🌿ELO System Updates

[miklomike]

I just want to reiterate how poorly these implement in different scenarios.

Games with small player pools, especially those with luck based games and large tables. Playing against 8 opponents, most of us have played a number of games against each other as there is not many people playing (this goes for most games in beta). Unless you come dead last, it will count as a win against players. Placing 3rd means I now have a win streak against 4th-8th, negating the points I should rightfully earn if I placed 1st next game.

In beta games, again with high player tables and large luck factor, it is impossible to play against people you have not played before simply because there are not enough players. It will get to a point where no one can get above a certain elo because wins will gain nothing and a loss will lose more than winning again after.

[LesserJester]

I just want to reiterate how poorly these implement in different scenarios.

Games with small player pools, especially those with luck based games and large tables. Playing against 8 opponents, most of us have played a number of games against each other as there is not many people playing (this goes for most games in beta). Unless you come dead last, it will count as a win against players. Placing 3rd means I now have a win streak against 4th-8th, negating the points I should rightfully earn if I placed 1st next game.

In beta games, again with high player tables and large luck factor, it is impossible to play against people you have not played before simply because there are not enough players. It will get to a point where no one can get above a certain elo because wins will gain nothing and a loss will lose more than winning again after.

Not just beta games, any game with a relatively small pool of players.

[jeroenh]

I just want to reiterate how poorly these implement in different scenarios.

Games with small player pools, especially those with luck based games and large tables. Playing against 8 opponents, most of us have played a number of games against each other as there is not many people playing (this goes for most games in beta). Unless you come dead last, it will count as a win against players. Placing 3rd means I now have a win streak against 4th-8th, negating the points I should rightfully earn if I placed 1st next game.

In beta games, again with high player tables and large luck factor, it is impossible to play against people you have not played before simply because there are not enough players. It will get to a point where no one can get above a certain elo because wins will gain nothing and a loss will lose more than winning again after.

How is this a problem? If we're looking at the gold standard of ELO rating (chess) then you have a situation where there is a reasonably small number of players playing each other all the time, and there is a limit to the maximum ELO rating.

Even then, saying that the BGA ELO system malfunctions in a small beta game is not really a good argument to say that the whole system is broken.

[Eocene]

I just want to reiterate how poorly these implement in different scenarios.

Games with small player pools, especially those with luck based games and large tables. Playing against 8 opponents, most of us have played a number of games against each other as there is not many people playing (this goes for most games in beta). Unless you come dead last, it will count as a win against players. Placing 3rd means I now have a win streak against 4th-8th, negating the points I should rightfully earn if I placed 1st next game.

In beta games, again with high player tables and large luck factor, it is impossible to play against people you have not played before simply because there are not enough players. It will get to a point where no one can get above a certain elo because wins will gain nothing and a loss will lose more than winning again after.

How is this a problem? If we're looking at the gold standard of ELO rating (chess) then you have a situation where there is a reasonably small number of players playing each other all the time, and there is a limit to the maximum ELO rating.

Even then, saying that the BGA ELO system malfunctions in a small beta game is not really a good argument to say that the whole system is broken.

Chess is being played by millions of people worldwide, I'd say that's not a small amount of players. And even if you only look at the top-ranked chess players, yes, they mainly play each other. But there's those millions of other players that, directly or indirectly, exchange ELO points with these top players. Also, there's no limit to the maximum ELO rating in chess.

As to why it's a problem: the system as currently implemented continuously leeches elo points. If someone beats another person two or more times in a row, that's elo lost that will not be replenished unless a new player starts playing the game. If someone doesn't play for a few months, that's elo lost that will not be replenished either. This essentially means that the amount of total elo available to players starts decreasing more and more over time.

Apart from that issue, there's the issue of fairness: why should someone be punished for beating someone twice or more?

This is not limited to small beta games. It's just the games where this issue is most visible.

[Stroom]

Chess is being played by millions of people worldwide, I'd say that's not a small amount of players. And even if you only look at the top-ranked chess players, yes, they mainly play each other. But there's those millions of other players that, directly or indirectly, exchange ELO points with these top players. Also, there's no limit to the maximum ELO rating in chess.

As to why it's a problem: the system as currently implemented continuously leeches elo points. If someone beats another person two or more times in a row, that's elo lost that will not be replenished unless a new player starts playing the game. If someone doesn't play for a few months, that's elo lost that will not be replenished either. This essentially means that the amount of total elo available to players starts decreasing more and more over time.

Apart from that issue, there's the issue of fairness: why should someone be punished for beating someone twice or more?

This is not limited to small beta games. It's just the games where this issue is most visible.

Elo is quite a pointless metric when you look into how it works. It depends on active player count (which constantly changes) and that all players who have entered, would keep playing (but many quit at some point, taking the points with them). The actual calculation of points per one game can be zero-sum in order not to inflate or deflate the value... but the value is inflating or deflating in other ways anyway.

The elo system is essentially a pyramid scheme. The more players you have, the higher someone can climb. If you only have 2 players who started at 1000 elo, then one player could get a max of 2000 points. With 3 players, it's 3000 etc. But this is not going to happen in practice. Note that in BGA the initial elo scores are working differently.

What actually happens is that there are lots and lots of players with below average elo, some players with medium elo, less players with high elo and very few with super high elo. In order for one group to gain any elo, they have to play with those who are one step below them and win, leeching elo for themselves. And then they lose to the players in a group higher than them.

Over time, as more players join the system, they bring fresh points into the system, which eventually allows the highest elos to gain more points (pyramid can grow if the foundation is wide enough). Similary, lots of players will stop playing the game and their elo is essentially cut out of the system (The players at the top can't gain more points by playing other players - better not to play or you would start falling off).

So, the system itself is not really as zero-sum as people think. Sure, each individual game kind-of is, but over time, it still either increases or decreases what the "active player base" total elo sum is.

So... elo over time is not really comparable. It doesn't really matter whether the system itself is adding or removing points into the system per individual games as long as everyone gets to play with the same rules. What you can compare is whether one player's current elo is higher or lower than another player's current elo... given that both players are actively playing the game within the same group and not choosing who they play with. If someone is taking a break, his old elo is not necessarily comparable to new player elo as the active player base might be much smaller or larger than the last time that player was actively playing.

I guess that BGA just wants the leaderboards to be more dynamic so that you would have to constantly prove yourself to the other active players. I am fine with that. The current approaches are not the best solutions to get there, but you can't really find the best system without actually testing it in real life. But overall, even with the new changes, the active players would still have the same ORDER in elo scores. It might be 100-200 lower than before but everyone will be lower. So overall it won't really matter. It's not about the score. It's about the comparison of one player vs another player.

[h_illes]

Before the changes, ELO was created at the bottom of the rankings due to the lower cap at 100, and ELO was taken away by players who stopped playing. But from the points created at the bottom, very little trickled up to the top of the rankings and very slowly, so, together with the ELO taken away by players who stop playing, there was still a balance. Consider e.g. FIDE ratings in chess to see the system works well and is comparable over time.

After the changes, ELO is leaking out of the system at every level, but more so at the top where winning streaks are more common. This effect will not be balanced out by the points created at the bottom, so everyone will go down in rating.

Essentially, whatever balance and comparability the system used to have is gone now due to the changes.

Also, since there seems to be a ~30 day cutoff when counting winning streaks, players who play relatively few games (e.g. 1 game per week from a given game) can avoid the penalty, while players who play more (e.g. 10 games per day) are penalized for winning. Why does the system promote playing fewer games?

[6element]

Also, since there seems to be a ~30 day cutoff when counting winning streaks,

How do we know it?

I thought Elo stealing is only applied to the games played after December 12.

[jeroenh]

Okay, I misspoke, there is no literal limit on chess rating, but in practice there is. Magnus Carlsen, the current highest rated chess player, has reached a FIDE chess ELO rating of 2800 in 2009. His current rating is 2830, and his rating has fluctuated around that number over the past 10+ years.

I can understand that it feels good to have a rating increase when you are winning, but ELO rating does not work like that

[Stroom]

Before the changes, ELO was created at the bottom of the rankings due to the lower cap at 100, and ELO was taken away by players who stopped playing. But from the points created at the bottom, very little trickled up to the top of the rankings and very slowly, so, together with the ELO taken away by players who stop playing, there was still a balance. Consider e.g. FIDE ratings in chess to see the system works well and is comparable over time.

After the changes, ELO is leaking out of the system at every level, but more so at the top where winning streaks are more common. This effect will not be balanced out by the points created at the bottom, so everyone will go down in rating.

Essentially, whatever balance and comparability the system used to have is gone now due to the changes.

Do you have any proof of elo leakage or is it just speculation? How much elo was actually entering and leaving the system before these changes?

If there is a constant influx of new players, the new plateau of max practical elo is just lower than before. Or maybe the developers just have to analyze what is happening to elo per game for active players and then adjust something. Gradual changes etc. Maybe introduce something that gives more elo into the system. Still, at the end of the day you are just comparing your current elo to someone else's current elo. If you say no to every new change then there will be no changes.

Also, most games here have quite low player bases compared to chess platforms. They will always have quirks of some kind.

[h_illes]

Do you have any proof of elo leakage or is it just speculation?

There is a penalty for every win-streak game. Before the changes, there was no penalty. That's the leakage. The actual penalty value is displayed per game, if that's what you need as proof.

If you say no to every new change then there will be no changes.

There have been plenty of suggestions in this thread. Just to repeat the one I like most, make the K factor decrease with a winning streak instead (so the ELO is still zero-sum).