The bot didn't know how to checkmate with a king and a rook and after a lot of moves the bot saw that the game was going to be a draw and resigned
https://lichess.org/q0zDaKX9jsp5
The bot didn't know how to checkmate with a king and a rook and after a lot of moves the bot saw that the game was going to be a draw and resigned
https://lichess.org/q0zDaKX9jsp5
It resigned on move 151... that is hard programmed by Lichess. (for reasons of preventing database and resource cascade issues in some fringe cases, the site needed to put in a hard cap on number of moves in these Bot games.)
That level 4 can not win this (unless it gets lucky) is easy enough to understand. At level 4, it has a hard cap on its horizon. It looks only 5 ply (3 of its moves and 2 of yours) deep. The code at this level hard-forbids it for looking deeper.
At say move 130, your K is in the centre, and it has a K somewhere near, and a R somewhere. Is king is centralised enough that it has full mobility, and on an open board, the K always has 14 mobility as long as it is not blocked by its own K.... From a static analysis, you can move the K around a bit, or move the R to more than half the square on the board, an the static evaluation remains the same (at least to within rounding errors).
Since it can only look 3 of its moves deeps, it calculates tons of lines for 3 moves, and finds that they lead to slightly different positions, all of which are statically identical in value, so it picks one at random. Unless it by shear luck happens to randomly guess a few good ones in series, it never makes any progress.
It resigned on move 151... that is hard programmed by Lichess. (for reasons of preventing database and resource cascade issues in some fringe cases, the site needed to put in a hard cap on number of moves in these Bot games.)
That level 4 can not win this (unless it gets lucky) is easy enough to understand. At level 4, it has a hard cap on its horizon. It looks only 5 ply (3 of its moves and 2 of yours) deep. The code at this level hard-forbids it for looking deeper.
At say move 130, your K is in the centre, and it has a K somewhere near, and a R somewhere. Is king is centralised enough that it has full mobility, and on an open board, the K always has 14 mobility as long as it is not blocked by its own K.... From a static analysis, you can move the K around a bit, or move the R to more than half the square on the board, an the static evaluation remains the same (at least to within rounding errors).
Since it can only look 3 of its moves deeps, it calculates tons of lines for 3 moves, and finds that they lead to slightly different positions, all of which are statically identical in value, so it picks one at random. Unless it by shear luck happens to randomly guess a few good ones in series, it never makes any progress.
@woodfantasy said ^
That’s a pretty funny way for the bot to resign I didn’t realize level 4 had such a short search horizon, so the random rook moves make more sense. K+R mate looks easy until you’re the one trying to force it. I sometimes use ChessMinds for a quick second look when an engine result feels stranger than the position.
Did you join lichess.org for the sole purpose of advertising ChessMinds? ATP you don't "sometimes" use it, you ALWAYS use it. T_T
@woodfantasy said [^](/forum/redirect/post/Oyo3wg7u)
> That’s a pretty funny way for the bot to resign I didn’t realize level 4 had such a short search horizon, so the random rook moves make more sense. K+R mate looks easy until you’re the one trying to force it. I sometimes use ChessMinds for a quick second look when an engine result feels stranger than the position.
Did you join lichess.org for the sole purpose of advertising ChessMinds? ATP you don't "sometimes" use it, you ALWAYS use it. T_T