Free online draughts server. Play draughts in a clean interface. No registration, no ads, no plugins required. Play draughts in tournaments, with friends, or with random opponents.
Play
Create a gameTournamentSimultaneous exhibitions
Learn
PuzzlesCoordinatesStudyRules & Variants
Watch
Lidraughts TVCurrent gamesBroadcasts (beta)
Community
PlayersTeamsForumQuestions & Answers
Tools
Analysis boardBoard editorImport gameAdvanced search
Sign in
Reconnecting
  1. Forum
  2. General Draughts Discussion
  3. Scan's endgame skills could and should be improved

I was testing Scan's endgame technique in International draughts by comparing human approach and AI approach.

There is a nice puzzle by Noome and Hortsman (I quote their names by cyrillic retranslation, and I apologize beforehand if I did it wrongly):

[FEN "W:W15,25,26,K27:BK14,17"]

Puzzle composers do offer very direct approach:

1. 27-49 17-22 {1. .. 14-28 2. 49-43 17-22 3. 16-21 white will clearly crown the checker and win via standard endgame} 2. 49-38 14--23 3. 38-47 22-27 {Otherwise white intends to attack 4. 47-36 and lead checker 26 towards crowning} 4. 47-36 27-32 5. 36-47 32-37 6. 26-21 23-12 7. 21-16 12-23 8. 16-11 23-18 9. 11-6 18-36 10. 6-1 37-41 11. 1-18! 36x4 12. 47x36 2-0

However, Scan decides to move white's king to 49 (which is key square) only in 3rd move, proposing 1. 27-16 17-22 2. 16-21 14-23 3. 21-49, but increases evaluation from initial +3.3 to +5.6 after immediate 1. 27-49.

I hope that this nice example shows that human approach can be sometimes more efficient.

Nice study, I find such positions to be very difficult (but then again I'm not a very good draughts player). I put your line in a study: lidraughts.org/study/pSUHzSDH

Maybe the author of Scan can comment on this (account Scan_X), but I think it's mostly a matter of search depth. If I put this position in the analysis board, with Scan on infinite analysis, it finds 27-49 as the best move around depth 22/23.

I'm not sure it would play all the same moves as in the line of the authors, but given enough depth it should be able to play this position (close to) optimally, especially with endgame tablebases.

Now you may look also into my study lidraughts.org/study/YiAZbWM6 .

Scan evaluates drawn endgame with -6.6 :) Please, see lidraughts.org/study/YiAZbWM6/uObpYzhs

Yes, that's strange, the evaluation should be +6.6, as it is on the analysis board and when you analyse this pdn as an imported game. It seems to be a bug with analyzing studies from the black side, I'll go fix that.

I'm not that surprised Scan doesn't see the draw here by the way. There are many examples like this from chess as well, fortress positions that cannot be won, but are given a huge plus score by the engine. Nice position though!

Here is a nice puzzle that Scan was unable to solve even at depth 37: lidraughts.org/study/foIbE6cs/FEuzIPfi (I dare to say that any upper-intermediate player can solve it easily).

Scan fails to see winning plan in famous Scouppe position: lidraughts.org/study/CNOQ5MNQ/TNY3BYaL .