Winboard Forum

Hi to all!
I have made several opening books :
The main advice Ihave got is the following:
" positions must be balanced"
But what is a balanced position for an opening book?
Probably not exactly the same as in a game analysis : so I should be glad if you can propose me some criteria :
For all those I can imagine , I find at once some counter-examples
Yet I must take this in account when I make an opening book

A possible definition might be something like this:

a position is statistically balanced, relative to a database of games, when
there is at least "n" games with this position in the database, and White has scored between 50% and 55% in those games.
(for example, n=100 looks sensible).

The database has to obbey some criteria. For example, we might rule that players must be rated above 2500 and there is a specified minimum of games in it.

An alternative -and very convenient definition in my view- is that the database should have only games where the position has arisen, and the players are top chess engines. "Top" is to be sure controversial and temporary. I'd prefer to have a stable list of engines who more or less everywhere agrees are top
(for instance: Shredder, The King, Hiarcs, Junior, Fritz, Nimzo, Gandalf, Ruffian, ProDeo, Crafty and the like).

The list could be actualized yearly, or at least yearly.

This way many people would be able to verify whether a due position is "balanced" or not.

In the absence of enough data, a position might be regarded as -likely-
balanced if someone who presumably knows about -a grandmaster, say- believes it is balanced.

It goes without saying that, for a fairly big opening book, you couldn't possibly be sure that all its positions are in fact balanced. You have to hand tuned it again and again.

Telmo

Hi Telmo!
I thank you for your answer , that is perfect for introduce the discussion
Of course , there is no problem with variations that lead to flat positions , but they are not the most interesting
With critical lines , various criteria are in conflict , and that lead to interesting problems: some examples:
2 Knights Defence 1 e4 e5 2 Nf3 Nc6 3 Bc4 Nf6 4 Ng5 d5 5 exd5
Theory assessment is that chances are equal ( maybe slightly better for Black) after 5...Na5 , 5...Nd4 , or 5...b5
But most engines chose 5...Nxd5 , so between engines the% of white wins is abnormmally high
Leningrad Nimzo 1 d4 Nf6 2 c4 e6 3 Nc3 Bb4 4 Bg5 h6 5 Bh4 c5 6 d5
There are a lot of "theoretical" moves , but most engines chose
6...Nxd5?! 7 Bxd8 Nxc3 8 Qb3 Ne4+ 9 Kd1 Nxf2+ 10 Kc1 (or Kc2) followed by Nxh1 ; unfortunately , after 10...Kxd8 11 Qg3 Nxh1 12 Qxg7 Re8 13 g4 , white wins (this refutation is quite different of the lines in ECO)
Latvian Gambit : this is the contrary of previous examples :
the line is given as winning for white , but with best play on both sides , there is a forced draw, but there are a lot of difficult moves to find for both sides
the analysis is rather long , and I am 78 , so I shall give you in my next post

Claude Le Page wrote:Hi Telmo!
I thank you for your answer , that is perfect for introduce the discussion
Of course , there is no problem with variations that lead to flat positions , but they are not the most interesting
With critical lines , various criteria are in conflict , and that lead to interesting problems: some examples:
2 Knights Defence 1 e4 e5 2 Nf3 Nc6 3 Bc4 Nf6 4 Ng5 d5 5 exd5
Theory assessment is that chances are equal ( maybe slightly better for Black) after 5...Na5 , 5...Nd4 , or 5...b5
But most engines chose 5...Nxd5 , so between engines the% of white wins is abnormmally high
Leningrad Nimzo 1 d4 Nf6 2 c4 e6 3 Nc3 Bb4 4 Bg5 h6 5 Bh4 c5 6 d5
There are a lot of "theoretical" moves , but most engines chose
6...Nxd5?! 7 Bxd8 Nxc3 8 Qb3 Ne4+ 9 Kd1 Nxf2+ 10 Kc1 (or Kc2) followed by Nxh1 ; unfortunately , after 10...Kxd8 11 Qg3 Nxh1 12 Qxg7 Re8 13 g4 , white wins (this refutation is quite different of the lines in ECO)
Latvian Gambit : this is the contrary of previous examples :
the line is given as winning for white , but with best play on both sides , there is a forced draw, but there are a lot of difficult moves to find for both sides
the analysis is rather long , and I am 78 , so I shall give you in my next post

Very sensible observation! I'm starting to think that the advice favoring "balanced" positions could be a mistake.
From your first example -say- one could deduce that the position after 1.e4 e5 2.Nf3 Nc6 3.Bc4 Nf6 4.Ng5 d5 5.exd5 can't be final, as it is not -statistically- "balanced" because most engines would err with 5... Nxd5.
So one must include more plies in the book. Are you making books in Arena format? In this case you should specify a very low probability for the move 5...Nxd5.
Yet, to be sure, any engine playing this opening as White has to know how to profit from 5...Nxd5?!, so you add a refutation for White. But, of course, the final position of the line that refutes 5...Nxd5 will be pretty far from balanced"

Telmo

Hi Telmo!
As you know , I am old , and I was somewhat tired , so I had to shorten my post
It's useful to complete it by my last example , beginning with the famous game Atars- Tomson:
1 e4 e5 2 Nf3 f5 3 Bc4 fxe4 4 Nxe5 Qg5 5 d4 Qxg2 6 Qh5+ g6 7 Bf7+ Kd8
8 Bxg6 Qxg1+ 9 Ke2 c6 10 Nc3 (*) Kc7 11 Bf4 Qxa1 (**)12 Nxd7+!! Kxd7
(***) 13 Qf5+ Kd8 14 Qxf8+ Kd7 15 Bf5#
(*)any other move loses!
(**)See infra
(***)if 12...Kd8 , it's mate in 18 , announced both by Junior7 and Shredder8 (but the mates are different )
Statistics are ugly : I have 18 examples in Harding's database , all 1-0
And yet , Black can improve by 11...hxg6 12 Qg5 Be7 13 Nxc6+! d6!!
14 Bxd6 Kxd6 15 Qc5+ Ke6 16 Qe5+ Kf7 17 Rxh1 Nxc6 18 Qxh8 Nf6 19 Ke3 Bf8 20 Nxe4 Nxe4 21 Kxe4 Bg7 22 Qh4 Bf5+ 23 Kf3 Nxd4+ 24 Kg2 g5
25 Qh5+ Bg6 26 Qg4 Bf5 27 Qh5+ Bg6 Draw
Thus this is OBJECTIVELY balanced , even if it's difficult for both sides
Everything is in the Purpose :
If my purpose is to judge the progress of some Amateur programs , I can remove questionable variations , even if it makes the tournament less interesting , but I can't be so indulgent with Fritz or Hiarcs
Anyway , the choice of best line in Fegatello is far from definitive , and I am interested by the way used by Shredder , Junior or Gandalf to refute Black play
So , I must not prevent an engine to play Nxd5 :if it does so , it's a bug
More generally , when I have the choice of length of book line , I stop just before the point where engine has an option
Now , I see that I failed to discuss one or 2 points:
About statistics , it is sufficient that a master adopt a rare line with success for completely change the statistics :
Cambridge Gambit:
1 e4 Nf6 2 e5 Nd5 3 d4 d6 4 c4 Nb6 5 f4 g5!?
Statistics depend on the number of games of Michael Schirmer that are in the database : he scores 74% with Black , other players only 28%
At last , I am quite sceptical about the consensus between Masters :
Their assessment are the reflet of fashion , and not the expression of the Truth

Claude Le Page wrote:Hi Telmo!
As you know , I am old , and I was somewhat tired , so I had to shorten my post
It's useful to complete it by my last example , beginning with the famous game Atars- Tomson:
1 e4 e5 2 Nf3 f5 3 Bc4 fxe4 4 Nxe5 Qg5 5 d4 Qxg2 6 Qh5+ g6 7 Bf7+ Kd8
8 Bxg6 Qxg1+ 9 Ke2 c6 10 Nc3 (*) Kc7 11 Bf4 Qxa1 (**)12 Nxd7+!! Kxd7
(***) 13 Qf5+ Kd8 14 Qxf8+ Kd7 15 Bf5#
(*)any other move loses!
(**)See infra
(***)if 12...Kd8 , it's mate in 18 , announced both by Junior7 and Shredder8 (but the mates are different )
Statistics are ugly : I have 18 examples in Harding's database , all 1-0
And yet , Black can improve by 11...hxg6 12 Qg5 Be7 13 Nxc6+! d6!!
14 Bxd6 Kxd6 15 Qc5+ Ke6 16 Qe5+ Kf7 17 Rxh1 Nxc6 18 Qxh8 Nf6 19 Ke3 Bf8 20 Nxe4 Nxe4 21 Kxe4 Bg7 22 Qh4 Bf5+ 23 Kf3 Nxd4+ 24 Kg2 g5
25 Qh5+ Bg6 26 Qg4 Bf5 27 Qh5+ Bg6 Draw
Thus this is OBJECTIVELY balanced , even if it's difficult for both sides

Interesting. I'd like to know if the natural move 13.Nf7+ (instead of 13.Nxc6+) has been studied as well.
This is "our" analysis (Fritz in Bahrain & me):

13.Nf7+ Kb6 14.Na4+ Ka6 15.Nc5+ Kb6 16.Rxh1 Bxg5 17.Bxg5

and Black looks in trouble.

Claude Le Page wrote:Everything is in the Purpose :
If my purpose is to judge the progress of some Amateur programs , I can remove questionable variations , even if it makes the tournament less interesting , but I can't be so indulgent with Fritz or Hiarcs
Anyway , the choice of best line in Fegatello is far from definitive , and I am interested by the way used by Shredder , Junior or Gandalf to refute Black play
So , I must not prevent an engine to play Nxd5 :if it does so , it's a bug
More generally , when I have the choice of length of book line , I stop just before the point where engine has an option

Now I think I can understand your point of view, namely, the purpose of having opening books are:

i) to ensure variability of play;
ii) to ensure that interesting variations (like the Fegatello) have an opportunity to be tested, even if they are regarded as too risky.

Claude Le Page wrote:Now , I see that I failed to discuss one or 2 points:
About statistics , it is sufficient that a master adopt a rare line with success for completely change the statistics :
Cambridge Gambit:
1 e4 Nf6 2 e5 Nd5 3 d4 d6 4 c4 Nb6 5 f4 g5!?
Statistics depend on the number of games of Michael Schirmer that are in the database : he scores 74% with Black , other players only 28%
At last , I am quite sceptical about the consensus between Masters :
Their assessment are the reflet of fashion , and not the expression of the Truth

I don't know who's that Schirmer. in my view, scoring 74% with Black doesn't imply the Cambridge gambit is a decent opening, if his adversaries have been weak.

Telmo

Hi Telmo!
Games played by Schirmer are postal games , played at highest level ;
Now , if Cambridge Gambit looks questionable , you have the same phenomenon with Boey and Jaenisch Gambit of Ruy Lopez , that becomes the "best" defence in correspondence databases , chiefly with the games won by Boey and his imitators
I analyzed your line of Latvian gambit with Junior7 up to depth 20 : White has an edge , but Black can possibly draw ; 13 Nxc6+ is held for best winning try , and 13...d6 is really a very beautiful move

Claude Le Page wrote:Hi Telmo!
Games played by Schirmer are postal games , played at highest level ;

Alas, how ignorant I have become lately! I'll try to find Schirmer's games, I imagine they might be something special.

Now , if Cambridge Gambit looks questionable , you have the same phenomenon with Boey and Jaenisch Gambit of Ruy Lopez , that becomes the "best" defence in correspondence databases , chiefly with the games won by Boey and his imitators

I have lost some games against the Jaenisch myself, so I agree that it is not too questionable

I analyzed your line of Latvian gambit with Junior7 up to depth 20 : White has an edge , but Black can possibly draw ; 13 Nxc6+ is held for best winning try , and 13...d6 is really a very beautiful move

My understanding was that you had established that 13.Nxc6+ is drawish. If this is so, then 13.Nxc6+ is spectacularly wrong

Best regards

Telmo