Elo
The Elo rating is a rating method that aims to reflect a player's true level. It tries to keep the rating close to the player's actual level, and it can be used to rank many participants and quantify the strength differences between them.
Its philosophy is completely different from traditional "points". The Elo system adjusts ratings according to the win rate between participants, so that ratings intuitively reflect the gap in strength. For example, when a high-rated player faces a low-rated opponent, a win adds only a few points, a loss costs many more, and a draw costs a few points.
Elo's philosophy is completely different from traditional points: Elo wants to reflect a player's true level, and pursues the idea that "whatever a player's real strength is, the rating should sit in a certain range". Many traditional point systems, in contrast, can grow without limit.
Elo reflects the expected score between two sides. For example, if the two sides differ by 100 points, Elo considers their win rates to be 64% versus 36%. If a high-rated player keeps drawing against a low-rated player, why should we still believe that player's level is high? It is the results that determine the Elo rating; a draw is treated as the two sides being equal in this game, which narrows the Elo gap between them a little. That draws are common is a property of the game itself and has nothing to do with the Elo rating.
Elo calculator
- Stronger side wins 39.0%
- Draw rate 50.0%
- Stronger side loses 11.0%
Drag the slider to see what an Elo gap means. The higher the draw rate, the fewer decisive wins the same gap produces.
When two engines in a ranking meet, if engine A's Elo rating is 71 points higher than engine B's, the Elo formula says A's expected score against B is about 60% (a draw counts as half a win and half a loss). If actual results show A scoring below 60%, A's Elo is adjusted down and B's up accordingly, to reflect the strength difference more accurately. The Elo of multiple engines is usually computed with tools such as Bayeselo and Ordo.
Side note: Elo has almost become a pejorative term in video games, but Elo is just a rating system and has nothing to do with matchmaking. How the score is used for matchmaking is up to the matchmaking system; however accurate the rating, it cannot stop a matchmaker from pairing a 1000-point player with a 100-point one. Also, video games basically don't use plain Elo either, but more complex Elo variants or rating systems created from Elo's ideas (such as MMR).
In one-on-one engine testing, Elo can be computed entirely from the win rate.
When it is not one-on-one, different Elo systems may differ in calculation method, in the engines (players) in the system, in the match methods and so on, so scores from different Elo systems cannot be compared with each other. You can refer to the relative strength order, but you cannot compare the numbers directly.