Cursed win at TCEC

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Harvey Williamson, bob

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Post Reply
Lyudmil Tsvetkov
Posts: 6036
Joined: Tue Jun 12, 2012 10:41 am

Re: Cursed win at TCEC

Post by Lyudmil Tsvetkov » Wed Nov 16, 2016 2:13 pm

Rochester wrote:
Lyudmil Tsvetkov wrote:
Rochester wrote:They gave the SYZYGY to the engine. The SYZYGY follow the FIDE ruler.

But they gave the GAVIOTA to the GUI. The GAVIOTA don't follow the FIDE ruler.

The GAVIOTA has the wrong data. It is the bug. They must give the same to engine and to GUI.

They say they follow FIDE 50 move rule. They must give SYZYGY to the GUI.

Game is draw by TCEC rules. GAVIATO is WRONG. TCEC judge is WRONG.
the end position is won for white. why adjudicate a won position as draw?
Need the 72 moves for faster mate. But the FIDE and the TCEC has the 50 move ruler. The GAVIOTA has the bug: tt don't know the 50 move ruler.

You say you want change the rule. But the rule is the rule. No exception on 50 move rulers any more. Long time ago, yes. Now no.

TCEC rule not say "GUI ask to Lyudmil".
But 'the BUG' already existed in the rules, did not it?
And it is valid for all engines, is not it?

So, I do not see a reason to change anything: bug or not bug, the end position is simply won for white and the GUI took the rigth decision, even if involuntarily.

Branko Radovanovic
Posts: 48
Joined: Sat Sep 13, 2014 2:12 pm

Re: Cursed win at TCEC

Post by Branko Radovanovic » Wed Nov 16, 2016 2:26 pm

Lyudmil Tsvetkov wrote: So, I do not see a reason to change anything: bug or not bug, the end position is simply won for white and the GUI took the rigth decision, even if involuntarily.
From www.thefreedictionary.com:

adjudicate - (Chess & Draughts) (tr) chess to determine the likely result of (a game) by counting relative value of pieces, positional strength, etc [emphasis mine]

In case of draw adjudication (0.05) and win adjudication by eval (6.50), the objective is to shorten the game by assigning the outcome that is likely to happen should the game continue. The problem with this particular case is that it produced a result that was extremely unlikely to happen in absence of adjudication.

If we accept the above TFD definition as essentially correct, it follows that the GUI took the wrong decision.

Lyudmil Tsvetkov
Posts: 6036
Joined: Tue Jun 12, 2012 10:41 am

Re: Cursed win at TCEC

Post by Lyudmil Tsvetkov » Wed Nov 16, 2016 2:38 pm

Branko Radovanovic wrote:
Lyudmil Tsvetkov wrote: So, I do not see a reason to change anything: bug or not bug, the end position is simply won for white and the GUI took the rigth decision, even if involuntarily.
From www.thefreedictionary.com:

adjudicate - (Chess & Draughts) (tr) chess to determine the likely result of (a game) by counting relative value of pieces, positional strength, etc [emphasis mine]

In case of draw adjudication (0.05) and win adjudication by eval (6.50), the objective is to shorten the game by assigning the outcome that is likely to happen should the game continue. The problem with this particular case is that it produced a result that was extremely unlikely to happen in absence of adjudication.

If we accept the above TFD definition as essentially correct, it follows that the GUI took the wrong decision.
you said yourself: by assigning the outcome that is likely to happen should the game continue; should the game have continued, unlimited by any specific rules, most likely SF would have won the game; indeed, such endgames are very easy to win for modern top engines.

User avatar
hgm
Posts: 22588
Joined: Fri Mar 10, 2006 9:06 am
Location: Amsterdam
Full name: H G Muller
Contact:

Re: Cursed win at TCEC

Post by hgm » Wed Nov 16, 2016 3:28 pm

Lyudmil Tsvetkov wrote:
hgm wrote:Actually they did write in the rules explicitly that the 50-move rule does apply.
they also wrote that Cutechess adjudicates automatically tablebase positions. what is the main rule and what the secondary one?
They did not write such adjudication could assign arbitrary results to the game that have no relation to the adjudicated position. This is sort of implied by the word 'adjudication'. If you assign a win to the bare King in KQQQK you cannot defend that by saying: "the rules stated that 5-men positions would be adjudicated".

So what happened was not adjudication. It was malfunctioning of the GUI during an attempt at adjudication.
I guess a won position is simply won, and the end position is won. FIDE rules concerning 50-moves draw is definitely a remnant from the past, and should be changed, the sooner the better.
Irrelevant. TCEC is played by FIDE rules, stupid as they might be. I'd also preferred to see that the captured pieces could be dropped back into the game, another defect of the current FIDE rules. But alas, these are the rules they play by...
So, what should be done, is for a specific engine protocol to change the 50-moves rule to a larger number, at least 100, maybe referring only to specifically known tablebase positions. The 50-moves rule migth hold true for any other case(not to make the game too long), but, in the case of a known tablebase position, the rule might be changed to 100 moves.

For example, if pawnless endgame (where primarily those positions occur) and tablebase win, 50-moves rule=2*50-moves rule. Not difficult to do, is it?
What should be done is stick to the rules as they were agreed upon. Whether better rules could be designed, or how much better they would be, is not relevant.

User avatar
hgm
Posts: 22588
Joined: Fri Mar 10, 2006 9:06 am
Location: Amsterdam
Full name: H G Muller
Contact:

Re: Cursed win at TCEC

Post by hgm » Wed Nov 16, 2016 3:31 pm

Lyudmil Tsvetkov wrote:you said yourself: by assigning the outcome that is likely to happen should the game continue; should the game have continued, unlimited by any specific rules, most likely SF would have won the game; indeed, such endgames are very easy to win for modern top engines.
Wrong. The outcome would have been most certainly a draw. Wanna bet? :roll:

basil00
Posts: 55
Joined: Thu Oct 22, 2015 12:14 am

Re: Cursed win at TCEC

Post by basil00 » Wed Nov 16, 2016 3:43 pm

whereagles wrote:Discuss :)
If Stockfish wins the tournament based on this game, then the tournament result will also be a "cursed win".

I think the result should be amended; the rules of chess should trump any legalistic interpretation of the tournament rules (which I don't really care about...)

bnemias
Posts: 370
Joined: Thu Aug 14, 2008 1:21 am
Location: Albuquerque, NM

Re: Cursed win at TCEC

Post by bnemias » Wed Nov 16, 2016 3:55 pm

All this talk of FIDE rules and TCEC rules misses the point.

I'll agree with those who say TCEC rules were followed, and the GUI should adjudicate this position. But it is wrong to feed engines that rely on TB one set that says the position is a draw, and then adjudicate based on another set that says it is a win for one side.

Debate the merit of the 50 move rule as it pertains to computer chess all you want. It is quite irrelevant to the situation.

The solution is very easy: admit this situation wasn't forseen, admit it is wrong to use different metrics for adjudication than the ones used by the engines, and amend the result so that adjudication was based on the same set the engines were using.

Branko Radovanovic
Posts: 48
Joined: Sat Sep 13, 2014 2:12 pm

Re: Cursed win at TCEC

Post by Branko Radovanovic » Wed Nov 16, 2016 4:30 pm

Lyudmil Tsvetkov wrote:you said yourself: by assigning the outcome that is likely to happen should the game continue; should the game have continued, unlimited by any specific rules, most likely SF would have won the game; indeed, such endgames are very easy to win for modern top engines.
But that is true only if "unlimited by any specific rules" meant "unlimited by the 50-move rule". Since the 50-move rule is implemented and would have taken effect automatically and without exception, that game would have surely ended in a draw.

Dirt
Posts: 2851
Joined: Wed Mar 08, 2006 9:01 pm
Location: Irvine, CA, USA

545 moves

Post by Dirt » Wed Nov 16, 2016 4:38 pm

Lyudmil Tsvetkov wrote:... what is the longest tb win without captures/promotions/pawn move?
Apparently, the longest known such win is 545 moves. There may be longer ones yet undiscovered.
Deasil is the right way to go.

S.Taylor
Posts: 7954
Joined: Thu Mar 09, 2006 2:25 am
Location: Jerusalem Israel

Re: Cursed win at TCEC

Post by S.Taylor » Wed Nov 16, 2016 4:46 pm

hgm wrote:You seem to miss the point. The position is not a win, but a draw. And both engines saw that correctly.

So the relevant question is: if the position is a draw, why declare a win?
I don't understand. The TB said the position is a win. And the TB saw that BEFORE the 50 moves ended.
End of story!

THIS is how things have been adjudicated until now.
No one cared what the engines say and no one cared what human common sense say.
And as Cato said very clearly, TCEC allows use of TB's.
End of story!

(It's not me, it's Cato and most other members both here and in TCEC... UNTIL now! The question is, are we going to change this right now? Right in the milddle of this superfinal? If so, who gives us permission?)

Post Reply