Improvements to elo calculation

[nzyuzin]

To rank players in ctf2 games it may be beneficial to match players against not every other player on the field but only against the opposite team players considering the average levels of both teams.

So it won't be easy to get high rank when the games are seriously unbalanced.

[needs]

Server state doesn't contain the team players are on. However, elo rating system is made so when there is too much skill differences between two players, the game doesn't count. Actually, when there is a difference of more than 800 elo points, the winner won't get any points, and the looser won't loose any too.

So in the worst case, the strongest team farm the weakistest one until there is a difference of 800 points. Plus the farming will be long because the more difference there is, the less point you will win (or loose). That's not a situation we wanna see, but is there a way to avoid it?

[nzyuzin]

There are developments in the field of rating calculation for team-based games. The most famous algorithms, however, are patented and are not fully known.
https://www.microsoft.com/en-us/research/wp-content/uploads/2007/01/NIPS2006_0688.pdf

However, I'm sure that there should also be more open sources of information on that.