How we check what we publish

The review runs Stockfish 18 inside your browser, so your games never leave your computer. By default it uses the Fast setting: every position searched to depth 10, and up to 8 of the moves that lost the most searched again to depth 14. Balanced (12/16) and Deep (14/20) are in Settings for a game you care about.

A move is graded by how much of your winning chance it gave away, the way Lichess measures it: 10 to 20 points is an inaccuracy, 20 to 30 a mistake, more than 30 a blunder. The full list of grades is in the guide.

What that does not promise: a quick search can misjudge a close call, and because the engine uses several threads, two runs of the same game can differ slightly. Clear mistakes are found reliably; the line between an inaccuracy and a mistake is where the deeper settings earn their wait.

Each of the 58 opening pages was checked by Stockfish at depth 24: the six best moves from the opening's position, shown on the page, and every trap line evaluated move by move. Anything the text claims about a position is held to that stored verdict by a test, so the wording cannot quietly drift away from what the engine says.

That check found real errors in our own pages. On 10 October 2026 we made 14 corrections: lines where the “mistake” was in fact a piece left hanging that the text never mentioned, a famous trick that does not work in the position shown, and claims that went further than the engine does. Each was re-checked at depth 22 before it was rewritten.

Every diagram on the rules pages and in the lessons makes a claim in its caption: castling is allowed here, this capture is en passant, this is stalemate, nothing protects this knight. Each claim is checked with chess.js, an independent implementation of the rules, every time the site changes. A mistyped position fails a test before it can reach someone learning the game.

The rules pages follow the FIDE Laws of Chess and cite the article each answer rests on.

The games in the blog are analysed with Stockfish to depth 16 for every move and depth 24 for the moves that lost the most ground. The accuracy figures and the moves a post singles out are checked against that stored analysis by tests, and every post draws its facts from more than one outlet.

An engine's number is its best estimate at the depth it searched, not a proof, except where it has found a forced mate. That is why the pages here say “the engine rates this about level” rather than “this is equal”, and why a position described as better for one side is a judgement with a depth attached to it.