Rumor is the balance mod is based on win/loss ratio and in practice it doesn't seem to work well.

Win/loss requires lots of random teams and lots of matches to coverage on how good the player is.

Ideally the balance ranking would count new matches as more important than old matches and look at the players "score" so it coverages quickly.

I think this can be achieved by using a EMA filter (an exponentially moving average) of a blend of team ranking (K:D ratio) & win/loss.

For this kind of stuff you want to use normalized numbers, 0.00 ~ 1.00, and this is also convenient because the sorting is easy.

You can multiply by 100 to give people a score; e.g. if your score is 100 then you have come in first place every game and won every game.

If your score is 0 then you have come in last place and lost every game.

Another fairly important concept in this kind of stuff is "zero sum" - otherwise the points "run away" and people who have simply played longer get a higher score.

So each game is worth 1.666 points. The winning team gets 1.000 point (100%) and the losing team gets 0.666 (2/3rd) of a point.

The first rank player on each team gets a team multiplier of 100%.

Each rank down drops a certain percentage* until the last player gets 1/nth (where 'n' is highest number of players of a team - this gives a boost to a team that has fewer players).

Take the 1 or 2/3 and multiply with your team multiplier and that's your score for that game.

Take the score for this game and multiple by .25 and add that to .75 times your running total score. (This is the EMA filter.)

That's your new ranking which would be something like 0.346 which is your NS2 "batting average".

This would make you score bobble around a bit each game which we want to that repeated use of balance does not produce the same teams over & over.

*

Let's say it's a game of 6 vs. 7 so we use 7 for "n".

First place is a 100% multiplier on both teams (7/7).

Second place is 6/7, third place is 5/7, etc... down to 2/7 for the 6 player team and 1/7 for the 7 player team.

Win/loss requires lots of random teams and lots of matches to coverage on how good the player is.

Ideally the balance ranking would count new matches as more important than old matches and look at the players "score" so it coverages quickly.

I think this can be achieved by using a EMA filter (an exponentially moving average) of a blend of team ranking (K:D ratio) & win/loss.

For this kind of stuff you want to use normalized numbers, 0.00 ~ 1.00, and this is also convenient because the sorting is easy.

You can multiply by 100 to give people a score; e.g. if your score is 100 then you have come in first place every game and won every game.

If your score is 0 then you have come in last place and lost every game.

Another fairly important concept in this kind of stuff is "zero sum" - otherwise the points "run away" and people who have simply played longer get a higher score.

So each game is worth 1.666 points. The winning team gets 1.000 point (100%) and the losing team gets 0.666 (2/3rd) of a point.

The first rank player on each team gets a team multiplier of 100%.

Each rank down drops a certain percentage* until the last player gets 1/nth (where 'n' is highest number of players of a team - this gives a boost to a team that has fewer players).

Take the 1 or 2/3 and multiply with your team multiplier and that's your score for that game.

Take the score for this game and multiple by .25 and add that to .75 times your running total score. (This is the EMA filter.)

That's your new ranking which would be something like 0.346 which is your NS2 "batting average".

This would make you score bobble around a bit each game which we want to that repeated use of balance does not produce the same teams over & over.

*

Let's say it's a game of 6 vs. 7 so we use 7 for "n".

First place is a 100% multiplier on both teams (7/7).

Second place is 6/7, third place is 5/7, etc... down to 2/7 for the 6 player team and 1/7 for the 7 player team.

## Comment