8

I'm curious what checkmating move is the most common in the Lichess database? That is, the properly algebraically specified move, for a given color, like "White plays Re8#".

Answers for other databases are also welcome.

(I'm guessing that Re8# for white may well be one of the top, as well as Qf7)

Improve this question
5
  • 1
    If you have programming experience, Python has plenty of chess libraries and lichess provides free dumps. If that's not fast enough there's C++ and Rust libraries too.
    – qwr
    Commented Mar 20 at 1:11
  • I'm curious as to why you want this. In games between decent players (and more so among better ones) an actual checkmate will be pretty rare. So, you will be collecting data mostly on weak players.
    – Peter Flom
    Commented Mar 20 at 10:20
  • 1
    Just curiosity. And your observation about checkmate holds true for classical matches between high rated opponents, but decreasingly less so at shorter time controls and lower rated opponents, which makes up the vast majority of the lichess database.
    – Steve Bennett
    Commented Mar 21 at 1:45
  • 1
    I have posted in chess meta on this chess.meta.stackexchange.com/questions/1220/…. Please opine
    – Laska
    Commented Mar 24 at 3:43
  • 1
    I am glad the question has been reopened. But the broader problems with inappropriate closure remain unanswered. I have added an EDIT to my question at chess.meta.stackexchange.com/questions/1220/…
    – Laska
    Commented Mar 25 at 7:31

1 Answer 1

Reset to default
12
+50

I've downloaded the Lichess database from 2013 to the end of 2022, and counted the final move of all the games, including whether it was a check or checkmate. For space reasons, I processed this in a streaming fashion (i.e not keeping any data after it was downloaded and processed), and I had forgotten to include recording whether the final move was by black or white; the only things I know are the number of games where the final move played was X. It's broken down by month, as well as collected into a grand total.

You can see the raw stats on the GitHub repo next to the counting program, and here's a Google Sheet with the final data and some sample calculations. In total, there are 4,004,175,546 recorded rated games in this period.

Here are a few highlights:

  • Only 26.5% of games ended in a checkmate, with the remaining 15.19% ending with a check and 58.31% with neither; thus, 73.5% of games ended in a timeout, draw or resignation.
  • The most popular game terminations are Q(x)g2# and Q(x)g7#: together, these account for a whole 3.03% of games and 11.43% of checkmates.
  • Of the checkmates, the most popular one is with the queen: these account for 64.78% of all checkmates. (Note that moving the pawn to the final rank to deliver checkmate counts as a pawn move, not the move of the promoted-to piece.) The other popular option is with the rook, which accounts for another 25.39%. All the other pieces combined account for only 9.83% of checkmates, with a king move only delivering 0.0258% of them (exclusively by means of a discovered attack, because the king itself cannot cause a check).
  • 32.43% of checkmates were also captures, and this is fairly consistent across the pieces: from 37.16% for the queen to 18.81% for the knight, it seems to hover somewhere around 20-30%. An exception is the king, which already has a rather small number of checkmates, but only 3.87% of those also capture a piece.
  • 1.88% of checkmates were caused by a promotion, most often into a queen ??=Q# (in 92.25% cases) but occasionally into a rook ??=R# (7.53%).
  • Checkmating by a castling move is very rare: this only happened 39,026 times in the data, or 0.000974%; and among these, O-O-O# is 67.01% while O-O# is half as frequent with 32.98%.
  • There are a lot of extremely rare game endings: there are 2058 different final moves that were each played only once (and 798 of them are checkmates). Most of these seem to be fully-specified: for example, only one game ended in Qe5xd4#. These rare moves account for a substantial amount of my data: I've collected 22661 unique game-ending moves, and there's 5322 (23%) moves that concluded less than 10 games, and 9028 (40%) that finished less than 100 games; and among the latter, there was only 51975 checkmates, which is less than 0.0049% of all recorded checkmates.

I haven't looked at the squares involved too much, but I imagine that there would be some more insights to be found there.

Improve this answer
6
  • 1
    Any cases of =N#? And how rare is en passant mate?
    – Rewan Demontay
    Commented Mar 23 at 13:22
  • 2
    @RewanDemontay =N# happened 23194 times, or 0.001828% of checkmates. I don't have info on enpassant mates because enpassant is notated as a regular pawn move, and you'd need to know the state of the board before the move to know if it's an enpassant or regular capture.
    – Danya02
    Commented Mar 23 at 16:18
  • 1
    Awesome answer, thanks. Are you missing draws in your 73.5% calculation? Also, is there a way you can share a longer list of the unique game-ending checkmate moves, perhaps with links to the games in question?
    – Steve Bennett
    Commented Mar 24 at 21:42
  • 2
    @SteveBennett Like I said, the amount of information in this analysis is very limited, so in order to extract more info than you see in the Google Sheet you'd basically need to rerun the analysis from scratch. I'm currently downloading the Lichess database locally (I bought the HDDs for it), so I can extract more info from it for my own projects, and when that's done I may be able to add more info here.
    – Danya02
    Commented Mar 24 at 21:46
  • 1
    So sorry, I completely missed the links. My bad.
    – Steve Bennett
    Commented Mar 24 at 22:46

Not the answer you're looking for? Browse other questions tagged or ask your own question.