Forums

Can Analysis tool on chess.com find this mate?

Sort:
jamerolle2

Can the Analysis tool on chess.com find this mate in the time it takes to review a game?   If it can not, then can we really trust the % accuracy given after the post-game review function? 

[FEN "3K4/2p2p2/2P2P2/4R3/R1p5/1prp2p1/1prp2P1/1kbB4 w - - 0 0"]


3K4/2p2p2/2P2P2/4R3/R1p5/1prp2p1/1prp2P1/1kbB4 w - - 0 0

{

CraigIreland
Stockfish 16 Lite knows it's a White win with a little help but doesn't find the mate within 70 moves.
Komodo 3.3 gets stuck in repetition.
 
Martin_Stahl

Both Stockfish 16 versions show it's 0.0 at depth 99, as does both Torch versions

CraigIreland

This is the fastest I can find in cooperation with the engine but it doesn't get there on its own. While Black's moves are forced, White has way too many options for the engine to evaluate and the best are too likely to be pruned for it to find the mate efficiently.

I'd love to know what Super GMs experience of this puzzle is like. Perhaps they go straight to the ideas or perhaps it's very challenging for them.

Very cool nevertheless. Thanks for posting.

Rocky64

This is a mate-in-23 composition, where the play is far too sophisticated and long for Stockfish to solve. Here's the solution. White must avoid stalemate and at the same time prevent Black's b2-pawn from promoting.

jamerolle2

So, can we say, the accuracy % given on post game review here on chess.com is not necessarily an accurate assessment? It appears there should be a disclaimer on that feature.

spacecatchess2007

I think the engine cant see a checkmate after a certain amount of moves

CraigIreland

You don't need to go to these lengths to demonstrate that the accuracy rating isn't very precise. It does a quick low depth search. It's no substitute for doing the analysis ourselves but for most players it'll give a good indication of how well we played.

jamerolle2
Rocky64 wrote:

This is a mate-in-23 composition, where the play is far too sophisticated and long for Stockfish to solve. Here's the solution. White must avoid stalemate and at the same time prevent Black's b2-pawn from promoting.

 
 
 0 
#3

Both Stockfish 16 versions show it's 0.0 at depth 99, as does both Torch versions

 

Is this analysis on your personal computer or is this chess.com server engine cluster working? Reminding all, I am evaluating the accuracy of the cluster engine being used by chess.com for post-game analysis.

Martin_Stahl
jamerolle2 wrote:
Rocky64 wrote:

This is a mate-in-23 composition, where the play is far too sophisticated and long for Stockfish to solve. Here's the solution. White must avoid stalemate and at the same time prevent Black's b2-pawn from promoting.

 
 
 0 
#3

Both Stockfish 16 versions show it's 0.0 at depth 99, as does both Torch versions

 

Is this analysis on your personal computer or is this chess.com server engine cluster working? Reminding all, I am evaluating the accuracy of the cluster engine being used by chess.com for post-game analysis.

Site engines running in the browser and positional cloud side analysis.

Game Review uses Stockfish 16 NNUE.

Martin_Stahl

That said, the Accuracy metric is not a one-to-one engine match algorithm.

https://support.chess.com/article/1135-what-is-accuracy-in-analysis-how-is-it-measured