A french defence game

[Lyudmil Tsvetkov]

This time with the correct diagram.

[d] r1bq1r2/p4pk1/1pn1p3/3pPnPQ/2pP3P/P1P1B3/2P2P1N/R3K2R w KQ - 1 17


the thing that I wonder is why Komodo varies between +100 and -100cps or so on each even/uneven ply?

SF does not do so, but its score also varies a lot.

I did not see any such oscillation of scores using the latest Komodo. The evaluation was negative, then 0.00 for a while, and then positive once Qf3 was found. Here's depth 40:

Code: Select all

+1.32  17. Qf3 Rh8 18. Ng4 Nxh4 19. Qf6+ Qxf6 20. gxf6+ Kf8 21. Bh6+ Ke8 22. Rxh4 Kd7 23. Rh5 Bb7 24. Rg5 Rh7 25. Rg7 Rah8 26. Ke2 Nd8 27. Rxh7 Rxh7 28. Bf8 Bc6 29. Bd6 Rh8 30. Ne3 Kc8 31. Rg1 Be8 32. Rg7 Rh4 33. Kf3 Rh1 34. Rg8 Kd7 35. Ke2 Rh2 36. Ng4 Rh3 37. f3 Nc6 38. Kf2 a6 39. Kg2 Rh5 40. Ne3 Na7 41. a4 Rh4 (depth 40, 0:16:34)

Isn't this the same line given by asmFish, at least for a dozen plies or so?

Now Komodo is just a bit more optimistic:

Code: Select all

+1.66 17. Qf3 Rh8 18. Ng4 Nxh4 19. Qf6+ Qxf6 20. gxf6+ Kf8 21. Bh6+ Ke8 22. Rxh4 Ba6 23. Rh5 Kd7 24. Rg5 Rh7 25. Rg7 Rah8 26. Kd2 Nd8 27. Rxh7 Rxh7 28. Bf8 Bb5 29. Rg1 Rh8 30. Be7 Nc6 31. Bd6 Nd8 32. Ne3 Kc8 33. Rg7 Be8 34. f4 Rh2+ 35. Kd1 Rh8 36. Ke1 Kd7 37. Kf2 Kc6 38. Be7 Nb7 39. Kg2 Kb5 40. Kg3 Kc6 41. Bb4 a5 42. Be7 a4 43. Kg2 Na5 44. f5 Kd7 45. Kf3 Nc6 46. Bd6 Na7 47. fxe6+ Kxe6 48. Kf2 (depth 50, 2:04:03)

same cr*p line.

could you plese check, when you have time, the h5 line?

[Lyudmil Tsvetkov]

Very interesting. The IDeA computers are working on something else at the moment, but they should finish in 4 days. I'll point them over to this variation at that time and let them spend a few days. Advantages in the early opening rarely hold up over the long-term. A static eval of the position may very well reach 70 or 80 cp for the initial board position. It will be interesting to see if extending the lines confirms that assessment or black squeaks out with a defense.

Thanks for your insights and commentary.
Mike.

you are most than welcome, thanks for your feedback.

I am certain engines you are analysing with will squeak out here, current top engines simply don't have sufficient enought refined eval for this, and more time does not help, as they will be pruning a lot.

whether an advantage holds up or not, will depend on the correctness of that eval advantage.

real eval advantages should hold up and increase, fake ones disappear into thin air.

what I know for certain is that black castling short loses the game, drawing chances give only castling long and Ba6 trade.

[Lyudmil Tsvetkov]

indeed, h5 is the correct solution, and wins easier.

Louis' SF, however, that is currently wasting a lot of expensive electricity, calculating irrelevant nodes evaluated incorrectly, will never see the winning line, or even come close to it.

why?

because both SF and Komodo don't see a line like that one, for example, that is vital for the assessment of the whole variation:

Qf3 Rh8 h5 Kf8(failing to retreat with the king to the queen side should win for white very quickly, I suppose engines see that) Ng4 Ke7 g6 Kd7 g7!(engines see that, but not the continuation, scoring it as 0.0, which is completely off target) Ng7(SF and Komodo both 0.0 ) Qf7 Qe7:

[d]r1b4r/p2kqQn1/1pn1p3/3pP2P/2pP2N1/P1P1B3/2P2P2/R3K2R w KQ - 0 7

engines here change queens and draw, but this is easily won for white after Nf6 instead, and then Kc7 Qg6!(why is this such a difficult move for engines to recognise?):

[d]r1b4r/p1k1q1n1/1pn1pNQ1/3pP2P/2pP4/P1P1B3/2P2P2/R3K2R b KQ - 0 8

this is easily won for white

white places a passer on h7, gets the queenside rook to support the onslaught and it is, like GMs like to say, 'Curtains'.

such obvious easy moves, my assessment is at least 300cps white advantage, but engines don't see it...

so, it is better no more electricity is wasted on this, better try to improve a bit SF code.

[Lyudmil Tsvetkov]

I suppose there are also a lot of alternative winning lines, also in the Ng4 instead of h5 variation, white evaluation advantage is simply tremendous, this is just an ilustration how engines easily go astray in any position requiring deeper search and more sophisticated eval.

[MikeGL]

indeed, h5 is the correct solution, and wins easier.

Louis' SF, however, that is currently wasting a lot of expensive electricity, calculating irrelevant nodes evaluated incorrectly, will never see the winning line, or even come close to it.

why?

because both SF and Komodo don't see a line like that one, for example, that is vital for the assessment of the whole variation:

Qf3 Rh8 h5 Kf8(failing to retreat with the king to the queen side should win for white very quickly, I suppose engines see that) Ng4 Ke7 g6 Kd7 g7!(engines see that, but not the continuation, scoring it as 0.0, which is completely off target) Ng7(SF and Komodo both 0.0 ) Qf7 Qe7:

[d]r1b4r/p2kqQn1/1pn1p3/3pP2P/2pP2N1/P1P1B3/2P2P2/R3K2R w KQ - 0 7

engines here change queens and draw, but this is easily won for white after Nf6 instead, and then Kc7 Qg6!(why is this such a difficult move for engines to recognise?):

[d]r1b4r/p1k1q1n1/1pn1pNQ1/3pP2P/2pP4/P1P1B3/2P2P2/R3K2R b KQ - 0 8

this is easily won for white

white places a passer on h7, gets the queenside rook to support the onslaught and it is, like GMs like to say, 'Curtains'.

such obvious easy moves, my assessment is at least 300cps white advantage, but engines don't see it...

so, it is better no more electricity is wasted on this, better try to improve a bit SF code.

How can white get +3 or +4 cps here when the position is still unclear? You need to replace your buggy engine and download the official SF8 dev build to avoid confusing other posters.

[pgn]
[Event "Computer chess game"]
[Site "Talkchess"]
[Date "2017.07.26"]
[Round "?"]
[White "?"]
[Black "McBrain_2017"]
[Result "*"]
[SetUp "1"]
[FEN "r1bq1r2/p4pk1/1pn1p3/3pPnPQ/2pP3P/P1P1B3/2P2P1N/R3K2R w KQ - 1 17"]
[Termination "unterminated"]
[PlyCount "8"]
[WhiteType "human"]
[BlackType "human"]

17. Qf3 Rh8 18. h5 Nxe3 19. h6+ Kf8 20. Qxe3 Ne7 {black holds}
[/pgn]


Black can hold on the final position of this line, since black is 1pc ahead
[d]r1bq1k1r/p3np2/1p2p2P/3pP1P1/2pP4/P1P1Q3/2P2P1N/R3K2R w KQ - 1 21

That's a friendly suggestion, download the official engine release and don't
use the buggy engine which you've compiled using a buggy compiler.


regards

[zullil]

could you please check, when you have time, the h5 line?

My Komodo has been checking for many hours. Unlike me, it never needs to sleep.

[d] r1bq3r/p4pk1/1pn1p3/3pPnPP/2pP4/P1P1BQ2/2P2P1N/R3K2R b KQ - 0 18

Code: Select all

+1.31 18... Kf8 19. Ng4 Ke7 20. Kd2 Kd7 21. Nf6+ Kc7 22. g6 a5 23. Qg2 Qf8 24. Rag1 fxg6 25. hxg6 Rxh1 26. Rxh1 Kb8 27. Rh5 Nce7 28. Rh7 Nxg6 29. Qxg6 Ra7 30. Bh6 Qd8 31. Rxa7 Kxa7 32. Qf7+ Qe7 33. Qg8 Qc7 34. Bf8 Kb7 35. Ke2 Bd7 36. Bd6 Qc8 37. Qf7 Kc6 38. Kf3 Nh4+ 39. Kg4 Nf5 40. Bf8 Qd8 41. Kf4 Qc8 42. Kf3 Nh4+ 43. Kg4 Nf5 44. Kf4 Qd8 45. Bd6 Qc8 46. Qf8 Nh4 47. Kg3 Nf5+ 48. Kg4 Qb7 49. Qd8 Bc8 (depth 51, 7:46:52)

[zullil]

[d]r1b4r/p1k1q1n1/1pn1pNQ1/3pP2P/2pP4/P1P1B3/2P2P2/R3K2R b KQ - 0 8

this is easily won for white

white places a passer on h7, gets the queenside rook to support the onslaught and it is, like GMs like to say, 'Curtains'.

such obvious easy moves, my assessment is at least 300cps white advantage, but engines don't see it...

so, it is better no more electricity is wasted on this, better try to improve a bit SF code.

I guess Komodo needs improvement too, since it doesn't see that "this is easily won for White."

Code: Select all

0.00 24... Nf5 25. h6 Bd7 26. h7 Raf8 27. Bg5 Qf7 28. Qxf7 Rxf7 29. Kd2 Nce7 30. Rag1 Ng6 31. Nxd7 Rxd7 32. Bf6 Rhxh7 33. Rxh7 Rxh7 34. Rxg6 Rh2 35. Ke2 Rh3 36. Kd2 Rf3 37. Rg2 Rh3 38. Bg5 Rh1 39. Be3 Kd7 40. Rg6 Ra1 41. Rf6 a5 42. Rf7+ Kc8 43. Rf8+ Kd7 44. Bg5 Rxa3 45. Rf7+ Kc8 46. Rf8+ Kd7 (depth 50, 0:38:07)

I'll let asmFish take a look too, but my guess is that it will agree with Komodo.

[Ignacio]

My analisis:
1.Qf3 Rh8 2.h5 and now:
a) 2... Nxe3 3.fe3 Qxg5 4.0-0-0 winning.
b) 2... Kf8 3.Ng4 Ke7 4.g6 Kd7 5.Nf6+ Kc7 6.Kd2 and white is better.

[zullil]

[d]r1b4r/p1k1q1n1/1pn1pNQ1/3pP2P/2pP4/P1P1B3/2P2P2/R3K2R b KQ - 0 8

this is easily won for white

white places a passer on h7, gets the queenside rook to support the onslaught and it is, like GMs like to say, 'Curtains'.

such obvious easy moves, my assessment is at least 300cps white advantage, but engines don't see it...

so, it is better no more electricity is wasted on this, better try to improve a bit SF code.

I guess Komodo needs improvement too, since it doesn't see that "this is easily won for White."

Code: Select all

0.00 24... Nf5 25. h6 Bd7 26. h7 Raf8 27. Bg5 Qf7 28. Qxf7 Rxf7 29. Kd2 Nce7 30. Rag1 Ng6 31. Nxd7 Rxd7 32. Bf6 Rhxh7 33. Rxh7 Rxh7 34. Rxg6 Rh2 35. Ke2 Rh3 36. Kd2 Rf3 37. Rg2 Rh3 38. Bg5 Rh1 39. Be3 Kd7 40. Rg6 Ra1 41. Rf6 a5 42. Rf7+ Kc8 43. Rf8+ Kd7 44. Bg5 Rxa3 45. Rf7+ Kc8 46. Rf8+ Kd7 (depth 50, 0:38:07)

I'll let asmFish take a look too, but my guess is that it will agree with Komodo.

asmFish:

Code: Select all

0.00 24... Nf5 25. h6 Qf8 26. h7 Nce7 27. Qh5 Bd7 28. O-O-O Ng7 29. Bh6 Nef5 30. Qg5 Nxh6 31. Rxh6 Qxa3+ 32. Kd2 Nf5 33. Rg6 Ba4 34. Nxd5+ exd5 35. Qxf5 Rxh7 36. Rg8 Bxc2 37. Kxc2 Qb3+ 38. Kc1 Qxc3+ 39. Kb1 Qb3+ 40. Kc1 Qc3+ (depth 54, 0:28:20)

[Lyudmil Tsvetkov]

indeed, h5 is the correct solution, and wins easier.

Louis' SF, however, that is currently wasting a lot of expensive electricity, calculating irrelevant nodes evaluated incorrectly, will never see the winning line, or even come close to it.

why?

because both SF and Komodo don't see a line like that one, for example, that is vital for the assessment of the whole variation:

Qf3 Rh8 h5 Kf8(failing to retreat with the king to the queen side should win for white very quickly, I suppose engines see that) Ng4 Ke7 g6 Kd7 g7!(engines see that, but not the continuation, scoring it as 0.0, which is completely off target) Ng7(SF and Komodo both 0.0 ) Qf7 Qe7:

[d]r1b4r/p2kqQn1/1pn1p3/3pP2P/2pP2N1/P1P1B3/2P2P2/R3K2R w KQ - 0 7

engines here change queens and draw, but this is easily won for white after Nf6 instead, and then Kc7 Qg6!(why is this such a difficult move for engines to recognise?):

[d]r1b4r/p1k1q1n1/1pn1pNQ1/3pP2P/2pP4/P1P1B3/2P2P2/R3K2R b KQ - 0 8

this is easily won for white

white places a passer on h7, gets the queenside rook to support the onslaught and it is, like GMs like to say, 'Curtains'.

such obvious easy moves, my assessment is at least 300cps white advantage, but engines don't see it...

so, it is better no more electricity is wasted on this, better try to improve a bit SF code.

How can white get +3 or +4 cps here when the position is still unclear? You need to replace your buggy engine and download the official SF8 dev build to avoid confusing other posters.

[pgn]
[Event "Computer chess game"]
[Site "Talkchess"]
[Date "2017.07.26"]
[Round "?"]
[White "?"]
[Black "McBrain_2017"]
[Result "*"]
[SetUp "1"]
[FEN "r1bq1r2/p4pk1/1pn1p3/3pPnPQ/2pP3P/P1P1B3/2P2P1N/R3K2R w KQ - 1 17"]
[Termination "unterminated"]
[PlyCount "8"]
[WhiteType "human"]
[BlackType "human"]

17. Qf3 Rh8 18. h5 Nxe3 19. h6+ Kf8 20. Qxe3 Ne7 {black holds}
[/pgn]


Black can hold on the final position of this line, since black is 1pc ahead
[d]r1bq1k1r/p3np2/1p2p2P/3pP1P1/2pP4/P1P1Q3/2P2P1N/R3K2R w KQ - 1 21

That's a friendly suggestion, download the official engine release and don't
use the buggy engine which you've compiled using a buggy compiler.


regards

the only buggy thing here is you.

besides, I don't rely so much on SF, but much more on my own understanding.

I already pointed out in the introduction to this thread that on your very buggy line 19.h6 is a major positional/tactical mistake, and instead 19.fe3 wins quickly and convincingly, but you don't care to follow the thread to at least try to input some meaningful thoughts, and instead spread false claims, much to my chagrin.

if you don't understand much of chess, better stick to plain endgames: top engines might be able to save you there from time to time.