Updated 23 August 2026
What is chess decision training?
Most chess tools tell you what the best move was; decision training is about improving the moment you choose, not the autopsy afterwards. You stay on the clock of a real position. You make a move you would make over the board. Then you see whether the engine agreed, and if it did not, you live with the position you created.
How the dual-board method works
You and the engine answer the same question at the same time, your answer on the left board, the engine's on the right, so every difference between them is a concrete, inspectable disagreement. Book moves play first when you start from an opening. After that, every user move is compared with Stockfish's top choice at the same depth. Matches pass quietly. Divergences open an explanation you must acknowledge before play continues.
Why there is no undo
An undo button turns a decision into a draft, and drafts teach nothing about consequence. After a weaker move, the engine board adopts your position and answers from there. The lesson is not "what would have been best in a game you did not play". It is what this position now asks of you.
What the explanations are built from
Every explanation is assembled from verified facts: engine evaluation, principal variation, and deterministic detections like pins, forks, and hanging pieces. A language model is only allowed to phrase them. Quality uses winning-chance buckets, not raw centipawns on the page. If no language-model key is configured, the same facts are composed on your device.
FAQ
What is Pathwise?
Pathwise is a decision trainer, not a puzzle set. You play a real position on one board while the engine plays its strongest move on a second board. When your choices differ, you get an explanation grounded in the engine's evaluation and verified tactical facts, then you keep playing from the position you actually created.
Why is there no undo?
Because taking a move back erases the lesson. Your move is permanent, exactly as it would be over the board. The engine board shows you the stronger path, the explanation tells you why it is stronger, and then the engine board adopts *your* position so the comparison always starts from reality.
Are the explanations trustworthy?
Every chess claim comes from Stockfish evaluations and deterministic detectors: pins, forks, hanging pieces, king safety, pawn structure, threats. A language model only phrases those verified facts at your chosen level; it never invents pieces, moves, or threats. Without a language model configured, the same facts are composed into a plain template explanation.
What do I need to run it?
A modern browser. The engine runs locally in your tab via WebAssembly; your positions are analysed on your device. Only when an explanation is written does a small packet of position facts go to our server, with no username, no history, nothing identifying.