About the Szachmaty project

Szachmaty was built by Igor Biały — a club-level amateur, rated around 1000 on chess.com and Lichess. He built this platform first for himself, to learn faster from his own mistakes, then opened it up to anyone learning chess from scratch or coming back to it.

In short: this isn't a commercial chess company or a team of grandmasters — it's an amateur's tool for amateurs, with its methodology and limits stated plainly below. The Stockfish engine calculates the variations, but it's heuristics — not a separate judge — that name your mistake in plain language.

Who built this project

Igor Biały plays chess as an amateur, rated around 1000 on chess.com and Lichess — exactly the level of a player learning or returning to the game, the same group this project is aimed at. Szachmaty isn't the work of a coaching staff or a chess publisher — it's a self-built app, made with exactly what was missing from his own learning: fast feedback after your own games, in your own language, without paying or creating an account.

How the key-mistake detection works

After importing a game, the Stockfish 18 engine (a lightweight build, running locally in your browser, never sending your game to any server) analyzes every move within a time budget — faster in quick mode, longer in deep mode, always interruptible. Each move gets a loss score relative to the engine's best move and lands in one of five categories (best, good, inaccuracy, mistake, blunder) — the same threshold system Lichess uses. Among the moves classified as inaccuracy, mistake, or blunder, the app picks one "key mistake": first by the most severe category, then by the largest loss, and if still tied, by the earliest moment in the game.

How the plain-language explanation is built

The explanation is NOT an independent, second tactical analysis — it's a set of pattern detectors (hanging piece, fork, discovered attack, pin, allowing mate, a bad trade) checked against what the engine already showed. When no pattern matches, the app falls back to more general positional pointers (an early queen sortie, the same piece moved twice, king safety, pawn structure), and as a last resort states the raw loss in numbers. The same mechanism picks the 3 practice puzzles matched to the recognized theme — and when no theme could be recognized, the app says so directly instead of faking a good match.

Limits worth knowing

The explanations are heuristic: they name typical mistakes accurately, but they won't replace a coach. Pattern detectors only check what the engine's recommended reply points to — they don't run their own independent tactical search. Quick and deep analysis both have a time budget (not infinite depth), so in very complex positions the evaluation may differ slightly from a longer analysis run outside the app. More on the analysis limits themselves: Game analysis.

Got a question, feedback, or found a mistake in the content? Write to mail@szachmaty.pl.

The app works offline after your first online visit on this device. No installation required.