If the OGS’s devs don’t mind I’ll try to make a layman’s summary of how rankings work for my own edification.
Please correct me where I’m wrong (which I’m sure I am).
First, We don’t care about kyu/dan. What we want is to give each player a number so that she/he can be matched with players with a similar number. I’ll call this the OGS number
At registration every user is given the same number: 1150. Then players play among them. Their OGS number goes up/down if they win/lose against stronger/weaker players.
Reasonably:
If you win against a stronger player, your number goes up
If you win against a weaker player, your number goes up but not so much
If you lose against a weaker player, your number goes down
If you lose against a stronger player, your number goes down but not so much
This setup will naturally sort every player after a number of games, resulting in an “empirical ranking” for the players.
Notice that this “algorithm” simply sorts the players according to win/lose.
Now, since there are different board sizes and time settings, each player actually have 9 OGS numbers instead of just one (a number per each board size and time setting) so when you win/lose a 9x9 blitz game only one of the numbers goes up/down. When you win/lose a 19x19 correspondence game another number goes up/down and so on.
One can make an overall number by taking the average of these 9 numbers (maybe weighted by the number of games in each type?)
Now, most players want to use the kyu/dan system because it’s cool and that’s what tradition wants us to use. So we need a way to map our distribution of OGS numbers to the 30k-9d scale. There are many many ways of doing this. For instance I can think of a couple of possible mappings:
- The strongest OGS player, with
OGS number=X is 9d. The worst player, with OGS number=Y is 30k. If you want to know your ranking, take your number, compare it with the best(X) and worst(Y) numbers, and see where you are in the scale (this would be a linear model).
Notice that we are randomly assigning the best player the 9d ranking (what if the best OGS players is considered a 20k by the rest of the GO world??!!)
- Some players play also in X external tournament, so go look their ranking in that tournament and use those relations to classify the rest of the players. This would be similar to the previous strategy, but instead of fixing only the best and worst
OGS numbers, you would have some other fixed numbers in between.
- Any other function that makes anyone happy.
- OP’s initial detailed explanation which statistically matches the low dan rankings with those of AGA/EGF and sets the mapping such that handicap stones in 19x19 are meaningful.
Notice that the mapping is somewhat arbitrary, and the only “measure of your strength” is the OGS number which compares you to other players in the server.
The mapping simply takes a OGS number and returns a kyu/dan number, regardless of the meaning of the OGS number inputted (9x9, 19x19,…).
Finally, as in any other server, the kyu/dan number provided by OGS it’s just your OGS ranking… play in other servers and compare… the differences among servers are not negligible!!
)
