🌿ELO System Updates

[teallite]

Does this actually go on? Like people actually hunting out players for ELO, cannot say I've witnessed it, and eg.

Absolutely. But they are not affected by consecutive win penalty, because they hunt for new/weak players and there are lots of them.

so you set your settings to beginners only to try and get Banzai and Beserker badge, but you're not going to gain any real ELO benefits from doing that.

It may be 1 Elo or less, but it's guaranteed. They play dozens of games and climb to the top.

[teallite]

Ironically, if I wanted to climb, the best way to do it now is to pick on weak players for elo. There are dozens/hundreds of new players I've never played before worth 1 elo each.

Exactly. As I said before, cheaters are fine, newbie hunters are fine. It's still unclear who benefits from these changes.

[h_illes]

The cutoff time for the win streak count, which used to be around 30 days, appears to have been modified to somewhere around 7 days.

I played and won 6 games against the same opponent within 10 days, but for my last game, a win streak penalty for only 3 wins was applied.
https://boardgamearena.com/gamestats?pl ... me_id=1480

We're always listening to your feedback and working to make Board Game Arena the best place to play. Keep those messages coming!

lordalx, such changes should be explained in detail, in advance, instead of the players having to figure it out with trial and error. We already noticed the loss streak mitigation (or whatever you want to call it), and are in the process of figuring out the finer details, but there has been no official communication yet.

So an official announcement is in order. Once you're at it, please also address the fact why it has been implemented in such a way that it still disincentivizes playing against players who you have a win streak against instead of, say, a reduced K factor, when this specific potential pitfall has been highlighted within 3 hours of your last official comment from a month ago, and also let us know when you are going to actually implement an acceptable solution. If you (or whoever makes decisions) have trouble understanding the issues with the current implementation, do not hesitate to ask questions.

[nik592]

lordalx, such changes should be explained in detail, in advance, instead of the players having to figure it out with trial and error. We already noticed the loss streak mitigation (or whatever you want to call it), and are in the process of figuring out the finer details, but there has been no official communication yet.

Well, technically there has on the loss streak mitigation, though it's not been super obvious (in the release summaries - https://boardgamearena.com/forum/viewto ... 58#p175958). But yes, these things should be more clearly communicated - given how much activity there has been here, it should have also been mentioned here. I haven't seen anything about your observed 7 day window for the streak (vs. the original 30 days) either way.

[h_illes]

lordalx, such changes should be explained in detail, in advance, instead of the players having to figure it out with trial and error. We already noticed the loss streak mitigation (or whatever you want to call it), and are in the process of figuring out the finer details, but there has been no official communication yet.

Well, technically there has on the loss streak mitigation, though it's not been super obvious (in the release summaries - https://boardgamearena.com/forum/viewto ... 58#p175958). But yes, these things should be more clearly communicated - given how much activity there has been here, it should have also been mentioned here. I haven't seen anything about your observed 7 day window for the streak (vs. the original 30 days) either way.

I didn't even know about this information outlet. Thanks! Alas, no details there either.

[anykidrated]

Hoever, the long term consequence of the current version is that strong players who keep playing through win streaks will be punished (their ELO lowered), while weaker players will have their ELO increased

This is probably the point. Make it easier for casual players to gain ELO and they are less likely to give up due to feeling bad. Make playing the same player over and over and strong players will be disincentivised from picking on the weak players for ELO.

Ironically, if I wanted to climb, the best way to do it now is to pick on weak players for elo. There are dozens/hundreds of new players I've never played before worth 1 elo each.

It's the players one can typically beat but who are good enough to win a few that are the worst elo drains. You'll get a winstreak, reap only part of the rewards, then lose it all.

This is exactly what I've been doing. My climb has barely been affected when I switched to doing this

The cutoff time for the win streak count, which used to be around 30 days, appears to have been modified to somewhere around 7 days.

I played and won 6 games against the same opponent within 10 days, but for my last game, a win streak penalty for only 3 wins was applied.
https://boardgamearena.com/gamestats?pl ... me_id=1480

Wow ok. I literally stacked so many games to try and play people 2-3x/month, now I should literally not be hit by the winstreak penalty at all

[ab in dc]

lordalx, such changes should be explained in detail, in advance, instead of the players having to figure it out with trial and error. We already noticed the loss streak mitigation (or whatever you want to call it), and are in the process of figuring out the finer details, but there has been no official communication yet.

Well, technically there has on the loss streak mitigation, though it's not been super obvious (in the release summaries - https://boardgamearena.com/forum/viewto ... 58#p175958). But yes, these things should be more clearly communicated - given how much activity there has been here, it should have also been mentioned here. I haven't seen anything about your observed 7 day window for the streak (vs. the original 30 days) either way.

I didn't even know about this information outlet. Thanks! Alas, no details there either.

Thanks -- I did notice the "ELO malus for consecutive games against same player now also applies to losses (instead of just victories)". Which...kinda sounds like a reduced K factor, but in a more complicated fashion?

[h_illes]

Well, technically there has on the loss streak mitigation, though it's not been super obvious (in the release summaries - https://boardgamearena.com/forum/viewto ... 58#p175958). But yes, these things should be more clearly communicated - given how much activity there has been here, it should have also been mentioned here. I haven't seen anything about your observed 7 day window for the streak (vs. the original 30 days) either way.

I didn't even know about this information outlet. Thanks! Alas, no details there either.

Thanks -- I did notice the "ELO malus for consecutive games against same player now also applies to losses (instead of just victories)". Which...kinda sounds like a reduced K factor, but in a more complicated fashion?

It is similar, but there is an important difference for the last game (when the win streak ends). For a reduced K factor, if player A has a win streak against player B but player B finally wins, the reduced K factor would applies to the game which B won, so the ELO exchange is still according to the reduced %. For the current implementation, if player A has a win streak against player B but player B finally wins, there is a full exchange of ELO. So actually as long as the win streak lasts, player A has nothing to gain and everything to lose against player B, which is unfair and a strong incentive to not play player B.

This way you badly beat strong players and inflate weak ones.
Assume I am much stronger than player B (say I win 19/20) and I am paired with him several times in good faith (it happened to me in arena where I have no control on the opponent ...).
With this mechanism I will gain almost no ELO and get a strong drop when occasionally I loose. Player B can almost only gain ELO. If this keep going eventually we will get the same ELO despite the strong unbalance. To achieve your goal lower the k factor for both player according to the number of games played (Better if this is done proportionally to the probability of having multiple games by chance so that small players pool are not penalized ). You still achieve the goal to lower the impact of multiple games with the same opponent but don't introduce a manifest bias in
ELO calculation

Very good point. A lowered K factor means that the K factor is lowered in advance, regardless of the outcome of the last game. If the K factor depends on the outcome of the actual game, that's very different, and will indeed bring players of different skills closer in rating. So even though the ELO leak will stop, it will still make ELO a much worse indicator of skill. I hope this is not what they intend to do.

[h_illes]

We're always listening to your feedback and working to make Board Game Arena the best place to play. Keep those messages coming!

Changing the win streak count cutoff time to ~7 days instead of ~30 days helps mitigate the issues somewhat, but they nevertheless remain, particularly in Arena (although they are problematic outside Arena as well). Let me reiterate these issues briefly:
- In the current implementation, if player A has a win streak against player B, they are strongly incentivized to refuse to play player B, as player A has nothing to win and everything to lose.
- The win probability formula displayed after the game is also incorrect, as it is based on the old ELO formula, while the current implementation results in a reduced difference in players' ratings.
- The playing field created by the current implementation is unfair; from among strong players, it favours players who play fewer games.

The current implementation is also baffling because this potential pitfall has been pointed out by the players well before the last change. In other words, we have arrived at a point where players actively anticipate your mistakes and point out potential pitfalls IN ADVANCE, with detailed explanation. For me, a situation like this would already bother my professional pride greatly, yet the worst part is that you STILL make those mistakes after. How do you justify such decisions? Also, communication and details please.

[h_illes]

Let me explicitly offer a variant that I believe is acceptable for most players (it has been posted before by several players to a varying level of detail, I am just summarizing it here).

Instead of a win streak penalty-loss streak mitigation system, go with a reduced K factor for frequent opponents, regardless of who wins. So if e.g. player A has played player B many times (in a given time window), reduce the K factor for the computation of ELO exchange between the two of them, regardless of the outcome, and regardless of prior results, just based on the number of games they played in a given time window. This change will not distort the old ELO formula, it will just slow down convergence of players to their true ELO. This way, there would be no situations where one player is disincentivized from playing another player; the stakes would be lowered, but for both players, in a fair manner.

This variant would still have the desired side effect that it would be hard to obtain a lot of rating from one single opponent, thus cheating this way would be infeasible.

The details of the K factor reduction can be essentially the same as for the current win streak penalty, e.g. 3% after 1 game, 7% after 2 games etc., to 100% after 6 games. (Personally, I still don't like a 100% reduction, but in this setting, I could live with that.) The cutoff time for the games count can also remain at 1 week.

Let me call out to other players: do you find this implementation acceptable?