Question about values of books in chess today

[Uri Blass]

Suppose that there is a tournament of 100 games in TCEC time control and hardware.

The opponent is stockfish18 with no book from move 1.
You also use stockfish18 and you are allowed to use a special book for the match.

Can you build a book in 3 months that you expect to get more than 50% in the match.
If not then what is the latest version of stockfish that you can build a book to get more than 50%?
What is the larest version of stockfish when you do not need to build a special book and there is already a book that does it.

In all cases the same engine play against itself.

[Peter Berger]

100 games is a very low number, so it might be difficult and a bit random to succeed.

You can probably kill all the black games to a draw in quite (or maybe even very) little time (definitely less than a month).

The real question is: can you reliably win at least one game with white? It is a challenge, and it will be work. You either have to find an extremely imbalanced, slightly favourable position you can get Stockfish to play and let the variance do the job for you, or you have to find a real hole. Maybe in the Catalan, here Stockfish tends to choose setups with black that feel like they could be challenged maybe.

I envision someone really working on this challenge for 3 months. You will only find out if you donate enough money for someone to give it a serious enough try. My bet is like 60-40 for this to still be feasible. All guesswork and IMHO of course.

[Aleks Peshkov]

Do not agree that SF = all chess. The question can be more interesting if will not use SF or SF-clone engine with a custom book against bare SF. There are several very strong candidates.

[Dann Corbit]

There are two ways to introduce quality into books. First, analyze the most frequently used nodes at a time control deeper than expected in the games to be run. Second, analyze the statistics for a collection of high quality games and incorporate this data.
Now, using this book, try matches against the same engine but with no book. Whenever there is a loss, analyze the game for the place where the book engine went wrong.

I think that Sedat's matches demonstrate the value of books. He sometimes puts in "no book" entries and they get trounced.

The main problem with books is that they get used in contests and then the resultant games get analyzed and the winning book moves are discovered. Therefore, the opponents can plan against them in their books.

It would also be possible to specialize against certain engines by looking for positions where they lost the game. Put these positions into a database and try to steer the engine into those positions.

A big win that nobody thinks about is time. Clearly time bonus is especially valuable in games with a time increment added per move. If I can make 20 book moves and fall out of the book even, I have a huge time advantage at this point compared to an engine with no book.

You remember Uri, how those Orangutan lines had a very positive outcome in the books I made for you? It had nothing to do with tactical supremacy, since all of those book lines fell out of book even, unless the opponent made a blunder. But the thing gained was time. If I have enough time to think twice as long, I have an Elo win. Nobody analyzes the Orangutan for their opening book, or other kooky openings. I have ten thousand analyzed Orangutan book nodes. As long as the book exit is even score, the engine that stayed in book longest is the benefactor, because of gained time.

[Peter Berger]

Uri's question ist way more edgy, Dann. please reread.

[Peter Berger]

I have thought about the question how we could get closer to an answer to your question with a challenge that could actually be tested in reality because it takes less time and doesn't need amazing hardware.

1.) Again you get three months to prepare an opening book (or some other decent period of time).
2.) This time your goal is to draw all games (or better just not to lose the match). Let's do 1000 games with a somewhat decent time control, 10 minutes/1sec increment game?
3.) First you create your opening book, your opponent again will be Stockfish 18 without book on some hardware setup both opposing engines will use.
4.) Now you choose your personal companion engine for the match in public - if you lose the match against Stockfish 18, you are out. If you don't lose the match, the human winner of this book challenge will be the one who chose the engine that has the lowest ELO on some esteemed rating list(s).

How low would you dare to go? I think we both agree that you can choose +some+ entry and succeed in this challenge (I guess you can do this with Torch e.g.). But can you go say 200 ELO points lower? Or even 300?

To not get too many useless entries that will endanger the world climate you can even have an entry fee, that would help create a somewhat decent prize money to make some serious people really work on this challege.

Best regards.
Peter

[Uri Blass]

There are two ways to introduce quality into books. First, analyze the most frequently used nodes at a time control deeper than expected in the games to be run. Second, analyze the statistics for a collection of high quality games and incorporate this data.
Now, using this book, try matches against the same engine but with no book. Whenever there is a loss, analyze the game for the place where the book engine went wrong.

I think that Sedat's matches demonstrate the value of books. He sometimes puts in "no book" entries and they get trounced.

The main problem with books is that they get used in contests and then the resultant games get analyzed and the winning book moves are discovered. Therefore, the opponents can plan against them in their books.

It would also be possible to specialize against certain engines by looking for positions where they lost the game. Put these positions into a database and try to steer the engine into those positions.

A big win that nobody thinks about is time. Clearly time bonus is especially valuable in games with a time increment added per move. If I can make 20 book moves and fall out of the book even, I have a huge time advantage at this point compared to an engine with no book.

You remember Uri, how those Orangutan lines had a very positive outcome in the books I made for you? It had nothing to do with tactical supremacy, since all of those book lines fell out of book even, unless the opponent made a blunder. But the thing gained was time. If I have enough time to think twice as long, I have an Elo win. Nobody analyzes the Orangutan for their opening book, or other kooky openings. I have ten thousand analyzed Orangutan book nodes. As long as the book exit is even score, the engine that stayed in book longest is the benefactor, because of gained time.

At long time control time does not help and I expect stockfish with no book (100 seconds per move) to get 100% draws against stockfish (1 second per move).

[Uri Blass]

I have thought about the question how we could get closer to an answer to your question with a challenge that could actually be tested in reality because it takes less time and doesn't need amazing hardware.

1.) Again you get three months to prepare an opening book (or some other decent period of time).
2.) This time your goal is to draw all games (or better just not to lose the match). Let's do 1000 games with a somewhat decent time control, 10 minutes/1sec increment game?
3.) First you create your opening book, your opponent again will be Stockfish 18 without book on some hardware setup both opposing engines will use.
4.) Now you choose your personal companion engine for the match in public - if you lose the match against Stockfish 18, you are out. If you don't lose the match, the human winner of this book challenge will be the one who chose the engine that has the lowest ELO on some esteemed rating list(s).

How low would you dare to go? I think we both agree that you can choose +some+ entry and succeed in this challenge (I guess you can do this with Torch e.g.). But can you go say 200 ELO points lower? Or even 300?

To not get too many useless entries that will endanger the world climate you can even have an entry fee, that would help create a somewhat decent prize money to make some serious people really work on this challege.

Best regards.
Peter

A possible way to go may be to give stockfish to play many games at this time control against itself and add the draws that stockfish play against itself to your book.

I wonder how many possible lines of x plies you are going to get for every x.
Stockfish may not be deterministic with many cores so games may be different.

Note that you can practically decide for buliding a book for black that black is a deterministic player in the book so after every draw you add the moves of black to the book and again the question is how many different games you get for 2x plies for exery x(deviation cannot be at ply 2x-1 because after the first game stockfish learned to repeat the same move).

[Uri Blass]

Another question is what is the maximal level of stockfish fixed nodes that engines can beat.

Stockfish18 1M nodes per move sungle core already usually draw based on my tests(tested against Reckless0.10 with 10M nodes per,11M nodes per move 12M nodes per move and 13M nodes per move and all results are draws).

[Uri Blass]

stockfish18 1M lost a game against ReckLess depth 22
Here is the pgn(reckless wanted to play Rd2 at depth 21 but had a problem to finish depth 22 and decided to adjudicate the game as a win for black after reckless refused to move for a full hour.

[pgn][Event "Computer event"]
[Site "Somewhere on Earth"]
[Date "2026.04.05"]
[Time "22:55:40"]
[Round "3"]
[Board "3"]
[White "Stockfish 18"]
[Black "Reckless 0.10.0-dup1"]
[Result "0-1"]
[Termination "adjudication by human"]
[ECO "C11"]
[Opening "French"]
[TimeControl "120+2"]
[PlyCount "113"]

1.e4 {+0.3/21 1919 1000423} e6 {+0.5/22 3279 2125214} 2.d4 {+0.5/22 1669 1000309} d5 {+0.5/22 2170 1477519}
3.Nc3 {+0.4/23 1506 1000289} Nf6 {+0.4/22 1608 1105335} 4.e5 {+0.5/23 1515 1000455} Nfd7 {+0.5/22 1574 983715}
5.f4 {+0.4/22 1400 1000278} c5 {+0.5/22 2098 1529638} 6.Nf3 {+0.4/24 1420 1000182} Nc6 {+0.5/22 4030 2685854; C11: French, Steinitz variation}
7.Be3 {+0.4/23 1595 1000382} Be7 {+0.4/22 4633 2839678; C11: French, Steinitz, Boleslavsky variation} 8.Qd2 {+0.5/21 1733 1000004} a6 {+0.5/22 3835 2415355}
9.Bd3 {+0.6/20 1713 1000399} cxd4 {+0.6/22 3421 2315693} 10.Nxd4 {+0.6/21 1602 1000109} Nc5 {+0.6/22 1858 1300960}
11.O-O-O {+0.7/22 1765 1000533} Qc7 {+0.6/22 8383 5877379} 12.Qf2 {+0.6/23 1616 1000560} b5 {+0.6/22 6297 4351592}
13.Qg3 {+0.6/22 1612 1000035} O-O {+0.1/22 4576 3021919} 14.Kb1 {+0.3/20 1460 1000261} b4 {+0.1/22 5307 3740449}
15.Nce2 {+0.2/18 1368 1000020} Kh8 {+0.3/22 9427 6739949} 16.Rhf1 {+0.3/18 1532 1000459} Nxd4 {-0.1/22 6381 4695139}
17.Nxd4 {+0.1/18 1544 1000292} a5 {+0.0/22 749 511555} 18.Rf3 {+0.0/19 1455 1000460} Nxd3 {-0.2/22 4159 3017766}
19.cxd3 {-0.4/19 1612 1000180} Bd7 {-0.3/22 3614 2552495} 20.Rc1 {-0.3/21 1954 1000469} Qd8 {-0.3/22 6054 4232358}
21.f5 {-0.4/21 1624 1000387} exf5 {-0.3/22 2617 2220373} 22.Nxf5 {-0.4/21 1337 1000393} Bxf5 {-0.4/22 3481 2951144}
23.Rxf5 {-0.4/21 1386 1000491} Qd7 {-0.5/22 1824 1626114} 24.Qh3 {-0.5/18 1725 1000162} Qe6 {-0.3/22 1270 960775}
25.Rf3 {-0.4/20 1483 1000546} Qxe5 {-0.5/22 5816 4403813} 26.Qf5 {-0.5/20 1861 1000169} Qd6 {-0.6/22 3555 2286408}
27.Rc5 {-0.3/19 1545 1000122} Rad8 {-0.6/22 2038 1717047} 28.Rxa5 {-0.4/19 1423 1000061} Rfe8 {-0.7/22 3774 3080384}
29.Bd4 {-0.5/16 1682 1000301} Qc6 {-1.1/22 1287 932146} 30.Rf1 {-0.7/18 1416 1000667} Rc8 {-1.1/22 3504 2628468}
31.Qf2 {-0.6/19 1490 1000529} h6 {-0.7/22 5341 4274276} 32.Ra7 {-0.5/19 1485 1000033} Qg6 {-0.5/22 3345 2700147}
33.Qf3 {-0.5/19 1312 1000432} Qc6 {-0.5/22 3203 2668100} 34.Qf2 {-0.5/19 1395 1000542} Qg6 {-0.7/22 4117 3410082}
35.Qf3 {-0.4/20 1321 1000361} Kh7 {-0.8/22 1159 1022505} 36.Rd1 {-0.4/19 1383 1000139} Bg5 {-0.6/22 4460 3488851}
37.h4 {-0.5/19 1449 1000242} Bxh4 {-1.4/22 2893 2271082} 38.a4 {-1.2/21 1505 1000393} b3 {-1.4/22 1485 1280107}
39.Bc3 {-1.4/21 1426 1000001} d4 {-1.6/22 2246 1970488} 40.Bxd4 {-1.3/22 1308 1000485} Rcd8 {-1.7/22 2529 2267911}
41.Bc3 {-1.3/22 1411 1000175} Re7 {-1.5/22 1680 1460657} 42.Rxe7 {-1.5/23 1210 1000216} Bxe7 {-1.7/22 3008 2968182}
43.a5 {-1.6/22 1414 1000217} Rd7 {-1.8/22 3424 3338274} 44.g4 {-1.6/25 1358 1000379} Bd8 {-2.0/22 2216 2266903}
45.Qf5 {-1.4/22 1221 1000010} Qxf5 {-2.6/22 3458 3538248} 46.gxf5 {-1.8/21 1126 1000482} h5 {-2.4/22 2083 2239162}
47.Kc1 {-1.8/20 1191 1000236} Bg5+ {-3.0/22 2997 3222465} 48.Kb1 {-2.0/20 1081 1000173} h4 {-3.1/22 2214 2184432}
49.Bb4 {-2.0/19 1301 1000243} Be3 {-5.4/22 3837 3906528} 50.a6 {-3.3/22 1166 1000565} Kh6 {-6.3/22 4008 4202534}
51.Rh1 {-3.7/21 1174 1000069} Rxd3 {-7.0/22 3675 3681399} 52.Bc3 {-4.7/19 1168 1000424} Kg5 {-7.5/22 4329 4809698}
53.Be5 {-5.2/20 1022 1000105} Kxf5 {-7.5/22 2851 3116519} 54.Bb8 {-5.4/20 1030 1000566} Rd7 {-7.6/22 1863 2332428}
55.a7 {-5.4/15 988 1000383} Bxa7 {-7.8/22 2534 2905140} 56.Bxa7 {-6.0/15 793 1000203} Rxa7 {-8.1/22 5507 7251300}
57.Rxh4 {-6.1/17 726 1000475} 0-1

[/pgn]