New KGS Ranking proposal
- Sounds fine, but doesn't it seem a little to complicated with the whole point-handicap-thing-cart? Wouldn't it be easiest for online go servers to just do what tygem does (x number of games before you rank up or down)?
Here I'll try to explain my (lobo) idea of how my new ranking system will work & compare it with current (deeply flawed & beyond any repair in my opinion) KGS ranking system. After reading please use the discussion page.
Entry point of my approach is point based system similar to EGF. Players ranks are directly influenced by how many points they have:
Range Rank Range Rank
0-999 30k 20000-20999 10k 1000-1999 29k 21000-21999 9k 2000-2999 28k 22000-22999 8k 3000-3999 27k 23000-23999 7k 4000-4999 26k 24000-24999 6k 5000-5999 25k 25000-25999 5k 6000-6999 24k 26000-26999 4k 7000-7999 23k 27000-27999 3k 8000-8999 22k 28000-28999 2k 9000-9999 21k 29000-29999 1k 10000-10999 20k 30000-30999 1d 11000-11999 19k 31000-31999 2d 12000-12999 18k 32000-32999 3d 13000-13999 17k 33000-33999 4d 14000-14999 16k 34000-34999 5d 15000-15999 15k 35000-35999 6d 16000-16999 14k 36000-36999 7d 17000-17999 13k 37000-37999 8d 18000-18999 12k 38000-38999 9d 19000-19999 11k
Another crucial part of my plan is a table expressing points change for win/loss situation under various circumstances:
Rank difference Actual handicap handicap +5 +4 +3 +2 +1 0 -1 -2 -3 -4 -5
-5 200 190 180 170 160 150 140 130 120 110 100
-4 190 180 170 160 150 140 130 120 110 100 90
-3 180 170 160 150 140 130 120 110 100 90 80
-2 170 160 150 140 130 120 110 100 90 80 70
-1 160 150 140 130 120 110 100 90 80 70 60
0 150 140 130 120 110 100 90 80 70 60 50
+1 140 130 120 110 100 90 80 70 60 50 40
+2 130 120 110 100 90 80 70 60 50 40 30
+3 120 110 100 90 80 70 60 50 40 30 20
+4 110 100 90 80 70 60 50 40 30 20 10
+5 100 90 80 70 60 50 40 30 20 10 0
Notice how simple it looks and works: you know exactly how many points are at stake of a particular game and also how many more you need to actually advance!!!
This is where current KGS system fails the most. Depending on how many games you played, the more consecutive wins you need to advance. In result some people need to get ridiculous amounts of wins to actually change a rank and if they want to do it fast they are digging their own grave because the more you play the more you have to win!!!! At this point some people bring the subject 'if he's really better then current rank he should win easily' and they couldn't have been more wrong. Since wining percentage between two ranks is in the ranges of 60/40 in favor of the stronger player and our hero just wants to 'be' that one rank stronger he will loose some games in the process of advancing. Keep in mind though that if he does he is pushing his goal even further because the lost game counts towards not only rank decline but also to his rank 'confidence' that translates to winning more consecutive games to advance.
Having a system is not all, there are issues that have to be resolved with a set of 'rules' appropriate to a particular situation:
- A new player arrives in the system?
First of all this new player is an 'added value', it means that his participation in the system can't generate any 'cost' (by joining the game he doesn't steal some players knowledge/ability and in result he shouldn't steal his ranking points). That's why a game between registered player with [?] ranking and a stable [k/d] or semi-stable [k/d?] rank won't provide any points for the 'old guy' and assign a rank of the 'new guy' depending on his win/loss in 1 range of his opponent.
For the new player beings a 'trial' period in which system will try to assign him a best suited rank (indicated by ? after the estimated rank). To actually achieve this I'd like to use a diminshing multiplier depending on how many games he played. If it's his 2nd game his points from the Mother of all tables will be multiplied by 9 and decreased accordingly (3rd x8, 4th x7, 5th x6, 6th x5, 7th x4, 8th x3, 9th x2, 10 x1). This generates faster promotion/demotion for new players and is quite flexible solution. If a 'new 9d' plays [30k] and is now [29k?] he can quickly estimate that he should play much stronger opponent and thanks to the multiplier he will gain much more points thanks to his next win against a much 'stronger' opponent.
- Player with [k/d] rank plays a [k/d?]?
Stable rank player won't gain/lose anything in this game, he is just helping the opponent to gain a stable ranking. This way we encourage players to help others gain rank.
Well there you go, whole ranking system!! Simple isn't it? Yeah it is but it's not his biggest advantage. Let me explain its main features:
- Clear rules, you know exactly what is at stake of a particular game.
- Faster rank advancement/decline that provides a stable (not locally but server wide) game enviroment. Your opponents will have more accurate ranking, no more '1d consists of players at about four different playing levels'.
- No penalty for playing too much, people who do so won't change their ranking 'chaotically' throughout a day because 10 consecutive wins doesn't just happen.
- Thanks to faster rank propagation on server level (9d will demote a player weaker than him faster than in current ranking and then he will do so with 8d, 7d etc. down to 30k) we will once again have double digit kyu's that right now are squeezed in 10k-9d range.