What is new
Since 18 September 2026, Board Analysis on ChessRank.org uses Syzygy endgame tablebases. Whenever there are seven pieces or fewer on the board, kings included, we look the position up in the tablebase. When the answer arrives, the panel stops showing an engine estimate and shows the exact result instead: White Win, Black Win or Draw, together with the moves that keep it.
For these endgame positions this makes the analysis faster and more precise. Faster, because the result is looked up rather than calculated, so you do not have to wait for the engine to search deeper. More precise, because it is not an estimate at all. It is the proven result of the position.
Stockfish still starts on every position exactly as before. With more than seven pieces on the board nothing changes, and if the tablebase cannot be reached, for example when you are offline, the board simply stays on Stockfish.
Engine versus tablebase
Stockfish searches a limited number of moves ahead and then judges the positions it reaches. Its evaluation is an estimate: a very good one, but it can still change as the engine searches deeper. In endgames that matters, because a large material advantage sometimes cannot be converted at all. King, bishop and a rook's pawn against a lone king, with a bishop that does not control the promotion square and the defending king already standing in that corner, is a draw, although White is a whole bishop and a pawn up.
A tablebase works the other way round. For a given set of pieces, for example king and rook against king and knight, a computer goes through every legal position those pieces can form and determines the result of each one with perfect play from both sides. Nothing is estimated and nothing is left out.
How a tablebase is built
The method is called retrograde analysis. It starts from the positions whose result is already known: checkmates, stalemates, and positions where a capture or a promotion leads into another endgame that has already been solved. From there the computer works backwards:
- A position is won if the side to move has at least one move into a position that is lost for the opponent.
- A position is lost if every legal move leads into a position that is won for the opponent.
- These two steps are repeated until no new position gets a result. Every position still undecided at that point is a draw, because neither side can force a win from it.
Why the result is mathematically certain
Because every position is checked, a tablebase verdict is neither an opinion nor a probability. It is a proof by exhaustion, the kind of proof mathematicians use when a statement can only be settled by going through every case. When Board Analysis shows "White Win", it is proven that White can force the win against any defence. When it shows "Draw", it is proven that neither side can force a win.
The same is true of the moves. Every move with a green dot keeps the best result the position allows: a win stays a win, a draw stays a draw. That, too, is proven, not estimated. A grey dot marks a move that throws a win away but still draws, and a red dot marks a move that loses.
To be completely precise: like every computer-assisted proof, this one relies on the programs that generated the tables having worked correctly. That is why the Syzygy files carry checksums and the generator comes with a tool that checks the tables for internal consistency.
And two things "correct" does not mean:
- There can be several correct moves. Often more than one move keeps the win. Board Analysis lists up to four moves, best first.
- The best move is not always the fastest mate. Syzygy ranks winning moves by DTZ (distance to zeroing): the number of half-moves, with best play, until the next capture, pawn move or checkmate. That is what keeps a win safe under the 50-move rule. Another move may lead to mate sooner, so the tablebase move can differ from the one Stockfish prefers even when both of them win.
Win, draw and the cursed win
Tablebases also take the 50-move rule into account. Under this rule a game can be drawn once each player has made 50 consecutive moves without a capture or a pawn move. Over the board a player has to claim the draw. Online the rule is usually applied automatically.
Some endgames are won in theory, but the win needs a stretch of more than 50 moves without a capture or a pawn move. Syzygy calls this a cursed win: won with perfect play, but the defending side can hold a draw by the 50-move rule. Seen from the defending side, the same position is a blessed loss. Board Analysis shows a cursed win as a verdict of its own and puts the evaluation bar in the middle, because under the rules of the game it is a draw.
How long can a perfectly played endgame be? In 2012, while the seven-piece Lomonosov tablebases were being completed, a position was found in which the fastest forced mate takes 549 moves. That is far beyond anything an engine can see by searching ahead.
Who came up with it
- 1913, Ernst Zermelo. The German mathematician published a paper on the application of set theory to chess ("Über eine Anwendung der Mengenlehre auf die Theorie des Schachspiels"), often regarded as the first paper on game theory. The result named after him states that in chess either White can force a win, or Black can force a win, or both sides can force at least a draw. A tablebase gives the concrete answer, one endgame at a time.
- 1970, Thomas Ströhlein. In his doctoral thesis at the Technical University of Munich he described the first computer implementation of retrograde analysis for chess and solved several endgames with three and four pieces, among them king and queen against king and rook.
- Late 1970s and 1980s, Ken Thompson. The Bell Labs computer scientist, known as a co-creator of Unix, built tablebases for many endgames with up to five pieces. His king and queen against king and rook database became famous when grandmaster Walter Browne, playing the side with the queen against the computer's perfect defence, failed on his first attempt to win within 50 moves, even though the position was won.
- 1991, Lewis Stiller. On a massively parallel computer he solved some endgames with six pieces.
- From the late 1990s, Eugene Nalimov. His tablebases, which give the distance to mate, came into wide use in chess programs, first for all endgames with five pieces and later for all endgames with six.
- 2012, Lomonosov tablebases. Vladimir Makhnychev and Victor Zakharov of Moscow State University computed seven-piece tables on the Lomonosov supercomputer.
- 2013, Ronald de Man and Syzygy. In April 2013 Ronald de Man released the Syzygy tablebases for all endgames with up to six pieces. Instead of the distance to mate, Syzygy stores two things: the result with the 50-move rule taken into account (WDL, for win, draw, loss) and the DTZ. That makes the tables much smaller and fast enough for engines to use during their search.
- 2018, seven pieces. In August 2018 Bojun Guo and Ronald de Man completed the seven-piece Syzygy tables. It was the first complete seven-piece tablebase, because the Lomonosov tables had left out the endgames of six pieces against a lone king. The set covers more than 423 trillion positions in 18.4 TB, and Lichess made it freely available to everyone.
Where the answers come from
The seven-piece tables are far too large to download into a browser, so Board Analysis asks the free public tablebase service run by Lichess. The request contains only the position, no account or game data. Each answer is remembered for as long as the page stays open, so going back to a position you have already seen shows its result instantly.
Precise limits
- Seven pieces at most, kings included. With eight or more pieces, Board Analysis stays on Stockfish.
- No castling or en passant rights. Syzygy does not cover positions in which castling is still allowed, so Board Analysis looks the position up without castling rights, and without en passant rights as well. In a real endgame this only matters in rare cases.
- A fresh 50-move count. A position set up on the board has no move history, so it is looked up as if the 50-move count had just started. If many moves without a capture or a pawn move have already been played in your real game, a win shown here may in fact come too late to beat the 50-move rule.
What comes next
As of 18 September 2026, tablebases work in Board Analysis. We plan to extend them to other parts of ChessRank.org as well, including Game Analysis, Report, Blunder Puzzles, Play Bots, Opponent Prep and more. Since 23 September 2026 they also answer in Game Analysis, through the optional Go deeper button (switch it on in Settings, under Game Analysis). There they only show you the position and never change your report.
Try it yourself: set up any endgame with seven pieces or fewer on Board Analysis, for example king and queen against king and rook, and you will see the exact verdict instead of an engine estimate.