A game is more than its result.
We select games for what they can teach a club player, roughly 800–2000 rating: clear development, patient improvement, difficult defence, active kings and the connection between pieces. Some are famous masterpieces. Others deserve attention because a quiet choice, a resilient defence or a well-played draw is useful to understand.
“Great” is an editorial judgement, not a measurable ranking. The launch collection spans romantic attacks, world championship struggles, women’s chess and rook endings. As the archive grows, the candidate queue prioritizes wider representation and contrasting ideas.
From recorded score to useful lesson
- Source. Store the game score, factual metadata and source links. Unknown dates remain unknown. Preserve factual scores without reproducing another author’s annotations.
- Validate. Replay every move with independent chess rules libraries, check special moves, compare final results and reject duplicate game histories.
- Analyse. Run pinned Stockfish 17.1 before publication. Store full position history, White-oriented evaluations, principal variations, node budgets, achieved depth, engine binary identity and NNUE identities.
- Explain. Add original commentary tied to actual moves. Ask a concrete question at each critical moment, then compare the historical move with engine-supported alternatives.
- Check and publish. Require complete analysis, legal lines, supported exercise answers and complete metadata. The scheduled workflow prepares reviewed content; it does not invent historical facts or silently publish an uncertain entry.
What engine numbers mean
A positive centipawn score favours White; a negative one favours Black. One hundred centipawns is conventionally one pawn of evaluation, not a literal promise of winning a pawn. Mate scores are separate. A finite search is not a proof of perfect play. Each lesson exposes the actual budget, depth and complete evidence file.
Three candidate lines are analysed at teaching positions. Exercises accept candidates within 50 centipawns of the best saved result from the player’s perspective. Saved winning mating alternatives also receive credit, even when they take longer. A historical move can differ from the engine’s preference. A move outside the accepted set receives an explanation of that limit, not an invented claim that it loses by force.
Authorship and AI assistance
This is an AI-assisted editorial project. The original teaching explanations were drafted with AI assistance and checked mechanically against legal moves and the saved Stockfish analysis. We do not claim review by a named human chess expert. Historical quotations are not included unless a verifiable attribution and reuse basis are supplied. No claim is made about what a player was thinking.
The move-linked explanations are the site’s modern analysis, not contemporary accounts of the games. Follow the source links on each game page to inspect the recorded score. Corrections should update the game file, retain the stable URL and explain the substantive change in the update history.
Designed for learning, wherever you are
The complete article and annotated transcript are present in the initial HTML. Interactive study adds a responsive board, hints, saved variations, personal notes and local progress. There is no account or runtime database. Your data stays on your device unless you export or share it. Offline downloads are optional and distinct from your study record.
Publication record
The launch edition was created on 20 October 2026. Each game displays its own substantive publication and modification dates. The collection is intended to grow by three carefully checked lessons per week; that target is an editorial cadence, not a claim that automated runs have already published games.
Explore the game library → · Privacy, licences and credits →