Arena 3.5: No Adjudicated TB ending?

[Matthias Gemuh]

(i.e. castle, pawn move, capture)

I hope not! In FIDE rules castling doesn't reset the reversible-move counter.

Frankly, I would have to look at ChessGUI source code to see how I handle it there, but in my chess engines, castling is irreversible.

[hgm]

I hope not! In FIDE rules castling doesn't reset the reversible-move counter.

Are you sure? I always thought that by definition irreversible move is one that cannot be repeated later, thus castling logically should be irreversible. What about 3-fold repetition and castling?

Yes, I am sure. This issue has come up many times before. I guess 'reversible-move counter' is really a misnomer.

Castling rights do count in deciding whether a position is repeated. But they do not reset the move count. Which makes sense if you think about it: castling cannot really be considered progress towards winning. Simplifying or pushing a Pawn towards promotion can.

It is actually quite troublesome to find a suitable replacement for the 50-move rule in games where reversible pieces can promote.

[mar]

Yes, I am sure. This issue has come up many times before. I guess 'reversible-move counter' is really a misnomer.

Castling rights do count in deciding whether a position is repeated. But they do not reset the move count. Which makes sense if you think about it: castling cannot really be considered progress towards winning. Simplifying or pushing a Pawn towards promotion can.

Ok thank you and Miguel for clarification. I learned something new and already fixed this in my program. I kept my code that resets repetition hash on fifty counter reset though as in 99.99% cases castling occurs early.

[RJN]

So far, I have had 25 games (out of about 70) that did adjudicate, so it is a nice feature to save time. However, every single game that was adjudicated via TB has been a drawn game, no wins/losses, which makes me suspicious.

I have not manually been through the PGNs to see if there are other won/loss games at 5 pieces not being adjudicated, I just happened to see that game I posted here.

Does anyone know if SCID (or whatever utility) has a way to filter games by the total number of pieces left at the end of game?

TIA

[hgm]

Of course I cannot know if this is intentional or a bug, but I would consider it very smart of Arena if it only adjudicated draws. For engines that do not use EGTs adjudicating wins would corrupt test results, as you might be awarding wins to engines that would not know how to win. While for engines that do use EGTs adjudicating gains you nothing, as they should move instantly, as the EGT tells them the best move in the root (be it shortes mate or best defense). Only draws between EGT using engines are a nuisance, where a draw is guaranteed, but they try to swindle each other.

[syzygy]

Of course I cannot know if this is intentional or a bug, but I would consider it very smart of Arena if it only adjudicated draws. For engines that do not use EGTs adjudicating wins would corrupt test results, as you might be awarding wins to engines that would not know how to win.

Why wouldn't the same apply to draws? Lots of tablebase draws are difficult enough that (at least some) engines won't be able to hold the draw.

[hgm]

I guess you are right. But in any case it removes 2/3 of the problem, where nothing was to be gained in terms of speed. In WinBoard I don't even adjudicate the draws; I just limit the search depth of designated engines (presumably the EGT-using ones) to make them move instantly. So a non-privileged engine would still have to play for the draw, against an EGT-using one that just plays by intuition, and of course would immediately punish any mistake that would cost the game. That should solve the problem of CPU wasting with minimal corruption of the test results.

[RJN]

Of course I cannot know if this is intentional or a bug, but I would consider it very smart of Arena if it only adjudicated draws. For engines that do not use EGTs adjudicating wins would corrupt test results, as you might be awarding wins to engines that would not know how to win. While for engines that do use EGTs adjudicating gains you nothing, as they should move instantly, as the EGT tells them the best move in the root (be it shortes mate or best defense). Only draws between EGT using engines are a nuisance, where a draw is guaranteed, but they try to swindle each other.

Thanks for your take, it makes sense, and I agree with that, but was curious since Arena 3.5 was not very specific. If I recall, TCEC (ChessGUI) adjudicated all 5-piece games (draws, and wins with DTM<30, I just looked it up), so maybe I was expecting similar. The way TCEC did it looks like a good method of making sure an engine does not get a freebie win.

I have no problem with Arena 3.5 only stopping 5-piece draws, it's just that I would like to know for sure if that's the way it works, without guessing. If I could find more instances of 5-piece wins not being adjudicated in my PGN, I could at least make an educated guess. What a pain to go through it all manually. Maybe I will email and ask, but I thought the Arena authors read this forum and would chime in.