[FR] Could we cap ELO loss at 0 for perfect games?

[Ant Stewart]

Would it be possible cap the ELO loss at 0 for a perfect win?

I understand that the scoring system (for all of BGA for coop games) is based on a theoretical draw against a player of a specific rank - which means when you reach a certain level, even a perfect game is an ELO loss.

I started today with an ELO of 642.

I played 3 games,

Game 1 with a Master and an Expert
Game 2 with two Masters
Game 3 with a Master and an Expert

All 3 games were Tricky: 6th colour + flams - otherwise base.

All 3 games were perfect score 30s.

Afterwards my ELO was 641.

Would it be possible to cap the ELO so that you can't loose ELO on a perfect game? I am not expecting to gain ELO in these situations since the players are of a rating where a perfect score is expected - but it seems odd that a perfect game can still loose ELO.

[I Love You More Than]

I agree that something in the difficulty calibration needs to be adjusted. It feels like the difficulty settings for 3‑ and 4‑player games are noticeably lower than for 5‑player games. Either that, or the average ELO of your group is simply too high compared to the difficulty level you selected. From my experience, Tricky is set easier than Avalanche of Colors, though that’s just my personal impression. I’m also not convinced that Black Powder actually increases the difficulty rating, I don’t think it’s factored in. Overall, there seem to be several arbitrary elements in how difficulty is defined, and those inconsistencies end up affecting how ELO is calculated.

[Ant Stewart]

Yeah, it's not so much about the numbers, even with good numbers, there will still be perfect games that loose ELO.

My request is more, can we adjust the algorithm, so that perfect games can be capped at 0 loss

[I Love You More Than]

Actually, at 5 players the system is designed exactly that way. You can’t lose elo for a perfect score, but you’ll sometimes end up gaining nothing. That might be the baseline they calibrated the model around, and the effect simply cascades down to other player counts.

[Ant Stewart]

How does that work.

Given a fixed ELO virtual opponent, even with 5 players, a draw should still result in a loss in ELO if the 5 players have a sufficiently high ELO?

[I Love You More Than]

Good question, maybe someone from BGA's ELO programming team can answer.

I'm just basing my answer from my experience with 5 player games where the team has an average ELO of 1050-1100 and the result is a gain of 0. I don't know what happens if the ELO is higher since unfortunately there aren't enough players above 1100 on BGA to test it out.

And since we can't track other player's ELO anymore, unless we directly ask them, it's almost impossible to do the math.

[Jellby]

I'm thinking that it may be good to have an elo decrease even after a perfect play. In normal play it should be rare enough that it should not matter much. If you play a lot with "easy" variants, lost elo accounts for your getting rusty (and yes, that happens). If you're teaching others, perhaps you should use friendly mode. If losing elo becomes a "problem" maybe it's because your own elo (or your partners') is inflated? possibly due to cheating? I don't think 1500 (even 1200?) or more makes any sense in Hanabi.