if we ignore first 6 games where user didn’t have rank yet, then
chart:
manually count by using colors in dark theme:
15 losses vs stronger
8 wins vs weaker
6 losses vs weaker
manually count by using colors in dark theme:
15 losses vs stronger
8 wins vs weaker
6 losses vs weaker
I don’t think you’ve counted the right number of games. From the image:
6 + 23 + 11 + 4 = 44
From your manual counting:
8 + 15 + 8 + 6 = 37
Although, it’s already clear that manually counting has shown more losses vs weaker opponents than the image and more wins vs stronger opponents.
I see, the last page of games list is not 50 games
9 wins vs stronger
20 losses vs stronger
8 wins vs weaker
7 losses vs weaker
9+20+8+7=44
chart shows:
6+23=29 games vs stronger
11+4=15 games vs weaker
automatic counting of games list in Excel shows:
23 games vs stronger
11 games vs weaker
8 games vs equal
2 games vs [?]
23+11+8+2=44
Just a thought - are we (and is it) talking about opponent rank at the beginning or end of the game (which can be rather different for correspondence).
(I’ll probably move this to a separate thread when I have a moment, I’d love to get to the bottom of this, but it’s not really about sandbagging 
)
1 Like
I distinctly remember a past discussion of this in which it was authoritatively stated that the rank in the player’s history for each game was at the beginning of the game. If the pie chart corresponds to the history, then it should be the same.
If the ranking system were more stable, it might discourage more new players, as well as players who play less frequently, because they would see less improvement in their rank. Also, players would be stuck in the provisional zone longer (which is widely disliked).
Right, but the glicko thing has “high uncertainty” for new players and it (uncertainty) is supposed to decrease with more games leading to slower change.
You can actually see this in rank graphs, where the +/- boundary of uncertainty is shown.
(I’ve never understood why it doesn’t converge more than it appears to do.)
AIUI glicko is also supposed to increase uncertainty with less frequent play (a time elapsed between data points) element, but we don’t have that, for some reason I didn’t absorb properly.
I read this argument very often in the discussions.
Sometimes I wonder if this is a good approach. Of course it is important to get new players to the game. But maybe for a Go-Server it might be even more important to satisfy players who already knew the game.
I can imagine how this volatile rating annoys players who than decide to rather play elsewhere.
Ratings are also not fixed. There is no 12k standard (for example) that once you are beyond means you are 12k its just where the maths thinks you are relevant to other active players.
Its a bit like a tide.
If lots of new players join then ranks will inflate. If lots of players (especially lower rank players) leave then ranks (especially ddk) will drop even with identical skill level.
Or to think of it another way with a standard go distribution (which for ogs has more lower skilled players) if the bottom 50% stopped playing and werent replaced then (say all ddks) then in a few years a lot of sdks would drop to ddks and some dans to sdk as the distribution reasserted itself. I ran an ai sim of this recently to prove the point.
A final way of viewing is i started before covid and my main account is 11. I cant get this one above 13 but i think thats because more people played back then and numbers have dropped as opposed to me suddenly sucking so the skill level of lower ddks is higher.
Clear as mud eh
1 Like
I think rating inflation / deflation is an interesting topic, but I think is not directly connected with sandbagging. Only in the sense that people new to OGS think they “should be able to beat a 3k easy” and get frustrated and may think they get sandbagged.
I think we should focus here on people purposefully sandbagging, meaning manipulating their rank to appear on a different level as they are.
There are problems in the ranking system, like how new player influence the rating system, drift, or that Glicko has some problems when people don’t play a somwhat distributed cohort of players and instead play most of their games against much stronger or much weaker opposition, but I think they are miniscule compared to people purposefully messing with it.
3 Likes
Well, not just “new to OGS”, I think people who have been playing Go for many years and are high-dan on Fox are still not able to wrap their minds around the fact their ranks will be lower on OGS, and thus they will occasionally lose even games to OGS players who are ~6 stones weaker than their Fox rank.
1 Like
Well, we have to take in consideration too, that high-dan players in fox, has no playerbase against to play on OGS. Once you reach certain rank you move to Fox to be able to get a game in short time.
There is a lot of discussion about this but has nothing to do with sandbagging at all
And I’m perfectly happy to play someone with multiple stones on me, but because of that something I run into a lot is people whose rank is obviously deflated seeking still lower-matched players, which means that instead of playing someone with 5 or 6 stones, I am playing someone with 10 or 11 on me. This is often the difference between a likely loss where I can follow how I’m losing and learn something and getting incomprehensibly crushed before midgame.