How we check your games

A chess engine checks every move you played. A deeper search checks every move it flags a second time. Then fixed rules, written by us, sort those mistakes into patterns. Below is each step, and one real position from a report, so you can check it yourself.

1. The engine checks every move

We use Stockfish, a free and open-source chess engine. For each position in your games it gives an evaluation: how far ahead one side is, counted in pawns. We read that number before and after each of your moves. If it got worse for you, the size of the drop is what the move cost.

Depth is how many moves ahead the engine looks. Every move you played is searched to depth 8. Paid reports can choose depth 12, which looks further ahead and takes longer.

2. Every flagged move is checked again, more deeply

A fast search can be wrong about a sharp position. So any move that looks like it lost 0.6 pawns or more is searched again, before and after the move, to depth 14. The deeper numbers replace the first ones. A move only counts as an error if it still lost 1.0 pawns or more after that second look, and as a blunder at 2.0 or more.

3. Fixed rules sort mistakes into patterns

A pattern is a kind of mistake that can repeat, such as leaving a piece where it can be taken for free. Each pattern has a rule, written by us, that checks the board after a move the engine measured as a loss. No AI decides which pattern a mistake belongs to.

Each move counts toward one pattern only, so one bad move is never counted twice. A single loss bigger than 10 pawns, which is a little more than a queen, is counted as 10, so one disaster can't outweigh a habit.

Patterns are ranked by three things: in how many of your games the pattern shows up, how much it cost you on average each time, and whether it came back in two or more different games. We do not estimate rating points.

One real example

This position is from a real report, with the names removed. The player had White, and it is White's 11th move. On the move before, Black's pawn on b6 had captured White's pawn on a5.

Before White's 11th move
The engine's better move, c5
The green arrow is the engine's better move: the pawn on c4 moves forward to c5.

The move played: Rxa5. White's rook on a1 took the pawn on a5. But Black's queen on d8 can reach a5 along the diagonal, through c7 and b6, and no White piece can take it back there. So after the queen takes the rook, White has given up a rook, worth about 5 pawns, for one pawn.

The numbers: before the move, White was ahead by 1.32 pawns. After Rxa5, Black was ahead by 3.34. That is a drop of 4.66 pawns, well past the 2.0 that makes a blunder. The engine's best move was c5, which kept White ahead.

Checking our own work: the report said this move cost 5.0 pawns. We checked it again with Stockfish 17.1 at depth 20, a much deeper search than the report uses, and got 4.66. A deeper search moves the exact number a little; here it did not change the verdict or the better move. You can set up the position in any chess engine and see for yourself that Rxa5 loses the rook.

If we get it wrong

If you bought a report and it was not useful, email us within 7 days for a full refund. The same goes if you think something on it is wrong. Tell us what you found; a person reads every message.

See what the engine finds in your own games. Your first 10 games are free, with no account.

Run my free report

Frequently asked questions

Which engine checks my games, and how deeply?

Stockfish, a free and open-source chess engine. Every move you played is searched to depth 8. Any move that looks like it lost 0.6 pawns or more is searched again, both before and after the move, to depth 14, and the deeper numbers replace the first ones. Paid reports can choose a deeper first search, depth 12.

Who decides which pattern a mistake belongs to?

Fixed rules written by us, not an AI. A rule only looks at a move the engine has already measured as a loss, then checks the board: for example, whether one of your pieces could be taken for free. Each move counts toward one pattern only.

How are my patterns ranked?

By three things: in how many of your games the pattern shows up, how much it cost you on average each time, measured in pawns, and whether it came back in two or more different games. We do not estimate rating points.

Is anything on the report guessed?

No. Every number comes from the engine's search of your own games. If there is not enough data for a card, the card is left out rather than filled in.