Something unnacceptable in TCEC?

[S.Taylor]

Game 17, Komodo-Laser
I was watching every secoind of it at the end, and the engines did NOT agree it was a win for White (Komodo).
Someone in the the comments section in chessbomb said it was a tablebase win in 53 more moves.
And indeed, Komodo seemed to give that TB verdict by evaluating it as about 280 pawns up. However Lasers evaluation did not even reach 3, as in 3 pawns up.
I thought that both machines had to to agree it is 6 or some number of pawns up, but not almost nothing.
If no human can see the win, and neither can both the engines see it, I strongly object to ending the game because the TB says so.
We who are w atching the games, are humans, and the main purpose is entertainment, so us humans need to follow the games. Either we see it with our own understanding who is winning, or the demonstration boards can show us, plus the evaluations of both engines, before a win is registered.
This USED to be the rule, so has that changed?

Also, whilst I'm speaking here, I'd like to mention too that Game 20, seems to have been a draw. The analysis lines on chessbomb did not see any win. So how come it is registered as a win to Houdini? It's just not right IS it? There is no demonstration that it is a win! Why, was the win demonstrated on main site and agreed by both engines at that point?

If i am wrong about these facts, please let me know. But if not, then I feel this needs serious attention asap.

[JJJ]

Game 20 :
White's connection stalls

The game you mention, Komodo would have win it easily. Yeah it can be one rare case you have to run the end game yourself to see how you win this.

The real disapointment here is the crash of Hannibal. Why the hell this engine is back while still crashing ?

[S.Taylor]

Game 20 :
White's connection stalls

The game you mention, Komodo would have win it easily. Yeah it can be one rare case you have to run the end game yourself to see how you win this.

The real disapointment here is the crash of Hannibal. Why the hell this engine is back while still crashing ?

I'm not at all sure that any program would see any win in game 17, it's only the TB, that i have to believe.
Both queens on with one pawn far back is an easy win??????????? You must be joking!
Perhaps there should be the 50 move rule to see if that pawn moves forward at all.
IF it does, only then might we assume that the losing engine will evaluate it as winning, eventually.

[Peter Berger]

Game 17, Komodo-Laser
I was watching every secoind of it at the end, and the engines did NOT agree it was a win for White (Komodo).
Someone in the the comments section in chessbomb said it was a tablebase win in 53 more moves.
And indeed, Komodo seemed to give that TB verdict by evaluating it as about 280 pawns up. However Lasers evaluation did not even reach 3, as in 3 pawns up.
I thought that both machines had to to agree it is 6 or some number of pawns up, but not almost nothing.
If no human can see the win, and neither can both the engines see it, I strongly object to ending the game because the TB says so.
We who are w atching the games, are humans, and the main purpose is entertainment, so us humans need to follow the games. Either we see it with our own understanding who is winning, or the demonstration boards can show us, plus the evaluations of both engines, before a win is registered.
This USED to be the rule, so has that changed?

I support this concern. It will certainly not make any change in tournament points, yet I feel it is still important,

Laser had no idea at all it was dead lost when the game was adjudicated by the referee. Then it shouldn't be defaulted.

Btw - a very interesting game. Komodo was damn lucky to win it.

Peter

[yurikvelo]

Tablebase positions are automatically adjudicated, no matter if engines can seal the win/draw or not.

Because CLI-Cutechess support only Gaviota, 6-men adjudication is not available.

If I could vote, I will vote for TB7-Lomo adjudication to be implemented in CLI-Cutechess-TCEC.

[Peter Berger]

Tablebase positions are automatically adjudicated, no matter if engines can seal the win/draw or not.

Because CLI-Cutechess support only Gaviota, 6-men adjudication is not available.

If I could vote, I will vote for TB7-Lomo adjudication to be implemented in CLI-Cutechess-TCEC.

Why? This makes no sense at all IMHO.

Let's assume we have TB8-Lomos at some point in time. This way it is easy to imagine ( or actually create btw) a scenario where NEITHER engine has any clue what is going on and the game result becomes absolutely random.

I think one of the virtues of TCEC is that we get things in a nicely automated way. So some 200 useless moves more or less will never really matter because no poor human sits there wasting his times inputting moves ( which would be like the only benefit I could imagine to avoid this scenario ) .

Peter

[yurikvelo]

Let's assume we have TB8-Lomos at some point in time.

We had TB7 at a known point in a past.

Why? This makes no sense at all IMHO.

Implementing EG heuristic makes no sense, since perfect 6-men knowledge is freely available for more than decade (Nalimov) and technically available for everybody for 3 years (since Syzygy-6 availability) and 7-men available for adjudicating (no probing) for 1 year (free android app or paid HTTP access).

Forcing engine authors to spend their precious time implementing bullshit EG-heuristics (otherwise his engine cannot be tested and proven strong) is a way to nowhere.

A lot of authors abandon their projects, because its very time consuming process.
Why force them to reinvent wheel, if 7-men is freely available for all mankind?

Why force kibitzers to watch bullshit EG-heuristics if perfect answer is known immediately?
Why make tournament incompatible (in ELO strength) with respectable rating lists (like CCRL) which all use TB adjudication wherever possible

[cdani]

I vote for 4 piece adjudication, and probably with exceptions if is not easy to achieve the supposed result on some positions, if any. Adjudication with more pieces is to assume the engine will be capable of doing something that no one knows.

[yurikvelo]

Adjudication with more pieces is to assume the engine will be capable of doing something that no one knows.

Also it assume redefinition of what is win or draw in chess.
Correspondence chess FIDE rules allow to claim a win/draw if position is tablebase win/draw.

Engine tournaments and rating lists should adopt correspondence rules, not OTB rules, because computers had been storing perfect knowledge for decade and there is no benefits for mankind to handicap them.

[cdani]

Adjudication with more pieces is to assume the engine will be capable of doing something that no one knows.

Also it assume redefinition of what is win or draw in chess.
Correspondence chess FIDE rules allow to claim a win if position is tablebase win.

Engine tournaments and rating lists should adopt correspondence rules, not OTB rules, because computers had been storing perfect knowledge for decade and there is no benefits for mankind to handicap them.

Depends on the objective. At Tcec I have the opinion that people wants to see the engines battling for the result.