Methodology
The data
Our data is based on more than 1.3 million rated games from Chess.com and Lichess, in blitz 3+0, blitz 5+0 and rapid 10+0. Each site and time control is measured separately, in rating bands 100 points wide: 1500 means 1500–1599.
Only rated games between players within 400 points of each other are counted. Bots and berserked Lichess arena games are left out, and the averages at each rating also leave out accounts that the sites have closed or flagged and players whose rating had not settled.
The analysis
Every game is analyzed with Stockfish 18 at 100,000 nodes a position, comparable to a depth-18 search. A member's games get exactly the same analysis as the games they are compared with.
The analysis that Chess.com and Lichess attach to games is not used. A site analyzes only some of its games, and not all of them the same way, so those numbers cannot be compared with each other.
What the terms mean
- Centipawn.
- A hundredth of a pawn, the unit a chess engine scores a position in. Giving up a pawn for nothing costs about 100.
- Centipawn loss.
- How much worse a move is than the best move in the position. The average (ACPL) counts every move a player makes.
- Blunder, mistake, inaccuracy.
- A move that loses at least 150 centipawns is a blunder, at least 70 a mistake and at least 40 an inaccuracy. Only live moves are counted.
- Live move.
- A move played while the game is not already decided. A game is decided when one side is at least three pawns ahead by the engine's score both before the move and after it.
- Winning chances.
- The engine's score turned into the chance of winning from the position. What a move costs in winning chances is how much of that chance it gave away.
- Only move.
- A position where one move is clearly better than every other. “Finds the only move” is how often a player plays it.
- Time trouble.
- Less than a tenth of the starting time left on the clock. “Games in time trouble” is the share of games in which a player's clock got that low on any move, however the game stood.
- Percentile.
- The share of a site's players who are rated lower. Blitz and rapid players are ranked separately.
Your report
A report reads your newest 500 rated games from the last six months in a time control, and only those played since your rating came within 200 points of where it is now. It compares you with the band most of those games were played in, on the same site and in the same time control. The band beside it is added when at least a fifth of the games were played there. A percentile in your report is where you stand among the players of that band.
Trends
Trends follow six skills game by game, over every game of yours we have analyzed. They appear once we have analyzed 900 of your games in a time control, starting with your last 18 months. Each skill is a score on which 100 is a typical player of your own rating band. That band stays fixed, so a score moves only when your play does, and you can reset it on your account page.
How Chess.com and Lichess ratings are converted
Each site rates its own pool of players, so one number means a different strength on each. How well people play can be compared, because every game from both sites gets the same analysis.
Four measures of move quality fall with every 100 points of rating on both sites: average centipawn loss, blunders per 100 live moves, winning chances lost per live move, and the same with blunders set aside. For each one we find the Lichess rating at which players score what a Chess.com rating scores, and the converted rating is the median of the four. The tables are on Chess.com vs Lichess ratings.
The check on the conversion
If two ratings mark the same strength, statistics that played no part in matching them should agree too. The table's first column is the largest gap found at any of the 46 matched pairs: every Chess.com band from 200 to 2400, in blitz and in rapid. The other columns show three of those pairs.
| Not used in the conversion | Largest gapof 46 pairs | Chess.com 1000Lichess 1376 | Chess.com 1500Lichess 1709 | Chess.com 2000Lichess 2000 |
|---|---|---|---|---|
| Move quality | ||||
| Mistakes per 100 live moves | 4.0% | 15.52 / 15.25 | 13.98 / 13.94 | 12.14 / 12.40 |
| Inaccuracies per 100 live moves | 5.2% | 12.20 / 12.08 | 11.95 / 12.04 | 10.96 / 11.35 |
| Finds the only move | 2.4% | 70.8% / 70.9% | 74.9% / 76.1% | 79.1% / 79.6% |
| Plays the engine's top move | 2.8% | 31.6% / 31.7% | 33.5% / 33.4% | 37.1% / 37.0% |
| How the games go | ||||
| Game length, in moves | 6.2% | 34.0 / 33.4 | 36.1 / 35.9 | 37.6 / 38.1 |
| Live moves per game | 8.2% | 16.9 / 16.3 | 19.6 / 19.6 | 22.5 / 22.9 |
| Checks per 100 moves | 12.0% | 8.25 / 8.61 | 7.56 / 7.60 | 6.91 / 7.04 |
| Captures per 100 moves | 4.0% | 25.9 / 25.9 | 24.6 / 24.8 | 23.8 / 23.9 |
| Blunders by the other side per 100 live moves | 6.9% | 16.06 / 16.41 | 12.55 / 12.61 | 9.46 / 9.59 |
| Games that reach an endgame | 13.3% | 30.8% / 29.0% | 32.3% / 32.2% | 33.7% / 34.1% |
The largest gap is a share of the Chess.com figure. The three pairs are blitz 3+0, with the Chess.com figure first in each cell.
The measures of move quality agree to within 5.2% at every pair, and the measures of how the games go to within 13.3%.