Experimental procedure to refute openings. Application to the Latvian gambit.

Here's my strategy. I take a position and generate a sufficient number of Explorer vs Explorer games taking the best moves at depth d=20. When white/black wins are overwhelming the opening can be considered as refuted.

In order to illustrate it I'll check whether the Latvian gambit 1. e4 e5 2. Nf3 f5 is really refuted by 3. Nxe5. I'm starting with 4 games with 4 best replies for black:

• 3...Qf6 (post #2) 1-0
• 3...Nc6 (post #3) 1-0
• 3...Nf6
• 3...Qe7

I'll inform U how these games progress.

First game, the current status:

3...Qf6 4. d4 d6 5. Nc4 fxe4 6. Nc3 Qg6 7. f3 Nc6 8. Nd5 Kd8 9. c3 Nf6 10. Nxf6 exf3 11. Qxf3 Qxf6 12. Qxf6+ gxf6 13. Bd3 Bg7 14. Ne3 Ne7 15. O-O Be6 16. Bd2 Kd7 17. Rae1 Rae8 18. a4 Bf7 19. Ng4 Ng8 20. Ne3 Bh6 21. Bf5+ Kd8 22. g3 Ne7 23. Bd3 Nd5 24. Nxd5 Rxe1 25. Bxe1 Bxd5 26. c4 Be3+ 27. Bf2 Bxf2 28. Rxf2 Bc6 29. Rxf6 Bxa4 30. Rh6 Rf8 31. Kg2 Bd1 32. Rxh7 Bf3+ 33. Kg1 Bd1 34. Bf1 a5 35. Bh3 Ba4 36. Bg4 Re8 37. Kf2 Re4 38. Rh8+ Ke7 39. Be2 Rxd4 40. Rh7+ Ke6 41. Ke3 c5 42. Rxb7 Bd1 43. Bd3 a4 44. Rb8 Ke7 45. Rh8 Kf7 46. Kd2 Bb3 47. Kc3 Bd1 48. h3 Bf3 49. g4 Be4 50. Be2 Kg7 51. Rd8 Bg2 52. h4 Re4 53. Bd3 Rxg4 54. Rxd6 Bf3 55. h5 Rh4 56. Rg6+ Kf7 57. h6 Be4 58. Bxe4 Rxe4 59. Rg7+ Kf6 60. Rc7 Re3+ 61. Kc2 Re2+ 62 Kb1 Kg6 63. h7 Rh2 64. Ka2 Rh3 65. Rxc5 Rxh7 66. Ka3 Rh2 67. Ra5 Kf7 68. Rxa4 and white wins in 26 moves (Nalimov tablebases) 1-0.

It was a long game with constant white advantage and 2 big exchanges. The advantage somewhat diminished during the middlegame then increased again during the more advantageous endgame.

Okay, now I'm switching to the 2nd test game.

3...Nc6 4. Qh5+ g6 5. Nxg6 Nf6 6. Qh3 Rg8 7. Nxf8 Kxf8 8. d3 fxe4 9. dxe4 Rg6 10. Qh4 d5 11. Bg5 Qd6 12. Bxf6 Rxf6 13. Nc3 Rh6 14. Qg5 dxe4 15. Rd1 Rg6 16. Qh4 Qf6 17. Qxf6 Rxf6 18. Nxe4 and with 2 additional pawns white will easily win the endgame, 1-0.

I've simulated only 2 test games but the conclusion is already clear. Black has not a single chance to win it with the correct play by white. Maybe, in the best case, they could rarely achieve a draw in some miraculous sublines of 3...Qf6. So, I guess that the Latvian gambit is clearly refuted.

An opening is called statistically refuted if one side has an overwhelming number of wins in a big sample of games of 2 masters / engines of the same strength.