Hanabi bots' ELO - are they still valid?

[Rhydon MK2]

I saw a post viewtopic.php?t=4814
It says that in multicolor mode, 2 people with perfect score is assumed to have won against ELO 2330.
That doesn't sound plausible based on my experience. ELO ~1000 players doesn't get >1 point in that scenario.
Is the table outdated?
Thanks.

[beri]

The way ELO is counted on BGA changed in 2017.
Current ELO is old ELO, minus 1300.

Pending a reply from BGA’s boss, I’ll update the tables in the topic you linked to.

[Rhydon MK2]

Thanks a lot. That explains everything.

[qwertasdfg]

That doesn't really explain it.
If two people around 1000 win against the 1030bot, that should give more than the one elo point started in the opening post.

Before, at least it was obvious the table was outdated. Now it just seems to be wrong by itself

[Rhydon MK2]

Well, actually you're right. I was bit rash to say that. Recently I was only playing 4/5p game.
I forgot that there is a ELO ceiling if you only play in 2p game with both player very similar ELO, and that ceiling is definitely lower than 1030 from my experience.
I'm not the expert of ELO calculation, but if ELO difference in two sides (bot and players) should make ELO change after the game, table still doesn't make sense.

[Travis Hall]

There’s not a ceiling. That’s not how ELO works. However, the calculations and recorded values use smaller increments than what is shown in the UI. The result is that after a certain point, your increases will often be too small for you to see in the UI. You’ll still get an increase, and there is no actual limit on how high your ELO can go, but there’s a point where you would have to play a hundred perfect games in a row for the increase to show up.

[Stroom]

There probably is a cap with the technical solution where your elo won't rise anymore. Numbers are not stored in infinite manner. There will be a point where you round the number down and it becomes 0. The calculation might result in 0.000123 in real life but if the system only handles 3 decimals then it's 0.000 so it won't influence your score. (In reality the cutoff point is probably much further than that, maybe 10-20 decimal points). In 2p you'd need to play with 0 elo players all the time then.

[Travis Hall]

Yeah, I guess there would be a point where the calculated rise drops below the system’s precision. (I understand ELO scores are recorded only to hundredths of a point, so that’s the precision of the system.) That would still require reaching much higher scores than any I’ve seen here, though.

(I find that, while I’m playing with a random selection of partners from among the population, and refusing to abandon, I top out at a bit over 1000 before other people’s poor play pulls me down again. Possibly I could push higher by playing only with a carefully curated group, but life is too short for that. That’s well short of any ceiling.)

[Wreckage]

Here are a couple a 3 player games where the ELO gain is 0.00 for a score of 30.
https://boardgamearena.com/table?table=143525546
https://boardgamearena.com/table?table=122874356
3 player games are the worst I think for ELO, but we only have an 896 average here, and cannot gain.

In 2 player games I think ELO can be gained until the table average is about 1035 ELO.
I don't know where the cap would be for 4 player games, but I'm certain it's better than 3 player games.
I have heard that the ELO for 5 player games match 2 player games, but I don't know this for sure.

So I believe, if an awesome team of 5 players started new accounts and only played with the exact same team, they would reach a ceiling of 1010-1030 with an absolute ceiling of about 1035.

[Travis Hall]

Here are a couple a 3 player games where the ELO gain is 0.00 for a score of 30.

How can you tell? The point gain on those pages is rounded to an integer.