OGS rank histogram (outdated)

I made dis:

Click on it for bigger resolution.

I downloaded some active players (I hope I didn’t make server sad with my requests) and plotted number of players for each rank.

Blue bars include all players.

Green bars represent only stable ranks (deviation < 100).

Orange curve is just outline for stable ranks.

Percents on the bottom show percent of the players weaker than certain rank (or exactly percent of weaker players + half of equal players), based on stable ranks.

So If you’re 32k or weaker, you’re roughly stronger than 0% of the server. If you get to 19k, you can beat 10% of the players. 10k is a median here. SDK players can brag about being stronger than most of the players. And dans are gods.

Props to 41k for keeping playing.




How many players did you sample? It looks like only a few thousand? Aren’t there many more active accounts on this site? How did you select which ones to sample?

1 Like

As long as they were randomly and independently selected (easily done with any bot / code) he only needs 20+ samples for statistically significant results.

1 Like

Well, it’s not a serious study of course. I just jumped through recent games. So I might have touched some timezones more then others.


Were you mostly sampling from live or correspondence games? Or did you get a balanced mix of both? I suspect that there are fairly large subsets of players that mostly play only live or only correspondence. It would be interesting to see if there is a significant difference between those two groups, or if those two cliques even firmly exist.

I’m not too sure what it would say if a statistically significant difference was found. Maybe those two cliques don’t really mix enough (via a third group of players that play both), so any difference could just be due to independent drift.

1 Like

Too complicated to separate probably. You’d need to look at each player, decide whether they play mostly corr or live.

I went through game randomly, so whichever kind gets to be played more was more represented.

I do wonder though whether 1k corr is the same as 1k live.


Thanks! Interesting!
Maybe OGS should have some automated procedure that regenerates this histogram every 6 months or so?
OGS has a lot of new players, and it would be interesting to see whether the mode of the distribution (the peak) moves to the right with time as players improve. (Assuming that the ranking system doesn’t keep the average rank of all users constant, as some broken ranking systems do.)
There’s a very outdated histogram for KGS here: https://senseis.xmp.net/?KGSRankHistogram


I hope you’re not implying that the decision to fit rankings within a normal distribution constitutes a ranking system being “broken”


Does it mean that ranks between 20k and 8k are calculated less accurate than other?

1 Like

At about 11k people start playing more 19x19 than 9x9. And no one plays 13x13, why do we even have this board size.


I know that was a rhetorical question, but I’m pretty sure the actual answer is that it is found by some to be a useful transition from 9x9 to 19x19, so it’s “worth having”. There are also some people out there who simply like variety, and they play other board sizes as well, which aren’t even on your radar :wink:

(I can’t get my head around one board size and the infinite variety there, let alone others, but to each their own!)




uhm … 14.3% is roughly one in seven games … definitely not “no one plays 13x13”, I’d say :wink:


Looks like someone still does it:

I suppose it’s a joke I did’n get…


Yes, I try to play 13x13 and know all too well that no one wants to play it. I bet that these 14% of players play 13x13 just because it exists as official ranked board size (and it has its own correspondence ladder). My point is that very small percent of people actually want 13x13, it’s more or less “hey, it exists, why not play it?” For example, if ogs didn’t have 19x19 as ranked board, people would want it implemented, if ogs didn’t have 9x9, people would want it. If ogs didn’t have ranked 13x13 (and it was just an irregular board size), no one would care in particular.

Edit: I wonder what numbers will we get if we exclude all ladder and automatic tournament games.


OGS by time controls


Most popular time controls:

System Description Percent
1 byoyomi 5m+5x30s ~10%
2 byoyomi 10m+5x30s ~7%
3 fischer 2m+30s/move up to 5m ~7%
4 byoyomi 20m+5x30s ~6%
5 fischer 30s+10s/move up to 60s ~4%
6 fischer 3d+1d/move up to 1w ~4%
7 byoyomi 10m+4x30s ~2%
8 byoyomi 15m+5x30s ~2%
9 simple 60s/move ~1%
10 byoyomi 5m+5x10s ~1%

And now filter by rank. :slight_smile:


You think it matters?

1 Like

I’d like to see a rank histogram of correspondence players vs. live players.

I think there’s a systematic shift (corr < live).


Apologies, because there’s an undocumented feature that exists for no reason whatsoever all ladder games were skipped in my code. Now that I included them, number of corr games is a bit higher:


That bumps up 3d+1d up to 3d in time controls rating:

System Description Percent
1 byoyomi 5m+5x30s ~9%
2 byoyomi 10m+5x30s ~7%
3 fischer 2m+30s/move up to 5m ~6%
4 byoyomi 20m+5x30s ~6%
5 fischer 30s+10s/move up to 60s ~4%
6 fischer 3d+1d/move up to 1w ~3%
7 fischer 3d+1d/move up to 3d ~3%
8 byoyomi 10m+4x30s ~2%
9 byoyomi 15m+5x30s ~1%
10 simple 60s/move ~1%

Now with a quick hack from size by rank chart we can get live/corr vs rank:

Notice that since I count players here, not games percent of correspondence is higher. Probably it means that corr fans play fewer games each, while live players pump out a lot of games.

I also wonder whether the way I download games affect this in any way. I go through game ids with a certain big step. And corr players often start games in a tournament simultaneously so games are created in batches. And my script can step over whole tournaments. Should be fine in the end, of course, but I keep my suspicions.