Why measuring beats grinding
Mated exists because the gap between a 1200 and a 1600 is almost never the thing either of them thinks it is.
Measure first, train second
Most improvement advice is generic because most coaches have not seen your games. A thousand of your own games contain the answer already; the job is to read them properly and rank what is costing you the most.
The engine runs on your machine
Stockfish is compiled to WebAssembly and runs in your browser tab. Your games are not shipped to a server to be evaluated, which is also why the analysis is free of queue times.
Your mistakes are the curriculum
Blunder Shield, Blunder Training and the Conversion Trainer contain no stock puzzles at all. Every position in them is one you played, taken one move before it went wrong.
Fifteen minutes, chosen for you
The daily plan is three items long on purpose. Consistency beats volume, and a plan you finish beats a plan you abandon.
What actually gets measured
414 measurements across five categories, each one computed from your own moves and sliced by phase, colour and time control.
Where does the game slip out of your hands?
How often you hand material or the game away, and exactly when it happens.
Are you losing to the position, or to yourself?
Clock use, snap moves, tilt streaks and the games you lose after the first loss.
What are you not seeing?
Wins you saw, wins you walked past, and the patterns you keep missing.
Do winning positions turn into points?
What happens after you are winning, and after you are worse.
Are you out of book before move ten?
Repertoire consistency, how deep your theory really goes, and where you drift.
Things worth being straight about
- We read public games through the official Chess.com and Lichess APIs. We never ask for your password and we cannot see private games.
- Move classification uses win-probability swing rather than raw centipawns, so losing 200cp when you are already three pawns up is not counted as a blunder.
- Benchmarks in the report are reference values for a 1200–1600 club player, not percentiles from a live population. They are there to give a number a direction, not to rank you against strangers.
- The bot ladder sets the engine's own strength limiter rather than making it drop pieces at random, so a win against Bruno means the same thing every time.
- The 399 puzzles in the tactics trainer were generated and then proved exhaustively. Every one is a forced mate with a verified solution, not a scraped position with an unsound answer.
The engine
Analysis is powered by Stockfish, licensed under the GNU General Public License v3. It is compiled to WebAssembly and runs in your browser tab, so no position is sent anywhere to be evaluated. The build we ship comes from the stockfish.js package, and its source is available from the Stockfish repository. Mated is not affiliated with the Stockfish project.
See it on your own games
The tools and the puzzles are free forever. Review a game right now. It takes about twenty seconds and no account.