What the number means
How ratings work
Every rated game here moves a number, and the rules it moves by are seventy years old and simpler than they look. This is what the number claims, what it takes to move it, and how far anyone has ever got. Every figure in the tables is computed from the rating code this site actually runs.
A rating is a prediction, not a score
Arpad Elo, a physics professor and a strong amateur player, built the system in the 1950s. The US federation adopted it in 1960 and FIDE in 1970.
The number does not measure how much you have won. It makes one claim, and only one: the difference between two ratings predicts how a match between those two players will go. Everything else follows from that.
A gap of 400 points means the stronger player is expected to score about nine games in ten. Level ratings mean an even match. The scale is deliberately arbitrary — 1500 is not twice as good as 750, and the numbers only mean something relative to each other.
After a game, your rating moves by the difference between what you actually scored and what the system expected you to score. Beat someone you were expected to beat and you gain almost nothing, because the prediction was already right. Beat someone far above you and the prediction was badly wrong, so it corrects hard in your favour.
K decides how fast it moves
The same result moves a newcomer's rating much further than a veteran's, and that is deliberate.
K is the multiplier on every correction. A high K means the system does not yet trust its own estimate of you and will move it a long way on new evidence; a low K means it thinks it has you about right and wants proof before shifting.
FIDE uses 40 for a player's first thirty games, 20 for most players, and 10 once someone has reached 2400 — at the top, ratings are meant to be slow and stable. This app starts everyone at 1200 and follows the same shape: fast at first, slower as it learns.
The practical consequence is worth knowing before you play: your first twenty games move your rating more than the fifty after them. That is the window where the number is really being set.
How to actually raise it
The honest instruction is shorter than people expect: play opponents near or above your level, and play enough games.
There is no streak that unlocks anything, and no bonus for consecutive wins — each game is scored on its own. What decides how fast you climb is who you play. Beating someone far below you was already predicted, so it pays a point or two; losing to them was not predicted at all, and costs ten times as much. Farming weaker opponents is close to worthless and quietly risky.
Beating someone above you is where the points are. The table below is computed from this app's own rating code, so it is what will actually happen to your number.
If you want a target: against evenly matched opponents, roughly thirteen straight wins moves an established rating a hundred points. In your first twenty games, when K is at its highest, five will do it.
| Opponent | Expected | If you win | If you lose |
|---|---|---|---|
| -400 | 91% | +4 | -36 |
| -200 | 76% | +10 | -30 |
| -100 | 64% | +14 | -26 |
| Level with you | 50% | +20 | -20 |
| +100 | 36% | +26 | -14 |
| +200 | 24% | +30 | -10 |
| +400 | 9% | +36 | -4 |
| Opponent | Expected | If you win | If you lose |
|---|---|---|---|
| -400 | 91% | +1 | -15 |
| -200 | 76% | +4 | -12 |
| -100 | 64% | +6 | -10 |
| Level with you | 50% | +8 | -8 |
| +100 | 36% | +10 | -6 |
| +200 | 24% | +12 | -4 |
| +400 | 9% | +15 | -1 |
The top of the scale
Nobody has passed 2882, and the number itself has drifted over the decades.
Ratings are a closed system: points come from other players, so the whole population can drift up or down over time without anyone getting better or worse. Comparing a 1972 rating with a 2024 one by the number alone is not safe, which is why direct cross-era claims about who was strongest tend to generate more heat than light.
Titles are awarded on rating plus norms — results in strong international events — and unlike ratings they are permanent. Grandmaster needs 2500 and three norms; International Master 2400; FIDE Master 2300, no norms required.
Highest ever
2882
Magnus Carlsen, on the FIDE list of May 2014. Still the record.
The previous record
2851
Garry Kasparov, July 1999. It stood for fifteen years.
Grandmaster
2500 + 3 norms
The rating alone is not enough — the norms have to come from strong international events.
FIDE's floor
1400
Raised from 1000 in 2024. Below the floor, FIDE simply does not publish a number.
The other three games
Elo travels wherever a game ends in a win, a draw or a loss — but the more luck a game has, the more games it takes before the number means anything.
Draughts uses Elo essentially unchanged; the international federation rates players on the same curve chess does. Its problem is the opposite of chess's: at the top, near-perfect play draws almost everything, and a rating built on draws separates people slowly.
Backgammon has to adapt it. A single game turns heavily on the dice, so one result is weak evidence — the established systems scale the rating change by the square root of the match length, which means a long match moves your rating far more than a one-pointer. That is the same instinct as K: move further on better evidence.
Dominoes has no world body and no single rating. Where online sites rate it at all, they use an Elo variant, and the same caveat applies more strongly still — with a hidden hand and a shuffled boneyard, a short run of results says very little.