Change CA Elo Calculation
The current implementation of elo in CA needs fixes, especially at the higher ends. The recent change to give a 20% boost only made the existing problems more prominent.
1. The current system doesn't seem to take into account that every server already auto-balances based on elo. Assuming this elo system is in any way accurate, that should result in a 50% win rate. This is a problem because:
2. The punishments in the system aren't fair. You lose far more elo for being 1 performance point under the next highest than you gain for being 10 points over. And because the winning team gets a 20% boost, It's almost impossible to be the highest performance on the losing team.
In almost every loss, I'll lose 3-4x as much as I'll gain for a win, regardless of how well I do when I win. I'd need a ~75% win rate to keep a stable rating, which is pretty ridiculous with given every game is also elo balanced.
The solution is to go back to a straight w/l for calculating elo. Scale the amount won or lost by how far an individual player's elo is from the mean when they win/lose to avoid abuse. That was the old system and, to my knowledge, nobody had any problems with it - it just unfortunately existed before elo shuffling did. Scrap performance - it was a nice idea, but it just doesn't work.
1. The current system doesn't seem to take into account that every server already auto-balances based on elo. Assuming this elo system is in any way accurate, that should result in a 50% win rate. This is a problem because:
2. The punishments in the system aren't fair. You lose far more elo for being 1 performance point under the next highest than you gain for being 10 points over. And because the winning team gets a 20% boost, It's almost impossible to be the highest performance on the losing team.
In almost every loss, I'll lose 3-4x as much as I'll gain for a win, regardless of how well I do when I win. I'd need a ~75% win rate to keep a stable rating, which is pretty ridiculous with given every game is also elo balanced.
The solution is to go back to a straight w/l for calculating elo. Scale the amount won or lost by how far an individual player's elo is from the mean when they win/lose to avoid abuse. That was the old system and, to my knowledge, nobody had any problems with it - it just unfortunately existed before elo shuffling did. Scrap performance - it was a nice idea, but it just doesn't work.
I think the idea of rewarding a win is great but again the penalization is completely disproportionate for the losing when the performance calculation is done post their perf comparison.
For those talking about ELO shuffling that's not really an issue for QLStats, that's more of a MinQL issue, I agree with the issues raised but probably not for this post.
It was a legitimate complaint that many players brought to me when they said that matches were often thrown by individual players because they focused entirely on getting a high Perf value to win Elo and did not care if their team wins or loses.
The game's objective is to win as a team and the rating system should not change that objective. That's why the bonus was added and not because I don't like bater and campers or want you to play the game the way I want it to be played.
qlranks was the extreme opposite. No matter how good you played, if you were in the losing team, you lost Elo. Always. So nobody wanted to switch in a 6v4 match to even it out to 5v5. That is why I fully intentionally did not have a bonus/penalty for winning/losing when qlstats started.
I'm not saying the system is perfect as-is. 20% was a rather arbitrary pick and it can be tuned.
With more servers than before using mid-game shuffle it makes sense to reduce or even remove the bonus again. No matter what I do, I will receive complaints. Without winner bonus people may throw games unpunished and with bonus it is unfair for players who got shuffled around.
BOOM - encouragement for winning the match without disproportionate penalty.
This is fundamental, because the Glicko rating system can only be used to handle 1v1 scenarios. There is no way to update ratings of individual players if you operate on team totals.
What the winner bonus does: as a loser you have to deal 20% more damage to have equal performance as someone in the winner team, to call it a draw.
I'm not saying that you should introduce a new 'team elo' system. I'm saying that you should calculate all players perf ratings BEFORE any bonus values are applied, then re-calculate the perf for the players of the winning team and now include the bonus - this will avoid disproportionate penalties but still have an incentive to win the arena.
https://qlstats.net/player/8658#ca
Rating: 1639 ± 36 (ca, 56 games)
Rank: -
I see you have a rank now, so it should be fine.
I think for the team game ca they should make a new elo damage that goes to players that are not in range of at least one team mate during the first 3 minutes, it should be possible to check in stats there percentage of taken and dealt damage in the time line with team mates that some people deserve a fix. the damage should be like -100 this will change hiding forever and waiting till all possible damage by team mates is done and profit from all the easier kills you can make after your waiting game.
I would fix it to be like 3 games in a match below the team average in taken damage in the first 3 minutes of the games you get the punishment. because be fair you could jump in if your team mates get hurt more than you lets say ,off a -100 elo
plus the rest of the ridiculous thinks that go on it will make the game more equal in elo if you get awards for your performance like your the static idiot in the game
implementation wise i would go a different path like only count the first 3 minutes of scores and damages in matches so the game gets up to enjoyable speed again.. and more fair team wise in my opinion
PredatHor, please consider removing this bonus altogether, the 10% is still really punishing the losing team in comparison to how it rewards the winning team...take this game: https://qlstats.net/game/4901970 it makes no sense to me. i.e. if you look at the 2nd place guy on Red- has 1549 glicko and scored 57pnts in 14rnds but team lost so he is -27 glicko, the 3rd place player on Blue with a 1555 glicko scored 59 pnts in 14rnds team won so he gets +17. I see it as these guys are equal glicko, played equal time and scored about equal, winner gets +17 loser gets -27...that seems like the loser who performed just as well should not have lost any points because he had a bad team. skill rating should not be based this much on if your team wins or loses, especially when you have that large of a team. Thats the other thing, why don't 2v2 count when 15v15 does? this is a skill rating, 2v2 takes way more skill than all these spam fests of 10+ teamsize, please start counting 2v2?
if you have your VPS, you can deploy https://github.com/em92/quakelive-local-ratings
It is like mini version of qlstats: http://stats.eugenemolotov.ru