Stockfish test position (& possible eval bug)

[carldaman]

Hi Carl.

I did not go deep enough into the position, so I do not know if Qh5 is good or bad, but I think that the logic behind such a seemingly strange move might be as follows: when Qh5 and g6 are played, as pawns do not go back, the engine would consider that, when the black king castles short (a very probable option), the king on g8 would be worse covered by the pawn on g6 than a pawn still on g7, as in this case the king will have a pawn less immediately adjacent, and this is an important pawn. Pawns do not go back

The position is still somewhat open, so g6 might be a good move, but might also be a weakness, I did not go deep enough and am undecided.

I have also been frequently 'baffled' by similar engine moves in the past, they sometime prove to have some logic behind, though.

Hi Lyudmil.

I respect your original and independent thinking on chess, but this move is no valid "novelty", so it can't be easily rationalized. Maybe if 1 tempo, and not 2, had been lost, I might concur that inducing a "weakness" might have a point. But here, g6 is a very useful move for Black, leading to a sort of KID setup, and then Black gets in Nf6 for free as well. Deep analysis aided by the underrated and forgotten Deep Fritz 10 engine shows Black can mount a promising and swift pawn storm. DF10 may be one of the few engines that truly plays well in closed positions with flank attacks and pawn storms on the Kingside.

In my database, and backed up by extensive analysis, Nc2 or even O-O are objectively better moves for White, and the masters who played them probably think likewise.

I can post more later when I have more time.

CL

[Lyudmil Tsvetkov]

Hi Carl.

I did not go deep enough into the position, so I do not know if Qh5 is good or bad, but I think that the logic behind such a seemingly strange move might be as follows: when Qh5 and g6 are played, as pawns do not go back, the engine would consider that, when the black king castles short (a very probable option), the king on g8 would be worse covered by the pawn on g6 than a pawn still on g7, as in this case the king will have a pawn less immediately adjacent, and this is an important pawn. Pawns do not go back

The position is still somewhat open, so g6 might be a good move, but might also be a weakness, I did not go deep enough and am undecided.

I have also been frequently 'baffled' by similar engine moves in the past, they sometime prove to have some logic behind, though.

Hi Lyudmil.

I respect your original and independent thinking on chess, but this move is no valid "novelty", so it can't be easily rationalized. Maybe if 1 tempo, and not 2, had been lost, I might concur that inducing a "weakness" might have a point. But here, g6 is a very useful move for Black, leading to a sort of KID setup, and then Black gets in Nf6 for free as well. Deep analysis aided by the underrated and forgotten Deep Fritz 10 engine shows Black can mount a promising and swift pawn storm. DF10 may be one of the few engines that truly plays well in closed positions with flank attacks and pawn storms on the Kingside.

In my database, and backed up by extensive analysis, Nc2 or even O-O are objectively better moves for White, and the masters who played them probably think likewise.

I can post more later when I have more time.

CL

Hi Carl.
I respect your searching after the truth even more, and, after looking more carefully at the position (my first suggestion was just a hypothetical assumption), my opinion is that 10.Qh5 is indeed a bad move (at least imperfect). I would play 0-0 instead (but those are just intuitive assessments, as I do not follow theory at all).

I think that after 9.Bd3 f4 black is actually better, although the position is not as much closed to empower the chain. I would play 9.ef5 instead (Bf5 Bd3), maybe white is better here.

[Lyudmil Tsvetkov]

Thanks for confirming my findings, Louis. In many of my runs, actually Qh5+ edges out the other (better) choices by a small eval margin. I'll post more on this later.

It edges them because Stockfish possibly does not consider chains properly, but considers fully king shelter weaknesses.

[carldaman]

Thanks for confirming my findings, Louis. In many of my runs, actually Qh5+ edges out the other (better) choices by a small eval margin. I'll post more on this later.

It edges them because Stockfish possibly does not consider chains properly, but considers fully king shelter weaknesses.

I think you are correct. SF, like other top programs underestimates and undervalues such pawn chains, directed at the Kingside (as White typically castles short in this variation). It is better to play 9.exf5, as you suggested in another post, which is in fact the main line theoretically. 9.Bd3 has its following, but I wouldn't rate it well, especially for engines. Among human GMs, Nick DeFirmian, Semyon Dvoirys, Andrei Sokolov, Simen Agdestein and John Nunn have played it.

Incidentally, another engine that falls prey to the Qd1-h5-d1 'temptation' is Hiarcs 14, which surprises me.

Regards,
CL

[Lyudmil Tsvetkov]

Among human GMs, Nick DeFirmian, Semyon Dvoirys, Andrei Sokolov, Simen Agdestein and John Nunn have played it.

You might scrap all of those then
But I do not know about John Nunn, as the Nunn lobby here seems to be very influential...

[carldaman]

Among human GMs, Nick DeFirmian, Semyon Dvoirys, Andrei Sokolov, Simen Agdestein and John Nunn have played it.

You might scrap all of those then
But I do not know about John Nunn, as the Nunn lobby here seems to be very influential...

Hah! -- and you wanna see something funny? I've just noticed that this month's Checkpoint Book Review column at the ChessCafe features Nunn's loss as white to Nataf, in the very same variation, where he played the dubious 10.g3?! (talk about synchronicity!).

http://www.chesscafe.com/hansen/hansen170.htm

[Lyudmil Tsvetkov]

Among human GMs, Nick DeFirmian, Semyon Dvoirys, Andrei Sokolov, Simen Agdestein and John Nunn have played it.

You might scrap all of those then
But I do not know about John Nunn, as the Nunn lobby here seems to be very influential...

Hah! -- and you wanna see something funny? I've just noticed that this month's Checkpoint Book Review column at the ChessCafe features Nunn's loss as white to Nataf, in the very same variation, where he played the dubious 10.g3?! (talk about synchronicity!).

http://www.chesscafe.com/hansen/hansen170.htm

Thanks, Carl, for the link.

Interesting info.

I think many widely acknowledged theoretical lines are at least a bit dubious, and therefore I prefer instead of learning theory to play something like 1.f3, or 1.e4/d4 f6 with black, in the hope that will swiftly beguile engines to inferior lines. I would even suggest those lines for Graham's book, but I am sure he will decline; although, please notice, that the best game in the 3 Champs so far (and I think this is game 6), the more interesting in content and with the most unusual setup, arose out of a 1.b3 line. So people definitely miss a lot of fun when they do not pay enough attention to 'off' choices.

In any case, I am almost certain that both 1.f3, as well as 1.e4/d4 f6, are objectively much better options than at least half of the lines you will find in most books. And they offer the engines better chances to show their real capabilities at that.

But things are like they are: for the time being I will be the only crazy bloke who on 1.e4 will consistently reply with f6 (not entirely true, because I choose f6 just sometimes; perfectly playable, btw., still almost equal, but some sweet talk will justify an exaggeration)

[carldaman]

Thanks, Carl, for the link.

Interesting info.

I think many widely acknowledged theoretical lines are at least a bit dubious, and therefore I prefer instead of learning theory to play something like 1.f3, or 1.e4/d4 f6 with black, in the hope that will swiftly beguile engines to inferior lines. I would even suggest those lines for Graham's book, but I am sure he will decline; although, please notice, that the best game in the 3 Champs so far (and I think this is game 6), the more interesting in content and with the most unusual setup, arose out of a 1.b3 line. So people definitely miss a lot of fun when they do not pay enough attention to 'off' choices.

In any case, I am almost certain that both 1.f3, as well as 1.e4/d4 f6, are objectively much better options than at least half of the lines you will find in most books. And they offer the engines better chances to show their real capabilities at that.

But things are like they are: for the time being I will be the only crazy bloke who on 1.e4 will consistently reply with f6 (not entirely true, because I choose f6 just sometimes; perfectly playable, btw., still almost equal, but some sweet talk will justify an exaggeration)

"Dubious therefore playable" - isn't that a famous quote by Tartakover?
Well, I think the context there was human play, but since engines don't always excel at the opening, forcing them out of book early with offbeat, dubious moves has its merits.

1.b3 isn't all that dubious; it's a big part of my repertoire, in fact

Anyway, you can check out my recent post about the Don Dailey inspired chess 'variant', where each side moves a pawn to the 3rd rank randomly, and then the game begins on move 2.

http://www.talkchess.com/forum/viewtopi ... 94&t=49059

[lucasart]

[d]r1bqk1nr/1p2b1pp/p1np4/4p3/2P1Pp2/N1NB4/PP3PPP/R1BQK2R w KQkq - 0 10

In this position, Stockfish4 (and the latest updates), tends to play the baffling 10.Qh5+ [not every single time], and after the obvious 10..g6, it will play 11.Qd1, thus losing 2 tempos for no good reason. Black can follow up with Nf6, and eventually mobilize a pawn storm on the K-side (if capable of it, of course).

I've done some re-tracing, and this tendency to favor the tempo-losing Qh5 began with version 2.3 and has been around ever since.

Qh5 is not a bad move. It loses tempo to provoque the weakening g6. Which is more important between the tempo and the g6 weakness could be discussed endlessly. But, at the end of the day, the proof is in the pudding: SF is a world class engine, rated 3200, at least!

I think there are many other possibilities. For example, DiscoCheck prefers more normal developpement moves here:

Code: Select all

info score cp -4 depth 21 nodes 46140751 time 49726 pv c3d5 g8f6 c1d2 e8g8 e1g1 g8h8 d5e7 d8e7 a3c2 a6a5 c2a3 c8e6 a3b5 a8c8 f2f3 f6d7 d3e2 d7c5 a1c1 b7b6 d1c2

[carldaman]

[d]r1bqk1nr/1p2b1pp/p1np4/4p3/2P1Pp2/N1NB4/PP3PPP/R1BQK2R w KQkq - 0 10

In this position, Stockfish4 (and the latest updates), tends to play the baffling 10.Qh5+ [not every single time], and after the obvious 10..g6, it will play 11.Qd1, thus losing 2 tempos for no good reason. Black can follow up with Nf6, and eventually mobilize a pawn storm on the K-side (if capable of it, of course).

I've done some re-tracing, and this tendency to favor the tempo-losing Qh5 began with version 2.3 and has been around ever since.

Qh5 is not a bad move. It loses tempo to provoque the weakening g6. Which is more important between the tempo and the g6 weakness could be discussed endlessly. But, at the end of the day, the proof is in the pudding: SF is a world class engine, rated 3200, at least!

I think there are many other possibilities. For example, DiscoCheck prefers more normal developpement moves here:

Code: Select all

info score cp -4 depth 21 nodes 46140751 time 49726 pv c3d5 g8f6 c1d2 e8g8 e1g1 g8h8 d5e7 d8e7 a3c2 a6a5 c2a3 c8e6 a3b5 a8c8 f2f3 f6d7 d3e2 d7c5 a1c1 b7b6 d1c2

I beg to differ, because after 10.Qh5 g6, the only move is 11.Qd1 so 2 tempos have been wasted, since White has nothing to show for the last 2 moves; g7-g6 is not a weakness, since Black wants to expand on the K-side, where White is likely to castle. These are also the ideas evident in master games with this variation.

I'm not in the camp of blindly trusting a 3200 engine in all positions. Even top engines have certain weaknesses and 3200 does not equal perfection. Losing 2 tempos in the opening for no compensation makes little to no sense. If we take the attitude that so-and-so 3200-engine played this, therefore it must be good, improvement in these problematic areas will come later rather than sooner.