities of winning, data analysis, etc are as we have access to so much.
EDIT 4… I kept meaning to ask you if you had re-done that reverse komi with other bots, or other data sets. I already plan on trying out your numbers, but I’m just curious.
5 Likes
I don’t know any other strong AI that evaluates points and is freely available. So I don’t have any other data of similar reliability. I only know about some experiments decades ago where pros played against each other with large handicaps to determine the komi value. I think they found about 140 points for 9 stones handicap.
3 Likes
No, it’s not. I’m not being antagonistic, I’m genuinely asking these questions. See my first question/post.
There have been many such posts, including the following one, which is why I’m trying to argue on their behalf here…
You’ll notice that I was fairly quiet in this thread until the flood of questions regarding 25k on OGS came in:
That’s only a small clip, but I have been getting a lot of PMs over the last few days since it was first brought up on there about a week ago.
5 Likes
I think it’s partly my fault 
My point was, if the real sticking point with extending ranks beyond 25k is that rank differences must correspond to handicap stones strictly as per anoek, then we
- Probably need more data to back up that 25+ kyus can use handicap stones equally well as anyone else (hence discussing other rating ranges and also suggesting a possible tournament :))
And
- We would want to be certain that at other ratings (and ranks) that the handicap to rank conversion really is working currently at all of the current ranks/ratings (at least approximately well), otherwise why is this being enforced? Why is it the sticking point if 10 kyus aren’t having roughly even games with 14kyus with 4 stones, or 1 kyus having roughly even games with 10kyus with 9 stones. Why should this matter for the new 25-35 kyus say (if it goes that far).
This does sound good, get a 25kyu+ survey maybe 
5 Likes
Ten just wrote blog post Is the ranking system facing a crisis?
Its not especially about OGS nor low ranks, but about rank system in general. It makes me think that it’s wise to recheck and recalibrate all ranking system from time to time if we really want ranks to have any meaning.
6 Likes
But we would only need to do that if the purpose for ranks was something other than just determining handicap 
4 Likes
In principle OGS could try get ahead of the game with trying to implement an anti-cheat feature using ai to flag games or to assist in figuring out if a player used an ai - since half of that article is just giving out about Tygem players using ai.
But that’s for another topic. The rest of that article though is either addressing systems where there’s no rank demotion (like he says in Japan) which seems obvious that you could have a problem after a number of years, or real life/over the board (Aga, egf etc) rankings not being identical with online rankings, which again is obvious - people can play easier and more frequently online.
2 Likes
Actually I think the follow up article to that one is a (small) bit more relevant and also interesting 
Of course I didn’t expect there to be a follow up article at the time it’s abour handicap stones and komi according to katago - however I feel like katago’s opinion on the utility and value of the stones is kind of like a pro or pro strength opinion of stones/values.
https://www.nordicgodojo.eu/post/8
The question I would then pose (while I’m thinking about this) is what should the basis for handicap/komi be for all ranks/ratings - should it be
- that if pros played with these settings then they would win by x points and therefore a stone is worth x points
Or - We find that for amateurs in the range x-y kyu/Dan games tend to be more even (low scoring or 50/50 win rates?) if there is n stones or m points komi?
The latter actually suggests that maybe a dynamic handicap/komi could be more appropriate, that is one that changes with rating - who says that two stones is equally effective at 1900 glicko2 (or elo etc) as it would be at 1000 points (I’m avoiding saying kyu Dan since that seems to already imply the handicap works) or that 30 points reverse komi works the same for players near 2000 points as it would be at 800 points.
I presume the reason is that evidence says so? That it’s more or less the same across all rank ranges?
Or is it just something we study more the more we rank up and become more accustomed to the handicap games the more we play?
1 Like
I think they should implement ranks behind 25k
2 Likes
it makes me think i am the same rank as a 34k
But you are a 23k. I don’t understand your logic.
it says i am 2 ranks apart