🌿ELO System Updates

[Stroom]

That's no proof tho. Just your general opinion.

Yes it is proof. It is called logical reasoning. It's never too late to improve and educate yourself.

Common sense would say that they are just decreasing the K factor for consecutively winning against the same person.

Yes, I agree! That's why many have been suggesting it. The trouble is, it's not what's implemented right now. You don't seem to understand what is going on.

Edit: Seriously... most players here are strongly negative and start to echo chamber off of each other's posts... without reading the original post anymore.

That's pretty rich. You are the one who doesn't understand what's going on, and have not made any effort to make a logical argument. I suggest you dedicate some of the effort you put into writing to reading and comprehending instead. Or, if you are trolling deliberately - please stop.

Proof is a concrete thing, not opinions. From what I read from OTHER user comments, sure, their system is flawed. But it is such a small amount that by the time the season ends, it seems to have only marginal effect on the system.

All the comments only say to remove this change. Not to improve it. Only about "Boo hoo, I'm losing elo now". The system has barely started working and there have been no-one actually coming out to say that he is now significantly worse off than before.

[Stroom]

In any case, the moment any people from an insular group plays the outside world generally it quickly corrects. Your problem is very marginal.

It matters a hell of a lot if ELO over time is not comparable. Some part of this might not be avoidable but it's insane to intentionally introduce this.

It also does not affect people equally. The more you play, and the more you win(streak), the worse you're hit by far.

Largely because of this, it doesn't better assume skill of players. First of all ELO decay is way way too big, if the concept of real decay even is worth it. Also ELO decay in real terms makes things less accurate(as other people have mentioned constantly). You can have "fake" ELO decay as in vacation penalty, however matchmaking and "hidden MMR" should not decay ever. Second of all since winstreaking really hurts, people who don't winstreak will have artificially relatively higher ratings than those who do. In the case where the population is small enough winstreaking is unavoidable, there will be less ELO over time so it trends toward 100 making it even less good of a measure since everyone will be close to 100 anyway.

Doesn't really correct much. 10-20 points per game. You'd need lots of interaction for all groups to get on the same baseline.

ELO comparable over time... well, already the system is messed up. Different player count can change the game a lot. Someone only playing 2p games could get a much higher elo than someone only playing 4p. Or vice versa. So the scores aren't really that comparable. Or if you have a bunch of players entering the system, then losing a bunch of games and then stop playing, the rest of the players elo is actually inflated now. So it is not comparable. Average elos from only active players 5 years ago vs elos now are just not comparable. And that is OK. You only compare elos of active players today. 100$ 30 years ago wasn't the same as 100$ today. Same thing.

With elo, similarly to get to the very top, you would have to play much more to get there. If they'd correct the power play to K-factor change, it would probably work better. Over several games, if you always beat another player anyway and the factor is near 0... if you occasionally lose to them and the combo resets, you can just win next time and gain the previous loss back anyway (assuming similar elo rating).

Elo decay similarly... I proposed to use something like Glicko instead. If it just naturally keeps taking the top players down for not playing and gives them points back later, it's fine. If they are just being inactive, their loss of elo does not matter to anyone. If they come back, they can easily play a few games and win all that elo back.

If you say that climbing the elo ladder is hard then maybe you earned that elo by gaming the system not by just playing with your skill? In which case that change is also working as expected. Sure, the 10% is rough, but you should understand that it only applies to inactive players. Active players aren't affected. Those who actually matter.

So much fearmongering.

[anykidrated]

When the season ends and new one begins, you start over, right?

No, you don't start over lol.

So... let's just ask them to only change the calculation. Not to completely remove something that in general seems like a good idea to stop the system abusers from transferring elo points from one account to another.

Doesn't stop this and it's such a minor irrelevant problem anyway.

[Trimpster]

Dear BGA, Please prioritize changes suggested and highly-upvoted by the community. The community would appreciate it, would feel that you're actively listening to them & respecting them, and it would be a win-win situation for everyone.

PSA: Stroom is a troll, I'd caution people from engaging them or taking them seriously.

[Ranior]

PSA: Stroom is a troll, I'd caution people from engaging them or taking them seriously.

Agreed, it would be nice if the admins did any moderation for obvious bad faith actors in meaningful discussions, but I'd love the admins to start with any response at all.

Which on that front, how have you not yet reverted the insane power leveling "fix"?

I'm all for finding an actual solution that will help alleviate bad actors, but right now the solution you've implemented is far far worse than the previous issues of a few bad actors exploiting the ranking system.

For anyone who finds it a compelling argument that everyone is on the same system now, so it doesn't really matter if everyone's elo drops a bit--it's absolutely NOT true everyone is on the same or equal footing.

The power leveling change negatively impacts multiplayer players far more than it hurts 2p players. In a 2p game there's just the one comparison. You only play one opponent. In a 4p game there are six pairwise comparisons, you have three opponents in each of your games.

A 2p specialist who plays 100 games will be far less likely to run in to any win streaks, and therefore their elo will be minimally impacted by the power leveling change. A 4p player who plays 100 games is much more likely to run in to various win streaks against opponents, and will be impacted in a much greater fashion by this elo decay.

This means in the long run, elo becomes less accurate as a measure of skill, because it will be conflating skill along with player count preference. Those who prefer to play games at a higher player count (7p 7 Wonders specialists) vs those who prefer to play at a lower count (3p) will find their elos hurt more, so elo will partially become a number expressing skill, and partially a number expressing player count preferences.

I'll repeat: Elo will partially become a number expressing player count preferences, as well as skill. That's obviously not a desirable outcome.

Again, I'm ultimately against most of these changes precisely because it does not seem to make your rating system more accurate. In so far as we're going to have a rating system and be able to filter for opponents based on skill, that number should be reflective of said players skill. I get that some were abusing the system and finding ways to make their skill appear higher than it should--finding a fix for that would be nice--but the current change is going to lead to dramatically worse estimations of skills for many more players than what was previously going on, which means this change is a complete failure of what it's trying to achieve.

Why therefore have you not reverted it for now while you work on a better plan??

[Stroom]

When the season ends and new one begins, you start over, right?

No, you don't start over lol.

Maybe it should, then. So everyone has equal chance to get to the top within a limited set of time. Rather than someone who has already hit the top, playing as few times as possible to keep being in the top every season...

[franko]

Your incentive to play is to... have fun. That's all there is. Whether your elo rises or decreases is meaningless. You are supposed to keep playing in order for elo to actually display your skill level compared to other active players. Elo is not supposed to become stagnant when you don't play. Elo is constantly inflating and deflating due to changes in the system. A score of 600 is different from a score of 600 5 years ago. Even if player skills were the same. So it makes sense that it decays if you don't play for a long time or don't play with different players.

Sure, then we should just remove ELO altogether from the site if the only reason to play is to have fun. If ELO is meant to be reflective of skill, then it should be implemented in a way that will reflect skill. The changes around "boosting" have entirely broken ELO's reflection on skill because there's an indefinite time horizon on what they consider a match. I have "win streaks" because in the last 12 months I've beaten an account 5 times, meanwhile I've played ~2000 tables of this game. That's not a win streak or a case of boosting by any reasonable definition, so the fact that their change applies to those games is what I've been very clear is broken.

I've also proposed a very simple fix, add a time horizon (# games or time) to when a "streak" resets. Or, simply add this inability to gain ELO in the only times where it is not zero sum (wins against 100 rated players, as they can't lose ELO).

Many players here complain that elo changes make scores lower... as if elo is supposed to increase indefinitely. But it would be the same system for everyone. So it is evened out.

ELO doesn't increase indefinitely, it was already zero-sum (barring wins against <100 ranked players). If I played 100000 games against a particular player, our ELOs would find the equilibrium point of what our ELOs should be. If we both started with 300 ELO, then the equilibrium would stay at a total of 600 ELO split between the two players (but again, in whichever way reflects the actual skill difference). In the new system, if I played 100000 games with someone and we both started with 300 ELO, we would end with far less than 600 ELO between the two of us, as instead of just modifying the K-factor the change as implemented "deletes" ELO during win streaks. In the most degenerate case, i.e. an equally skilled opponent where we would have previously ended at 300 ELO after those games, in the new system we'd both end at 100 ELO.

I think that's a fairly clear and obvious "proof" as to what we mean by this system being broken. The new system only grants winners of a streak a decreasing % of the ELO they would have previously won, while still removing 100% of what is lost. So in the past you'd see ELO_win - ELO_loss = 0. A zero-sum game. Now you have ELO_win * streak_factor - ELO_loss <= 0. It's zero sum for the first game (streak_factor = 0), but for any game thereafter it is net negative on ELO (as streak_factor >= 0). If you take this and look at any game, it means the EV (expected value) of ELO for a game is net negative, which is why I keep saying that it literally disincentivizes playing the game for ELO. The expectation over time is that even the best player on this site will get to 100 ELO at some point after enough streaks delete all the ELO in the community pool. That's nonsense. Sure it might take an insanely long time for some games, but there are plenty of niche competitive games on the site where this will meaningfully depress scores without much time. For Kingdom Builder, I will guarantee this means no one will ever be able to hit Master again. That might not be important to you (and arguably not for me as I've already done it), but there are community members that do have enough skill to technically hit that rank, and because of these changes they will never be able to.

The rest of your comment just explains how elo works... Same as now. And random games are random anyway. Over time, the elo diff still will even out. Show actual mathematical proof that it would be worse rather than bandwagoning with everyone

You did not seem to try to comprehend the rest of my post, so I'm not sure how to respond to this. It showed that I would gain 0 ELO for beating a player who has ~620 ELO right now (even though I have ~590 ELO right now). That is not how ELO works today, that is explicitly what the change has done, and it sounds like you're somehow arguing that makes sense because... me beating a player better than me shouldn't give me any rating at all?



Also to conclude, you seem to be trolling by deliberately misconstruing others posts. I don't think the ELO Decay change is "bad", I think it's entirely meaningless. I gave an example of the #1 ranked by ELO player for Kingdom Builder who the community knows is parked in that spot. The decay change does nothing to stop their parking as it requires you play 1 game every 3 months.

[franko]

When the season ends and new one begins, you start over, right?

No, you don't start over lol.

Maybe it should, then. So everyone has equal chance to get to the top within a limited set of time. Rather than someone who has already hit the top, playing as few times as possible to keep being in the top every season...

You're describing Arena Seasons. They do reset every season, and you have to re-climb from a blank slate. That already exists. None of the changes in this post are to the Arena System.

*Moderation Edit: please do not use the Forums to call out specific tables or players by name, this is what the Moderation Reports are for. The Forums should remain a general discussion.

[flying_halflings]

Hi! In this thread you have people talking about a huge variety of different types of games: 2-player vs multiplayer, cooperative vs competitive, 5 minute vs 90 minute, predominantly luck based vs not...

Is "one size fits all" an underlying problem here?

Could changes like these be instead looked at on a per game basis in the same way Arena settings are? This would allow different game communities to implement changes:
(1) faster and with easier discussion and consensus
(2) with solutions that can be much more targeted to their specific problems
(3) without causing issues for other games that don't share the same problem or where because of the different type of game another solution would make more sense

This would also mean that reputation should change to be on a per game basis as well.

[Stroom]

Sure, then we should just remove ELO altogether from the site if the only reason to play is to have fun. If ELO is meant to be reflective of skill, then it should be implemented in a way that will reflect skill. The changes around "boosting" have entirely broken ELO's reflection on skill because there's an indefinite time horizon on what they consider a match. I have "win streaks" because in the last 12 months I've beaten an account 5 times, meanwhile I've played ~2000 tables of this game. That's not a win streak or a case of boosting by any reasonable definition, so the fact that their change applies to those games is what I've been very clear is broken.

Playing while being afraid to lose elo is also boring though. What I mean is that you should just let it be and it will just naturally eveolve to show: (1) your current score, (2) compared to other players, (3) at this moment. All these 3 aspects can only happen if you are actively playing, with different players. If you stop playing or if you are only playing in a closed circle, your score is not fairly comparable to other players.

If the rules are same for everyone and we can guide the devs to make some minor changes to the calculations then I don't see any problem behind the idea of these implementations. Sure, some tweaking might be needed but it isn't something that needs to be so worked up over.

Elo can fluctuate a lot in some games where luck is in play. Catan for instance... So it is a very rough indication of skill sometimes.

What is currently missing is to track score based on the game settings - player count, expansions/variants etc.

I don't see boosting changes as much of a break... if they would use K factor. It just makes it slower for you to leech off elo from some other player.

ELO doesn't increase indefinitely, it was already zero-sum (barring wins against <100 ranked players).
...
In the new system, if I played 100000 games with someone and we both started with 300 ELO, we would end with far less than 600 ELO between the two of us, as instead of just modifying the K-factor the change as implemented "deletes" ELO during win streaks. In the most degenerate case, i.e. an equally skilled opponent where we would have previously ended at 300 ELO after those games, in the new system we'd both end at 100 ELO.

I think that's a fairly clear and obvious "proof" as to what we mean by this system being broken. The new system only grants winners of a streak a decreasing % of the ELO they would have previously won, while still removing 100% of what is lost. So in the past you'd see ELO_win - ELO_loss = 0. A zero-sum game. Now you have ELO_win * streak_factor - ELO_loss <= 0. It's zero sum for the first game (streak_factor = 0), but for any game thereafter it is net negative on ELO (as streak_factor >= 0). If you take this and look at any game, it means the EV (expected value) of ELO for a game is net negative, which is why I keep saying that it literally disincentivizes playing the game for ELO. The expectation over time is that even the best player on this site will get to 100 ELO at some point after enough streaks delete all the ELO in the community pool. That's nonsense. Sure it might take an insanely long time for some games, but there are plenty of niche competitive games on the site where this will meaningfully depress scores without much time. For Kingdom Builder, I will guarantee this means no one will ever be able to hit Master again. That might not be important to you (and arguably not for me as I've already done it), but there are community members that do have enough skill to technically hit that rank, and because of these changes they will never be able to.

I know how elo works. I've even implemented one as a personal project. Some people here think that it should always keep increasing and they get mad when they lose elo. That's why I propose players to ignore the elo score and to just keep playing. The less you think of it, the more reflective of your skill it is. As you are not intentionally trying to game the system - choosing players to play with etc.

Elo in the current system can also inflate. Someone could just create 4-5 different users and then make one of them always win against the others. And when no more score can be won out of them then he creates more. So he can gain a lot of elo out of them and then just not log in with them anymore. So at the end of the day, the system as a whole, if you throw away inactive players, will have more active elo points than the average zero-sum would be. So... we have some mechanisms that introduce elo into the system and some that bring it out. Theoretically, sure, it might seem like a balanced system but in reality it is not.

But.. why bother? You just have to compare the current elo between players. The active players. Elo earned a few years ago would not be a good representation of skill today. Strategies might be improved a lot in this time. Only look at what elos currently active players have and compare it to your own. That's all you need. Whether the score is inflating or deflating over time, does not matter in the grand scheme of things.

You did not seem to try to comprehend the rest of my post, so I'm not sure how to respond to this. It showed that I would gain 0 ELO for beating a player who has ~620 ELO right now (even though I have ~590 ELO right now). That is not how ELO works today, that is explicitly what the change has done, and it sounds like you're somehow arguing that makes sense because... me beating a player better than me shouldn't give me any rating at all?

It's just an anti-boosting measure. Sure, it sucks for specific strawman-cases but overall I think it helps the system as a whole to prevent abusers. You are supposed to be playing with different players anyway. And if you do win a lot vs a higher elo player, maybe his elo has been inflated unfairly before, indicating that the system was broken in the first place.

Also to conclude, you seem to be trolling by deliberately misconstruing others posts. I don't think the ELO Decay change is "bad", I think it's entirely meaningless. I gave an example of the #1 ranked by ELO player for Kingdom Builder who the community knows is parked in that spot. The decay change does nothing to stop their parking as it requires you play 1 game every 3 months.

I'm not trolling. I see people so worked up that I want to give an opposing view. Let's try to improve the system, not only having a static boring elo score that has been easy to abuse.
Being able or not being able to reach master does not matter. Maybe the old system itself was broken to even allow players from reaching master. Inflated values and whatnot. Overall, the more players there are, the higher the "pyramid" of elo can become. So given enough players, the master ranking would be possible for any game. When it comes to decay... Maybe it needs to be reviewed as well. For public top 10 player lists, sure, maybe you should be playing more regularly, once a week or so.