0.544rating-length correlation
+1,066rating: one to four moves
2median solver moves

TL;DR: Puzzle length and rating have a Pearson correlation of 0.544. One-move puzzles average 932, while four-move puzzles average 1,998.

Average rating by solution length

Average puzzle rating by moves required from the solver
1 move932
2 moves1,391
3 moves1,739
4 moves1,998
5 moves2,167
6 moves2,274
7 moves2,360
8 moves2,424
9 moves2,495
10 moves2,569

I expected the averages to become noisy fairly quickly, but they rise at every length from one through fourteen moves. There are only six puzzles beyond that, so I would not read much into the shape of the far end.

The longest puzzle is not the hardest

The Pearson correlation between solver moves and rating is 0.544. A value of 1 would mean that longer puzzles were always harder; 0 would mean no linear relationship. So length matters, but it is nowhere close to being a rule.

The highest-rated puzzle in this snapshot takes four moves, not seventeen. A short puzzle can still hide a very difficult first move.

There are not many long puzzles

Two-move puzzles make up 50.68% of the database. Another 26.24% take three moves, and 14.84% take one. Only 2.37% require five moves or more.

That imbalance matters when reading the later averages. The ten-move figure comes from 616 puzzles; the two-move figure comes from 3,091,724. The ten-move average is therefore much noisier.

Method

I analysed the official Lichess puzzle export dated 10 September 2026. The first UCI move sets up the puzzle, so I excluded it and counted the solver's alternating moves. Ratings are a changing snapshot, not permanent labels.

The full derived results are available as JSON, and the analysis script is kept with the TacticBar site.