Chess960 tactics: what changes when the patterns go
I could not find published motif frequencies for Chess960, so I counted them. This site mines its puzzles from real Chess960 games, and the pool now holds 21,019 engine-verified positions. Pins turn out to be about twice as common as in standard chess. Forks are not.
What is different about tactics in Chess960?
You calculate what you would otherwise recognise. That sounds smaller than it is. A 1600 does not find Bxh7+ by listing candidate moves. He finds it because a bishop on d3, a knight on f3 and a queen with h5 available is a shape he has met four hundred times, and the shape raises its hand before he has decided to look for anything. Chess960 keeps every rule of chess and takes away the four hundred meetings.
What it does not take away is geometry. A knight on an outpost still hits two pieces at once. A rook on an open file still catches a king with a queen behind it. So there are two honest guesses about what Chess960 tactics are, and they predict different numbers. Either the tactical content of the game is genuinely different, or it is the same content with the recall stripped out. A pool of twenty thousand puzzles is enough to tell them apart.
What did the count find?
The Chess960 side is the site's own pool: 21,019 positions mined from real games by two passes of Stockfish, one to find the blunder and one to confirm the punishment is unique and forcing. The standard side is the Lichess puzzle database. The same tagger reads both, and mate length is derived from the line rather than trusted from a label.
Two corrections matter before the numbers mean anything. The pools sit at different points of the game, median ply 25 against median ply 55, and that gap alone would produce fewer back-rank mates and promotions on the 960 side without any of it being about Chess960. So the count uses only ply 16 to 40, which leaves 16,270 puzzles on each side. The tagger also recognises the two pools at different rates, 63 per cent of the Chess960 puzzles and 86 per cent of the standard ones, so every figure below is a share of the puzzles the tagger recognised, not of all of them.
| Motif | Chess960 | Standard | Ratio |
|---|---|---|---|
| Fork | 70.7% | 78.2% | 0.90× |
| Quiet move | 25.4% | 18.1% | 1.40× |
| Back-rank mate | 14.9% | 36.4% | 0.41× |
| Sacrifice | 13.2% | 17.5% | 0.76× |
| King hunt | 12.2% | 26.9% | 0.45× |
| Skewer | 12.0% | 12.3% | 0.97× |
| Pin | 10.0% | 4.9% | 2.03× |
| Discovered attack | 1.3% | 1.6% | 0.80× |
| Double check | 0.2% | 0.7% | 0.35× |
| Promotion | 0.0% | 0.1% | 0.34× |
Four rows carry the result. Pins double. Quiet moves rise by about two fifths. Forks and skewers do not move. Back-rank mates and king hunts fall to under half, and that last pair is the one to hold loosely, for reasons set out below.
Why are pins twice as common in Chess960?
A pin wants a long line with a valuable piece at the far end of it. In standard chess the far end is usually a castled king, and a castled king is the hardest target on the board to pin anything against, because every line into g1 runs through f2, g2, h2 or the rook on f1. In the same stage-matched window the defending king stands on the g-file in 51 per cent of the standard puzzles and 33 per cent of the Chess960 ones. In Chess960 it is spread over the b, c, d and e files. A king on c1 has the whole c1 to h6 diagonal in front of it, and anything of yours standing on that diagonal is a pin waiting to be made.
NRKRBQNB · Black to playThen the bishops. In ordinary chess they start on c1 and f1 and leave along diagonals you have watched since you learned the game. In 900 of the 960 arrangements at least one bishop starts somewhere else, and in 420 of them one starts in a corner, on a1 or h1, already lined up along a diagonal that crosses the whole board.
RNBQKBNRBRNQKNRBThere is a plainer version of the same point. These puzzles are mistakes, caught. A position enters the pool because somebody played a losing move and the engine found the unique punishment. So the table is not a census of the tactics available in Chess960. It is a census of the tactics that punish Chess960 players, which is narrower and, for a trainer, more useful. Pins double partly because the lines are open and partly because nobody looks down them.
Why are back-rank mates and king hunts rarer?
A back-rank mate is not really about the back rank. It needs a king shut in by its own three pawns, a first rank swept empty, and a rook or queen that reaches it. All three are habits as much as positions, and Chess960 breaks two of them.
The box goes first. In that same window the defending king has all three squares in front of it occupied by its own pawns in 22 per cent of the standard puzzles and 15 per cent of the Chess960 ones. A bishop that starts on h1 has one diagonal and one pawn in the way, so the g-pawn moves early and often, and it moves for a reason that has nothing to do with king safety. The king that later castles short arrives behind a shell with a hole in it.
NNBRQKRBThe rank does not empty either. In 33 per cent of the Chess960 positions the defender still has four or more of his own pieces on his first rank, against 21 per cent in standard chess, because a rook that starts on b1 or a knight on e1 is often already where it wants to be and never travels. One spare piece on the first rank turns mate into check.
King hunts thin out for a related reason and for one that is not about chess at all. The tagger calls a solution a king hunt when the solver gives two or more checks in the line, and a mined 960 puzzle punishes a single blunder, so its line is usually short. That is a fact about the miner rather than about the game.
What is a quiet move, and why are there more of them?
A quiet move here means the winning move captures nothing and checks nothing. Those are the moves your eye never offers you. In a position you half recognise the eye at least hands you the captures and the checks, and you can start there. In a Chess960 middlegame it hands you very little, so a decisive silent move gets missed more often, and a missed move is what puts a position in the pool. A quarter of the tagged Chess960 puzzles turn on one.
How far should you trust these numbers?
Four things are wrong with the study, and you should know all four before you use it.
The two pools were built by different machines. Lichess builds puzzles from games real people played to the end; this site mines blunders that have a unique, forcing, stable punishment. The gap shows. About a third of the standard puzzles in that window finish in mate, against about one in ten of the Chess960 ones, and the median standard solution runs two solver moves against one here. Back-rank mate and king hunt are exactly the two tags that need a mate or a long line, and they are the two that fell furthest. I would not bet much on either ratio. The pin, quiet move, fork and skewer figures do not depend on solution length, and those I trust.
One heuristic tagger defines every motif on both sides. It calls a move a pin when the move newly and absolutely pins an enemy piece against its own king. A definition that also counted relative pins, a knight held against a queen rather than against a king, would give a different number, and I do not know which way it would move.
Thirty-seven per cent of the Chess960 puzzles match no motif at all, against 14 per cent on the standard side. That third is unmeasured. Whatever tactic is doing the work in those seven thousand positions is missing from the table, and the hole is more than twice as big on the side this article is about.
Ratings are not controlled for. The two pools are not drawn from players of matched strength, and motif frequency moves with strength.
How do you train Chess960 tactics?
Start with what needs no training. Forks and skewers appear at the same rate in both games, so the fork you learned on f7 is the fork you need on b6, and ordinary tactics puzzles keep working for it.
The rest follows the table. Find the enemy king before you calculate and name the lines that reach it, because it is on c1 more often than you expect and much less often on g1. Look along every long diagonal for one of your own pieces standing in front of something dearer. Force out one candidate move per position that neither captures nor checks. Stop scanning for back-rank mates, which here run at well under half the rate your eye is trained on. None of this replaces knowing where the pieces want to go, which is the subject of the piece on Chess960 opening strategy.
Then practise on positions that behave this way. The Chess960 puzzle trainer runs on the same mined pool the count was taken from, so its motif mix is the mix in the table. The position generator deals you an arrangement to look at cold, which is the drill for the first paragraph of this article. The printed books hold a thousand positions each in the same proportions. What none of them gives back is the four hundred meetings, and that is the point of the game.