Rating system • page 1/1 • General Chess Discussion • lichess.org

Pepeta

I would like to know technically how the lichess rating system works.
Does the lichess have the same FIDE system?
I say this because, I see that in other sites of chess my rating is very different from the lichess. I have this doubt, can anyone help me?

I would like to know technically how the lichess rating system works. Does the lichess have the same FIDE system? I say this because, I see that in other sites of chess my rating is very different from the lichess. I have this doubt, can anyone help me?

sheckley666

Lichess uses the so-called Glicko-2 system, which is an enhancement to the Elo system, used by FIDE. It was developped by Mark E. Glickman.

Note: Even if lichess used the exact same sytem as FIDE, the ratings could heavily vary, because all these systems are relative. I.E. if all other players on lichess were much weaker than the FIDE players, your lichess rating would be higher, and vice versa.

Lichess uses the so-called Glicko-2 system, which is an enhancement to the Elo system, used by FIDE. It was developped by Mark E. Glickman. Note: Even if lichess used the exact same sytem as FIDE, the ratings could heavily vary, because all these systems are relative. I.E. if all other players on lichess were much weaker than the FIDE players, your lichess rating would be higher, and vice versa.

Pepeta

@sheckley666 Got it. Thanks!

bufferunderrun

This should help: https://lichess.org/qa/6/how-does-the-rating-system-work-on-here

cp560

Online Chess sites cannot possibly use full Elo though, as one enormous and massive major drawback of Elo is the abnormally long time needed to fully rate a new player joining (the proverbial unknown factor). In online Chess though, you do have such huge amounts of new players, every single day, which Elo system just cannot handle well. Instead almost all online Chess sites use the Glicko(-2) system instead named after a Harvard professor.

So, your Lichess rating is actually not Elo-based at all, but instead Glicko-based. Glicko ratings are however very deliberately designed to very closely approximate real-world Elo stuff. For example if an online Chess site has 100 thousand members and you are ranked exactly 400th, then you will auto get a rating of 2200, which is almost identical to Elo stuff. Same idea here.

Another factor to consider in both Elo and Glicko systems is the semi-elusive K-factor that is being awarded within both systems. At Super-GM level, FIDE deploys a K-factor of a very small 10 (which means maximum points won-or-lost on any one single game, simplistically). FIDE also has a sliding scale of 40-20-10 for the K-factor - for unranked, up-to 2400+ and lastly GM levels.

Online Chess sites are completely free to explore whatever K-factor they want to use, aka K=40,32,24,20,16,12,10. The higher the K-factor being used behind the ratings algorithm, the higher the swings in ratings upturns, the easier that it then become to progress up the ratings ladder, or to fall from said ladder too. Knowing the K-factor is very important. Under FIDE, you enjoy K=40 for a very short and limited time-period of your first 30 games, then your volatility falls sharply to K=20. Then giant leaps upwards just instantly disappear overnight and instead tiny gains become the more actual reality. K=10 for SuperGM level is protective here, as no massive jumps down or up are allowed anymore. All gains must thus be 100% fought for and earned and also a safety-net is deployed too here. It's hard to fall back radically, under k=10.

Other higher-level concepts used by Online Chess sites include your own actual and persistent ratings floor, any new players assigned floating floors, discarding inactive (or fly-by-night) players aggressively, running rotational and computational series of always moving averages, rate volatility, rate degradation, rate computation.

Glicko might seem like it's one-on-one versus Elo, but that's fully by design alone. Everyone knows Elo, so Glicko(or mark 2) very deliberately tries to follow sane and very well-known Elo markers. A very different system, but one that tries it's best to appear to be compliant. The underlying math is so very different, but to all intents and purposes the end result appears to be about the same. Glicko functions very differently from Elo, but it always approximates Elo's numbers schematics quite closely.

As to your Glicko ranking on Chess. com being quite different compared to Lichess, that is not so difficult to explain. On chess.com, for example, all random visitors (ever) and even for just one quick game and then fast-resign and never reappear again are just counted forever. Say, 3 years ago you beat someone (provisionally) rated 1800, who only played 2 games on the site. That victory still counts and that player's estimated rating still counts (for you). I do not like this system so much.

Lichess's algorithm just intrinsically prioritizes actually active players far more and mostly discounts fly-by-night and random opponents. You will instead get rated against those who do play quite a bit and you will also normally get paired with those on similar sequences too. A bit more realistic is that model. Not retaining a far-flung victory (or loss) from some years ago versus a once-upon a time player.

FIDE just-about exclusively uses basic Elo (with normalized math stuff, which promotes meta variables) stuff which so very heavily punishes lower-rated players and helps higher-rated ones. Most National or Club Elo ratings use the far-kinder logical system of interpreting Elo, which is far kinder to lower ranked players. Part of why your various FIDE and National ratings can often differ hugely. Same system, different interpretations. Difference between logical and normalized Elo is only a single percentage point, but the difference grows forever under cumulative growth angles. FIDE rating is very nasty stuff for the lower rated player. Online Chess sites cannot possibly use full Elo though, as one enormous and massive major drawback of Elo is the abnormally long time needed to fully rate a new player joining (the proverbial unknown factor). In online Chess though, you do have such huge amounts of new players, every single day, which Elo system just cannot handle well. Instead almost all online Chess sites use the Glicko(-2) system instead named after a Harvard professor. So, your Lichess rating is actually not Elo-based at all, but instead Glicko-based. Glicko ratings are however very deliberately designed to very closely approximate real-world Elo stuff. For example if an online Chess site has 100 thousand members and you are ranked exactly 400th, then you will auto get a rating of 2200, which is almost identical to Elo stuff. Same idea here. Another factor to consider in both Elo and Glicko systems is the semi-elusive K-factor that is being awarded within both systems. At Super-GM level, FIDE deploys a K-factor of a very small 10 (which means maximum points won-or-lost on any one single game, simplistically). FIDE also has a sliding scale of 40-20-10 for the K-factor - for unranked, up-to 2400+ and lastly GM levels. Online Chess sites are completely free to explore whatever K-factor they want to use, aka K=40,32,24,20,16,12,10. The higher the K-factor being used behind the ratings algorithm, the higher the swings in ratings upturns, the easier that it then become to progress up the ratings ladder, or to fall from said ladder too. Knowing the K-factor is very important. Under FIDE, you enjoy K=40 for a very short and limited time-period of your first 30 games, then your volatility falls sharply to K=20. Then giant leaps upwards just instantly disappear overnight and instead tiny gains become the more actual reality. K=10 for SuperGM level is protective here, as no massive jumps down or up are allowed anymore. All gains must thus be 100% fought for and earned and also a safety-net is deployed too here. It's hard to fall back radically, under k=10. Other higher-level concepts used by Online Chess sites include your own actual and persistent ratings floor, any new players assigned floating floors, discarding inactive (or fly-by-night) players aggressively, running rotational and computational series of always moving averages, rate volatility, rate degradation, rate computation. Glicko might seem like it's one-on-one versus Elo, but that's fully by design alone. Everyone knows Elo, so Glicko(or mark 2) very deliberately tries to follow sane and very well-known Elo markers. A very different system, but one that tries it's best to appear to be compliant. The underlying math is so very different, but to all intents and purposes the end result appears to be about the same. Glicko functions very differently from Elo, but it always approximates Elo's numbers schematics quite closely. As to your Glicko ranking on Chess. com being quite different compared to Lichess, that is not so difficult to explain. On chess.com, for example, all random visitors (ever) and even for just one quick game and then fast-resign and never reappear again are just counted forever. Say, 3 years ago you beat someone (provisionally) rated 1800, who only played 2 games on the site. That victory still counts and that player's estimated rating still counts (for you). I do not like this system so much. Lichess's algorithm just intrinsically prioritizes actually active players far more and mostly discounts fly-by-night and random opponents. You will instead get rated against those who do play quite a bit and you will also normally get paired with those on similar sequences too. A bit more realistic is that model. Not retaining a far-flung victory (or loss) from some years ago versus a once-upon a time player.