I downloaded some game metadata of random players. Games ended from May to November, 100 latest games (for that period) for each player. And we can plot game lengths to see how long an average game takes. From downloaded games I excluded annulled games and games with width or height not equal to 19. And plotted histogram.
Very nice
What is the skill level (distribution) of the players?
There are way more live games than correspondence. I am in the minority 
I think it’s more likely that number of live games finishing would be fairly consistent whereas I theorise that correspondence games might finish in waves.
“100 latest games (for that period) for each player” should be able to get rid of any wave.
when i was playing corr, I’d have a few months where I’d maybe only finish a few games, and have other months where dozens of games finish. Maybe that was just me though.
It’s about quality not quantity though isn’t it? 
The vast majority of the life games are blitz it seems to me. I’d compare live games over an hour or more to correspondence quantities.
It’s the games in between 2 hours up to around 1 day or a bit more that fascinated me the most. How do people play such games?! Maybe “blitz correspondence” in between normal life? Or actually sitting down for a pro style game of Go?
I’m not convinced.
Where would you draw the line for “blizty”? A 20 minute game certainly is not blitz, so the blitzy line is to the left of the peak of the graph. Admittedly, there are also more live datapoints to the left of the peak, but it’s a big call to say “the majority are blitzy”. At best, its about 50/50…
I don’t think it’s unreasonable to say most of the 20 minute games would be 5 min main + w/e byo yomi… I was under the impression 5 min main was usually classed as blitz…?
Anything less than 30 mins per player seems fast to me! So an hour for the game. That’s about the cut off for EGF class C games, which I think might be called “rapid” so less than that seems blitz. But I know everything is faster on line than IRL! And getting old has a similar impact on one’s perception…
OGS’s default setting for “live” is 10m+5x30s Byo-Yomi.
The label “blitz” gets every game with time settings allowing for less than 10 seconds per move.
Yes, I suppose what I have in mind is a distinction between games allowing time for some thought and games faster than that. With my slow brain I feel much less than an hour per game would mean uncomfortably fast thinking for me!
This is a subset of players from this [Outdated] Improved OGS rank histogram
I think no matter how we count correspondence games will be in minority. Correspondence takes more commitment.
I like slow games. But there’re a lot of fast players out there. Many times I’ve seen review requests on games that last like 15 minutes and wondered why bother reviewing if a player didn’t bother to think about the game for more than that. Many times I picked a game for kibitz that turned out to be played blitz-like. This is why I figured to plot this graph - to see what’s the general speed of play in go.
I played live games that last more than 2 hours but these are rare. It’s more often that fast correspondence games when both opponents are online end up taking less than a day.
I also notice that longer time controls don’t necessarily produce longer games - players often play them like it’s blitz anyway.
Also lots of people resign on the first sight of trouble, and that’s why otherwise long game may take less time. Maybe we should plot only games that went into counting?
I’ll look into different time controls but I need time to figure out “time_control_parameters” field. It’s a bit of a mess.
Good point, I didn’t think about games ending in resignation being shorter.
Is that the reason why there is a mini peak at around 1 day.
Plotting 20m 5x30s - most popular live 19x19 time control still gives kind of the same shape. People don’t use their time.
Can you plot number of moves?
Can you use different colours for resign vs count score?
