Excellent job

[Lyudmil Tsvetkov]

I would like to express my big satisfaction with the work done by people responsible for the book creation of current TCEC.

This is really a wonderful book, especially compared to the books used in previous editions. I noticed only 1 or 2 dubious positions of all available. Most importantly, all 4 engines right after book exhibited scores in the range of 10-15-20cps white advantage, precisely the way it should be. I saw only a couple of positions where one of the engines showed more than 30cps white edge, and none where an engine would evaluate first move right after book with more than 40cps!

My conclusion: all of the positions were fully equal, a necessary precondition for a just competition.

It could also be noted that the book has a slight tilt towards symmetric or even more closed positions, where engines not always understand fully the best way to proceed.

Some statistics as of round 66:

I counted 26 wins in 66 games, making a total of 40% average win rate. I think all people who thought that more equal positions would produce dull and drawn games, are herewith proven wrong. 40% win rate is a huge number for a tournament at such a long TC, and comprising the 4 strongest engines of more or less equal strength. So that equal positions do not produce more draws.

Out of those 26 wins, 21 went to white, and only 5 to black, so some 85% white wins, confirming the rising white advantage with increasing level of competitor strength.

I also calculated the average game length. Engines played 5001 moves in 66 games, totalling an average of 75.8 moves per game. This confirms the hypothesis that with increasing competitor strength the average game length also increases. So that higher quality also means longer games.

What is your opinion of the current edition TCEC book?

How do you regard the above statistics?

As by now it is almost clear who will qualify for the Superfinal, who do you think will win it?

[yurikvelo]

My conclusion: all of the positions were fully equal, a necessary precondition for a just competition.

For Stages, where engine play only 1 game with this opening - this is must precondition.

For Stage 4 - more unbalanced positions could be allowed, to lower draw rate.

It is still questiioable if it is fair contest for Swiss round robin...

Imagine 2 starting positions: one is so unbalanced that any Stage 4 participant win any opponent. With 2 games each get 1 point.
Another position is very balanced, with 60% draw probability, but there is a chance for one side to win and show its superiority - get 1.5 vs 0.5 points of 2 games.

[Lyudmil Tsvetkov]

My conclusion: all of the positions were fully equal, a necessary precondition for a just competition.

For Stages, where engine play only 1 game with this opening - this is must precondition.

For Stage 4 - more unbalanced positions could be allowed, to lower draw rate.

It is still questiioable if it is fair contest for Swiss round robin...

Imagine 2 starting positions: one is so unbalanced that any Stage 4 participant win any opponent. With 2 games each get 1 point.
Another position is very balanced, with 60% draw probability, but there is a chance for one side to win and show its superiority - get 1.5 vs 0.5 points of 2 games.

As said, the status of balancedness/unbalancedness does not influence the win/draw ratio.

However, it is fully unappealing, both chesswise and ethically, to win a game that is already won. That is why you need balanced positions, at least at this high quality of chess.

[yanquis1972]

i try to give him a hard time whenever i can in TCEC chat, but cato/nelson is probably the authority on comp-comp openings. i know that is/was his niche in his centaur team which i believe was the best around, & i dont believe anyones disputed his claim to the most exhaustive database in the world. so i had full faith in these last year when it was announced. huge thanks to nelson for volunteering ofc but also to martin for going out & getting the best guy for the job. TCEC is really something special.

[lkaufman]

I would like to express my big satisfaction with the work done by people responsible for the book creation of current TCEC.

This is really a wonderful book, especially compared to the books used in previous editions. I noticed only 1 or 2 dubious positions of all available. Most importantly, all 4 engines right after book exhibited scores in the range of 10-15-20cps white advantage, precisely the way it should be. I saw only a couple of positions where one of the engines showed more than 30cps white edge, and none where an engine would evaluate first move right after book with more than 40cps!

My conclusion: all of the positions were fully equal, a necessary precondition for a just competition.

It could also be noted that the book has a slight tilt towards symmetric or even more closed positions, where engines not always understand fully the best way to proceed.

Some statistics as of round 66:

I counted 26 wins in 66 games, making a total of 40% average win rate. I think all people who thought that more equal positions would produce dull and drawn games, are herewith proven wrong. 40% win rate is a huge number for a tournament at such a long TC, and comprising the 4 strongest engines of more or less equal strength. So that equal positions do not produce more draws.

Out of those 26 wins, 21 went to white, and only 5 to black, so some 85% white wins, confirming the rising white advantage with increasing level of competitor strength.

I also calculated the average game length. Engines played 5001 moves in 66 games, totalling an average of 75.8 moves per game. This confirms the hypothesis that with increasing competitor strength the average game length also increases. So that higher quality also means longer games.

What is your opinion of the current edition TCEC book?

How do you regard the above statistics?

As by now it is almost clear who will qualify for the Superfinal, who do you think will win it?

I agree with your comments except for your description of the openings as "fully equal". As shown by the 21 to 5 White stats, they are not at all equal, and not intended to be equal, except in the theoretical sense that the White advantages are probably not enough to win with perfect play. If "fully equal" openings were chosen (let's say ones within 5 centipawns of a zero score) there would be many more draws I'm sure. I think they picked openings with just enough White advantage to give winning chances but not actual wins, which made the 40% decisive results possible. This is one reason that rating differences in comp vs comp games (which tend to use openings where White has a normal advantage) tend to be larger than in human vs. human games, where Black (at top level) usually aims for slightly worse endings that are easy to hold.
As for the final, Komodo has added a few more elo since the semifinal version, hopefully enough to offset any SF gains in the same period. It should be close, but based on the semifinal results Komodo must be at least a slight favorite.

[Lyudmil Tsvetkov]

I would like to express my big satisfaction with the work done by people responsible for the book creation of current TCEC.

This is really a wonderful book, especially compared to the books used in previous editions. I noticed only 1 or 2 dubious positions of all available. Most importantly, all 4 engines right after book exhibited scores in the range of 10-15-20cps white advantage, precisely the way it should be. I saw only a couple of positions where one of the engines showed more than 30cps white edge, and none where an engine would evaluate first move right after book with more than 40cps!

My conclusion: all of the positions were fully equal, a necessary precondition for a just competition.

It could also be noted that the book has a slight tilt towards symmetric or even more closed positions, where engines not always understand fully the best way to proceed.

Some statistics as of round 66:

I counted 26 wins in 66 games, making a total of 40% average win rate. I think all people who thought that more equal positions would produce dull and drawn games, are herewith proven wrong. 40% win rate is a huge number for a tournament at such a long TC, and comprising the 4 strongest engines of more or less equal strength. So that equal positions do not produce more draws.

Out of those 26 wins, 21 went to white, and only 5 to black, so some 85% white wins, confirming the rising white advantage with increasing level of competitor strength.

I also calculated the average game length. Engines played 5001 moves in 66 games, totalling an average of 75.8 moves per game. This confirms the hypothesis that with increasing competitor strength the average game length also increases. So that higher quality also means longer games.

What is your opinion of the current edition TCEC book?

How do you regard the above statistics?

As by now it is almost clear who will qualify for the Superfinal, who do you think will win it?

I agree with your comments except for your description of the openings as "fully equal". As shown by the 21 to 5 White stats, they are not at all equal, and not intended to be equal, except in the theoretical sense that the White advantages are probably not enough to win with perfect play. If "fully equal" openings were chosen (let's say ones within 5 centipawns of a zero score) there would be many more draws I'm sure. I think they picked openings with just enough White advantage to give winning chances but not actual wins, which made the 40% decisive results possible. This is one reason that rating differences in comp vs comp games (which tend to use openings where White has a normal advantage) tend to be larger than in human vs. human games, where Black (at top level) usually aims for slightly worse endings that are easy to hold.
As for the final, Komodo has added a few more elo since the semifinal version, hopefully enough to offset any SF gains in the same period. It should be close, but based on the semifinal results Komodo must be at least a slight favorite.

By equal I meant not offering win possibilities for either side with perfect play. A non-native speaker sometimes has hard time using precise wording, especially in pre-Christmas times.

I think book positions showing 10-20cps advantage for white are the right way to do it, as this reflects the natural white right to move advantage. 0.0 score positions would rather be kind of a distortion if constituting a majority, but some 0.0 positions are just OK.

Come on, Larry, you are the clear favourite! And you know why? Precisely because of the openings. My impression is that they somewhat tend towards symmetrical/closed play, and you know very well that Komodo plays such positions better than SF. You have reiterated this several times, and here I fully agree with you. My only remark is that Komodo plays such positions better than SF, but far from perfect.

The good thing is that SF fans still do not have a clear idea about the above, otherwise I do not imagine the outcry that could follow.

Btw., good luck to both teams in the final.

[lkaufman]

I don't think that the choice of openings is the main reason Komodo did better in the semifinal. It's pretty clear to me what the reason is, but I can't divulge it for obvious competitive reasons. I'll just say that while latest Stockfish is still a bit stronger than latest Komodo in single-core bullet testing, this has little in common with TCEC conditions or even with typical home use.

[Laskos]

Come on, Larry, you are the clear favourite! .

I would call slight favorite. Last time it was a clear win of SF. The excellent starting positions were by Nelson too that time. Let's see, but I wouldn't say "clear" outcome in the Superfinal as of now.

[Lyudmil Tsvetkov]

Come on, Larry, you are the clear favourite! .

I would call slight favorite. Last time it was a clear win of SF. The excellent starting positions were by Nelson too that time. Let's see, but I wouldn't say "clear" outcome in the Superfinal as of now.

So what are your probabilities about the final?

I am sure we would all like to hear about it.

[Milos]

I don't think that the choice of openings is the main reason Komodo did better in the semifinal. It's pretty clear to me what the reason is, but I can't divulge it for obvious competitive reasons. I'll just say that while latest Stockfish is still a bit stronger than latest Komodo in single-core bullet testing, this has little in common with TCEC conditions or even with typical home use.

I thought it was quite obvious. You tuned eval only for the pool of opponents of SF, H and G (only SF for the final) and on the set of openings which have been chosen to be as similar as possible to type of opponings Cato is choosing.
TCEC is the best marketing thing for your engine and you can use it to your advantage.
SF with its test methodology is clearly in disadvantage.