🌿ELO System Updates

[Stroom]

It's another good example for why elo in the current form is not "a system to gauge all players, all the time" - your own words.
It's an example for why elo is a pointless metric. It's a very rough metric.

Absolutely it's a rough metric. To make it even rougher is not the answer. It's doing the best it can.

ELO as an eco-system. Players gain/lose by the same amount. The overall numbers are meant to stay equal.

The decay, and frequent player pairings will simply take away numbers in quite a large amount. EVERYONES ELO will decrease - to what end?
Personally speaking, and there are a lot like me. All my ratings will in the future hover around the 200 mark, since I currently rank 200-500 in 70 odd games.

You won't be able to tell anything about my skill level in any of the games.
Where it's 500 in Innovation now, it will be 200 in a year or 2.
Where it's 200 in haggis, it will be 200 in a year or 2.
That will be a catastrophic failure.

I'm a better Innovation player than haggis. And I still will be in 2 years time.

As more players join the system, they will bring in points... but if they stop playing forever after having given some of those points to active players, the system will gradually deflate the elo score. Or if a player reaches a high elo and stops playing, these points can't be taken back and given into the system. So there is no real balance within the system this way. Only by individual game outcomes. But the actual elo score does not represent a fixed skill level that is comparable in time A vs time B. You could fairly compare scores of the active players. Inactive player vs active player score is not comparable. So a decay placed upon inactive players would do nothing for you when you compare elos. If you use something like glicko, you can have a system where elo is not blindly removed form the system but the display score is just lower for inactive players. If you are afraid of elo inflation/deflation...

Decay is not muddying it up. It is slightly punishing players for inactivity. Only for inactivity. But elo scores at current time will still be roughly comparable. 600 going down to 540 - not a big deal.

If you don't play and your elo decreases... boo hoo. Who cares? Just play a few games and your elo naturally rises back up. You used to be a good player so prove it again.

That's why I ask why do you care about elo. Are you afraid that you can't really climb back up? And if you have not played and have let your elo decrease, why do you think that your skill is comparable to those who are still actively playing?

Elo is a rough metric. Losing it is not a big deal.

[Patrick of the Isles]

So the problem is with 10% decay. What if it was 5%? 2800 going down to 2280 seems more reasonable.

Expecting a 2800 rated player to play at a 2280 level after a single year of inactivity on a particular site would not seem reasonable to 99% of chess players. Gary Kasparov retired at that rating 17-18 years ago and would almost certainly have the ability today to play at a 2280 level. Anything more than a 0.5% annual decrease would seem totally disproportionate to my experience with most games of skill.

[Stroom]

Stroom, you barely play anything else than Hanabi. So it makes sense that you can't process the reasons why people who play many different competitive games and worked their way up in those want to keep that proof of their past on the profile even if they don't play much anymore. Why people want their opponents have a good rating when they are good players and not have them 200 points less than they are by skill. Why we want to see valuable information on our friends profiles. Why we want the system to work and not leak points into nothingness leading to even more inflation.

You are right with glicko would be good and showing the difference of trueskill and uncertainty. It would work well for both our goals. But we don't have glicko and plain decay is not the next best thing. Saying glicko would be good does not defend the implemented decay.

I play Hanabi on BGA and complex board games in real life. No big deal.

What proof of skill is in your current elo? It does not say what your highest has been. And as I have said before, elo from 2 years ago is not necessarily representative to the elo scores of players who are actively playing. The system has changed a lot in the past as well. What value is in keeping a pointless score this way?

You could just look at the player's elo rank evolution chart to see the highest value he has had. The current value is not a representative of the highest skill the player has had. If you want, maybe there could be a different value next to your name of what your highest elo has been through the years. But trying to keep the current value as some kind of indication of "hall of fame" score is not really meaningful.

Someone who has not played for a long time is not necessarily as good as he was back then. So having a decay/uncertainty would be something worth trying. If it doesn't work out, it would not be that hard to reverse it.

Either way, give it a chance and see what actually happens.

[RicardoRix]

If you don't play and your elo decreases... boo hoo. Who cares? Just play a few games and your elo naturally rises back up. You used to be a good player so prove it again.

That's why I ask why do you care about elo. Are you afraid that you can't really climb back up? And if you have not played and have let your elo decrease, why do you think that your skill is comparable to those who are still actively playing?

Elo is a rough metric. Losing it is not a big deal.

Complete misunderstanding.
Just re-read SIlene's reponse, it's the best succinct answer.
Nobody is trying to prove anything. We are simply trying to restore the integrity of the ELO system the best we can.

[Stroom]

So the problem is with 10% decay. What if it was 5%? 2800 going down to 2280 seems more reasonable.

Expecting a 2800 rated player to play at a 2280 level after a single year of inactivity on a particular site would not seem reasonable to 99% of chess players. Gary Kasparov retired at that rating 17-18 years ago and would almost certainly have the ability today to play at a 2280 level. Anything more than a 0.5% annual decrease would seem totally disproportionate to my experience with most games of skill.

But... they can climb back up. In BGA most games are not really that hardcore as chess anyway.

[Silene]

If you don't play and your elo decreases... boo hoo. Who cares? Just play a few games and your elo naturally rises back up. You used to be a good player so prove it again.

I am actively playing Stone Age about twice a year. It's the amount of practice that I need to keep my skill sharp. Also I am actively playing other games and am generally around.

I wouldn't mind if my shown rating would decrease through showing of a glicko-difference because then I won't hurt my opponents in those 2 games and my way back up would be quick if i return to playing more. Big difference to just reducing my score with nothing in return (no speedy adaption after return, no protecting my opponents).

[Patrick of the Isles]

In BGA most games are not really that hardcore as chess anyway.

I don't mean to keep the discussion going beyond the point of adding new useful information, so I'll finish my contribution by saying that for me the bottom line is that the site will be less enjoyable for me if this change is not reversed, and it's good that the site decision-makers are aware of that fact. It sounds like the same is true for others in this thread. I hope that site leadership will reconsider for that reason. The only reason most of us are here is to have fun. Fundamentally in my view if many of us will have less fun with the changes, then they aren't good changes, no matter what arguments for or against are made.

[Wreckage]

In 2017 and earlier, the top players in 6 Nimmt reached about 850 ELO. Today the best players reach about 600. (There are several ranked higher, but they are doing things….that’s a whole different discussion). 20th place was about 750 ELO then. Today 20th place is about 525 ELO So players today with the same amount of success have about 200-250 less ELO than they did 6 years ago. The K factor was 25 then, compared to a K factor of 20 now. If an old player returns to playing, their old score doesn’t reflect their current skill, and gives them a huge unfair advantage.

Another example:
In Hanabi, the rules changed a few months ago regarding abandoning games. As a result, the average player that played 1 year ago gained more ELO than the average player does now.

Another example:
On this thread, I think there is general agreement, that ELO will be gained at a different rate after these proposed changes.

Scores from before a system change don’t accurately reflect skill compared to scores after a change. There are two different topics being called decay here. One is about time decay, and the other about the new proposed - less ELO from pairing with the same opponents (Which I agree with most of you, isn’t good).

I get the arguments that we have arena for competitive players, but that’s not true for all games. In 6 Nimmt almost none of the best 50 players compete on arena because the settings aren’t right for higher skilled players. So the leaderboard is all we have. Except the few leaders at the top who don’t actually play games with the rest of us, kind of ruin it.

[Silene]

So players today with the same amount of success have about 200-250 less ELO than they did 6 years ago.

If this is the case then it means points are leaking out of the system (maybe because new players start at 0 rather than somewhere in the middle).

If that is the case then the last thing you should do is find more ways to leak points out of the system. And now 2 of those were implemented. You should instead try to find more ways to feed points in. Or fix the leak (suggestion: reciprocal effect of k-factor ==> don't get hit so hard when a new player beats you)

If a player turns inactive, their score isn't hurting anyone. Let them keep their hall of fame profile. If they return, they're gonna donate their points to others, which is good. Better than having them feed on others and then later puff those points into nothing again on their next inactive season.

[6element]

To sum it up, while the intent of the change was good, there are at least two serious issues with its design. Luckily, they can be fixed by just changing a few numbers.

Rating decay affects way too many people, and the effect is totally negative. Folks who used to play game X, got their Elo to 250 or 350 or 400, but nowhere close to top 100, will eventually get their Elo reset to 200. Now their Elo is artificially lowered misleading their potential opponents, and also they need to "prove themselves" if they decide to resume playing game X for no good reason. This serves no purpose (if these users paused playing more than 60 days ago they were not in the leaderboards, and their Elo was quite low comparing to top 100), just annoys one people and misleads others. Probably limiting decay to 500+ Elo (or to folks who are in the top 20) would solve the issue. Alternatively, changing the decay to 0 and changing "the delisting time" from 60 days to 20-30 (or even 15) solves the problem with "parking on top of leaderboards" without introducing issues.

The second issue is seemingly infinite memory ("no cooldown") of the Elo correction applied when playing against the same opponent. Number of possible opponents is limited and eventually this is going to become annoying. For example, I enjoy playing with certain opponents and while games against each of them don't represent more than 1% of my total games, eventually the new system will teach me to play less and less against them. So it just reduces fun. I suggest setting the memory limit to 48 hours (so after 48 hours of not playing against certain opponent there is no Elo correction), or applying it only when games against certain opponent represent more than ~5% of total games of a player.