DANGER, rating system will break if bots play rated !

Looks like they fixed the issue. Human ratings will be affected from now on.

https://github.com/ornicar/lila/issues/4297

@Dopaminergic_Chess
this is just for blitz and it looks like only anecdotal evidence

did the staff themselves at least allege that it is due to engines? why then just blitz?

also your rating on icc being so significantly lower than top rated players doesn't really tell much, the rating curve there might be significantly different from lichess - there appear to be a multitude of factors that affect rating that are less related to actual chess playing prowess
for example I've seen countless 1600/1700 rated players on lichess that seem to play near perfect/flawless/blunderless games yet somehow can't even reach 2000
there could be another explanation - for the "club of high rated players" - maybe they are mainly titled players who just choose to play among themselves and their ratings have become deflated much more than the non-titled ones - since they don't play much with non-titled players their high ratings don't "leak" much to the non-titled player group

@BermudaGhostShip , no they haven't blamed computers accounts for the rating distribution distortion in ICC. As a paid service, they making such affirmation is not part of their best interest.

I was not comparing my rating to top players. That would be silly. I was comparing my ICC rating (1300-1600) to my lichess rating (1600-2000). It's much harder in ICC to push up your blitz rating.

And a crew of GMs playing only against themselvez won't ever nicrease their everage rating. This is not how glicko rating works. The points are basically transfered betwee opponents, not created out of nothing.

The reason why this happens mostly in blitz is because computers are best in blitz time control, and most computers are dedicated to play blitz only.

@Dopaminergic_Chess

does ICC use glicko too? quick search for "icc chess glicko" does not seem to support it - I thought they use some other more usual ELO type system.
Also of course I meant 'inflated' not 'deflated' - deflated is when ratings become smaller. Could it not be that nowadays titled players go to ICC mostly only to play other titled players, the high rating being in blitz would partly support it since titled players are known to be less interested in longer time controls online, especially against non-titled players while they could spare the occasional minute against amateur players in bullet. I've heard ICC has struggled with player numbers for last years that would increase chances for titled players to go there only to play other titled players not as a general play-all like lichess.
From what I know it is very hard to gain high rating by playing low rated players, there is just too much randomness in chess, look at Leela, she is basically destroying everyone but struggles for a long time to even maintain 2700.
There is good chance that whatever new titled players came to ICC for last years played only other titled players - especially in blitz - since ICC struggled with player numbers, especially getting new members since it is a paid chess server there could have been extreme disproportion between new non-titled and new titled because IMs and GMs are free to play on ICC - so there could have been relatively high influx of new IMs and new GMs there for last years while struggling getting new non-titled players. All titled would play only other titled in blitz and thus the result - new player introducing new points into the pool, the segregation leading to something almost like two different rating scales within the system.

will write the main idea in short - ICC is free for GMs and IMs, I think they do not use glicko and are perfectly susceptible for rating inflation, - they struggle getting new players when there are so high competition from free to play sites, - they could be getting disproportionate influx of new titled players compared to new non-titled - the titled players bringing in new points into the system that would be mostly shared only among titled in blitz, not leaking much to non-titled. It is well known that titled players are much less interested in playing non-titled especially in anything longer than say 1+0. That would explain why rating inflation is only for blitz (and before you say why not in the longer than blitz time control - titled players hardly ever play anything longer than blitz online). There could be relatively few games among non-titled, playing with much smaller inflation. This would explain all the observed facts, there is no proposed explanation so far why engines could cause it.

http://www.chesschat.org/archive/index.php/t-5222.html
quote: "From a practical point of view online, ICC has experienced huge inflation with the ELO system, whilst FICS has experienced considerable deflation using Glicko."

from this forum post it follows that ICC uses ELO system and experiences rating inflation - again they struggle getting new players because they are paid site, while GMs/IMs are free to play, GMs/IMs gain little in playing amateurs, so they play them rarely on anything above bullet, thus the high titled player rating inflation only in blitz, and titled hardly ever play rapid/classical etc. online

this much more recent reddit post again states that ICC uses ELO system:
https://www.reddit.com/r/chess/comments/2wwvvj/chesscom_vs_icc/
quote: "ICC uses the Elo system"

@BermudaGhostShip ,

Thanks for researching into it. You are right about ICC using Elo. It is basically the same equation except that for non-established ratings Glicko uses the RD factor while Elo uses the k-factor. And scales are the same since both set 400 as the maximum measurable difference between two players playing against each other.

Lichess uses Glicko too, not as precisely as FICS though. In FICS after a long period of inactivity your rating will be unistablished again and move in larger steps due to your RD increase. In Lichess however, it doesn't seem to take inactivity time span into account.

Sorry but the idea that a group of players (say titled players) could inflate their ratings by playing mostly against themselves, makes no sense for me. Especially not in ICC where the K-factor only changes during your first 21 games, then it remains steady 32. So titled players with established accounts playing each other will just give and take ratings from each other,, in equal amounts.

That rating inflation can only be the result of climbing up over steady reference points in the rating scale. Very much like when you climb a mountain. I can only think about two steady sources: (a) the initial rating of fresh accounts and (b) the steady ratings of computer accounts.

So a new super GM may have been provided with an initial rating equivalent to his FIDE rating (such as 2750). Yet, the highest ratings seen (above 3600) are more than two worlds of difference above that. I say that because both Elo and Glicko define 400 as the maximum measurable difference between two players.

Dropping and recreating titled accounts, if allowed, wouldn't produce the inflation. Instead, it would just pull the inflated ratings back to FIDE scale.

Finally, I only see computer accounts as the steady reference points that where used to climb up into 3600. I don't know how they managed to win computers rated over 3200, but somehow they did.

@Dopaminergic_Chess

I don't have much time anymore for this discussion so most likely this is my last

quote: "In Lichess however, it doesn't seem to take inactivity time span into account."
as far as I know inactivity is taken in account here

quote: " I don't know how they managed to win computers rated over 3200, but somehow they did."
this is just hard to believe and would need some evidence to be believed, computers vastly destroy humans at chess nowadays, this is just too much to believe without any evidence that GMs/IMs suddenly started to destroy stockfish like engines and gain super ratings on ICC

quote: "Sorry but the idea that a group of players (say titled players) could inflate their ratings by playing mostly against themselves, makes no sense for me"
I am not an expert on rating systems but as far as I know ELO rating systems do inflate especially at the higher echelon, ICC is really struggling getting new players, it would make perfect sense that GMs and IMs there mostly go to get some strong titled opposition nowadays while amateur newcomers are not so many
FIDE uses ELO rating system and has been constantly inflating over time at least for the top players - there is again perfectly easy and clear explanation how server like ICC can get vastly inflated ratings for top players at blitz, I will list them point by point, and I don't see point to reply unless you can explain how it does not make sense - which of those points are not true:

1. ICC has a relatively low number of players (largely owing to it being a paid site)
2. ICC has been known for a place where GMs/IMs can still get some strong opposition as some titled players play only there
3. ICC is free for GMs and IMs, not for other players - thus it makes sense that it might have disproportionate amount of titled players to non-titled (as compared to other sites), especially nowadays when it has been losing ground to other chess sites for years - GMs and IMs still see point going there because it is a free extra for them where they can still get some good opposition
4. Titled players are known to be reluctant to play amateur players in anything but bullet (a GM can't learn much from a typical amateur) - thus the high inflation is in blitz not bullet where they mind less to spare a minute
5. if an amateur player reaches high rating (is overrated) that can make a titled player interested in playing against him on site like ICC, no much other reason, thus the lower rated/non-titled echelon will be constantly feeding their rating to titled players thus deflating their ratings and inflating titled player ratings - this is my point and I don't see how you can deny its plausibility
6. the reason why titled players will be much less interested in playing amateurs on icc than say lichess is because in lichess they can at least gain visibility, friends, contacts etc. in sites that are struggling to gain players, have always had relatively low player count (like ICC), no one might even find out about a GM playing some amateur - and GM won't learn anything from beating him/her so there is little point thus only the high rated amateurs will get to play titled players in blitz resulting in feeding titled player ratings (inflating them)

In my opinion you cannot say Glicko leads to „a“ and regular Elo calculating to „b“.

The stability of the system with billions of games played is a tough issue. It’s like a bath-tube with some inlets and outlets - it‘s pretty difficult to keep it on a constant level over years. And playing games on the server is like high-frequency trade...

I am pretty sure that sometimes a deviation in the fourth digit after the comma leads to inflating/deflating effects in the long term. A big questions remains: how to handle newbies and dead accounts? Is there a rating flow in or out?

@BermudaGhostShip

essentially killing off the whole bot API idea as it will become irrelevant

How about a separate bot rating for the players? Lichess already separates ratings for variants and time controls. It would be silly to let the bots skew the ratings between humans. Personally I would see that as worse than irrelevant and would call it detrimental even.