lichess.org
Donate

Forced checkmate sequence can increase in length as analysis goes deeper

So “forced” means “irresistible in practice” but not “irresistible in theory”. Is that right?
been a while that most exhaustive search engine are not within strict alpha beta... I thought I mentioned that by talking about late move reduction earlier. I think "go to Discord" reflex is not best solution, as we like to stay in chess land, and not have to make sense to devs in their language, but stay in chess interpretable language. I suggest some of the Discord folks come to chess instead.

Edit: also even if it were strict alpha-beta pruning, not being a complete tree means leaf evaluations are part of the search tree that are not legal outcomes. Each a priori new encountered mate being legal and strict mate immediate positions (not problem there) would likely (and not guessed-ly, just that partial trees and probabilities in big chess world instills prudence in my speech) have to be backpropagated min-max-ly based on those other leaf evaluations.

And if those are only as cheap as possible to gain ELO via speed to reach depths over width in general, then a forced notion is only as good as combined leaf evaluations can be. But then, we could hope to improve the leaf evaluations somehow.

So, I would conclude, probably and guessed, that divination is not yet here... do try to look through engine goggles more, if only we could have that easily without contorting to command line stuff, and bring the knowers (if existing) to talk in chess (be it using some math. it might help taking the time to use codeless mathematics perhaps, using pictures, seen some good ones lately myself.... from over devland...).
@Hott said in #51:
> So “forced” means “irresistible in practice” but not “irresistible in theory”. Is that right?

pretty much, but I think at those depths, a human might have some clues to resign before then.

Although I would refrain from using the word theory in chess context.. It has a diversity of meanings not all of which I am clear about, still.
@schlawg said in #50:
> Stockfish mate results are viewed as guarantees in practice, but are not necessarily proven. Numerous examples (such as this one) demonstrate they can be ephemeral even within a single search.

Having those two statements in the same paragraph might confuse some people. Stockfish is very careful not to return wrong mates but of course there might be a bug in the code that no one found yet and that is why Vondele said "there is no rock-solid guarantee".

But it has to be clear that losing a mate does not mean that the previous mate was wrong.

discord.com/channels/435943710472011776/882956631514689597/1103746641409867839
> Vizvezdenec — 04/05/2023 20:15
> mate 9 for stockfish is at least mate in 9
> sf has a lot of guarding against false mates
> maybe even in places where it's not really needed
> so if it reports mate in 9 you can be 99,999% sure it's actual mate in 9
> I don't think I've seen a single case of sf false mate since I've started to get interested in computer chess, so for 6-7 years
> it can even sometimes lose mating scores cause it doesn't trust it own transposition table if it returns mate scores

To prove that there is a bug in Stockfish you have to find an example where Stockfish says mate in at most X and in reality it must be mate in at least >X.

This topic has been archived and can no longer be replied to.