P. S.
In my run, the anomalies were spotted at around the 42’nd and 46’th depths.
@Hott said in #21:
> P. S.
>
> In my run, the anomalies were spotted at around the 42’nd and 46’th depths.
Doesn't happen for me. The distance to mate for the best line is steadily decreasing. I suspect that you are misinterpreting the multiple line analysis.
Hrm. I always looked at the value in the top left corner of the engine panel.
@Hott said in #23:
> Hrm. I always looked at the value in the top left corner of the engine panel.
Are you sure? In you first post you claimed that the engine said #23. That means "checkmate for white in 23 moves or less". In your FEN its black to move, and black is winning. My engine says #-26 at depth 65 which means black has a checkmate in 26 moves or less.
I get "#23" for second variant starting 1. ... Ke5, "#-26" for the best starting "1. ... bxa4". Could the problem be that
@Hott missed the difference in sign (which completely changes the outcome)?
BtW, at levels 42 and 46, I still had evaluation of -7.7 or so, forced checkmate (by black) only appeared later and the number of moves kept decreasing monotonically.
From what I recall, it all went somewhat like this:
I have spotted the thing (the First time) on 25’th of December in the puzzle #3GiCS, and forgot it. Then, on 26’th, I encountered it again (the Second time) in some other position, and decided to ask about this here. But at that moment I was hoping to find the First position, because I remembered that it exhibited the phenomenon more clearly than the Second; thus I didn’t post the Second. Today, on 27’th, I discovered that puzzle history exists, quickly found the First position and posted it here. So the numbers in the original post probably come from the analysis of the Second position.
When looking at the second line in the multi line analysis the distance to mate (for white) can increase, even for the same move. I tested it with Stockfish outside Lichess, and it is indeed the case.
E.g:
Depth 42: Ke5 +M48
Depth 43: Ke5 +M52
I don't know if this means that +M48 is wrong. That would be surprising. The alternative is that Stockfish can find the mate in 48 moves at depth 42, but not at depth 43, but that would also be surprising.
Here is the recording:
drive.google.com/file/d/117t4UMWiQaqO7aGj0emzoyw9zqK-_8zM/view?usp=drive_linkat 8′43′′ it founds #−53
then at 10′15′′ it claims #−55
twelve seconds later it’s #−57
then monotonically decreases to #−40
at 12′48′′ jumps to #−45
[...]
That counts two; there is at least one ahead, and ma-a-a-ybe even more, but I don’t feel the need to transcribe it further. Though I will do if you ask.
Well, that proves it. To me it looks broken, or there is something fundamental I don't understand about chess engines. I would like to know.
I'm interested to see if schlawg can come up with something.
Just the same as
@Mendelfist.