New engine releases & news H1 2024

[Frank Quisinsky]

Hi Graham,

I compare the two versions with Total Commander 11.02.
Can see small differences ... sure the programmer fixed a bug or changed anything what he forgot.

Good that you saw that.
Without your message I would have started the wrong version for my test.

Best
Frank

[jgilchrist]

Thus I've been testing this religiously in my engine (I even test refactors), and keeping scores of newly developed engines how many points they gain per feature.

Yes, I have been trying to be better about this recently too - everything since v1.1 is tested at STC and LTC. I'm also hoping to be able to get an OpenBench instance set up. One thing I struggle with though is the possibility that I make changes that are a nice optimisation on my M1 (for ARM) but end up being a regression on X86. I do like that you've put the Elo gains of each change in the changelog - I've put them in each commit but it's much nicer to see them listed like that.

So you are saying that if you match Tcheran's node counting to Rustic's, your engine is 50% (!) faster than mine? That would be massive. Could it be that your full legal move generator makes the engine 50% faster? If that is true, I'll have to look at this at some point, because it would gain at least 35 Elo. With regard to node counts... it's hard to measure engine speed by that, because everybody does it differently. Still, if I could scrap the legality check (and maybe even the is_check function, that uses the same square_attacked function) from the engine after the move generator is fully legal, then it could maybe become 50% faster. It could also still just be a difference in node counts. First though, I'll look into easier to add features, including staged move generation.

Of course it's always possible that there's some bug with how I'm counting nodes, but from a cursory check it looks that way. As you say, it is hard to compare. I suppose it's also possible that our features, despite being the same on the surface, differ slightly in the details. For example, I noticed that for check extensions, you extend at any depth which requires calculating whether you are in check in every node. By contrast, I only extend at the horizon, before entering quiescence (i.e. depth == 0). I did try modifying Rustic to do the same which appears to add ~200knps. Of course, that might still decrease its Elo, but it's just one example of how we might explain the differing NPS.

I doubt legal movegen can explain the difference, at least not to that extent. When I switched from pseudo-legal movegen it was roughly a 13% reduction in perft time, but I didn't test the resulting impact on search speed from no longer having to check for legality during search. I should go back and run more tests

Our feature-set is the same, and the only two node-cutting features both engines have are PVS and the Transposition Table. The TT adds 150-160 Elo which is... uh... exact the amount you seem to be missing.

...

Off the bat, my guess would be that your transposition table is having an issue.

Well, as it turns out, you were spot on. I looked back over my code for pruning via the TT and noticed that I was only pruning if I had TT entries from *higher* depths, but not from the same depth (i.e. > instead of >=). So I was still getting the benefit of the TT for move ordering, but missing out on a ton of pruning. One minuscule equals sign and the 8+0.08 test I'm running already has the fixed version up ~135 Elo after 124 games (and counting). Wow.

I can only say a huge thank you again. Time to release v2.1 I suppose.

[petero2]

I don't know what tbs it actually used though. Based on how it plays, maybe it is not using dtz tables and only relies on wdl tables.

A search in the source code confirms that Obsidian does not probe DTZ tables, which explains why it needs to search deep enough to win in complicated tablebase positions. It did manage to win the KQKR position also on my computer when given more thinking time. However, consider the following position:
[d]N7/3B4/8/8/8/8/5k2/7K w - - 0 1 I was able to draw two out of two times with black, when Obsidian was using 12 threads and 10 seconds per move. The first time I played DTM-optimal moves and additionally tried to obstruct the white king whenever possible. The second time I just retreated immediately to h8 and stayed in that corner as long as possible.

[pgn] [Event "?"] [Site "?"] [Date "2024.01.20"] [Round "?"] [White "Obsidian 10.0"] [Black "Player"] [Result "1/2-1/2"] [FEN "N7/3B4/8/8/8/8/5k2/7K w - - 0 1"] [SetUp "1"] 1. Kh2 Kf3 2. Nb6 Kf4 3. Bc6 Kf5 4. Nc4 Kg4 5. Bd5 Kf4 6. Bb7 Kg4 7. Bd5 Kf4 8. Bc6 Kg4 9. Kg2 Kf4 10. Bf3 Kf5 11. Kh2 Kf4 12. Bd5 Kg4 13. Kg2 Kf4 14. Kf2 Kg4 15. Bf3+ Kf4 16. Bg2 Kg4 17. Be4 Kf4 18. Bb7 Kg4 19. Ba8 Kf4 20. Ke2 Kg4 21. Bb7 Kf4 22. Kf2 Kg4 23. Ke2 Kf4 24. Kd3 Kf5 25. Ba8 Kf4 26. Bh1 Kf5 27. Bf3 Kf4 28. Be4 Kg5 29. Ke2 Kf4 30. Bc2 Kg4 31. Be4 Kf4 32. Bd3 Kg5 33. Bh7 Kh6 34. Bc2 Kh5 35. Kd2 Kg4 36. Be4 Kf4 37. Bh7 Kg5 38. Ke3 Kh6 39. Be4 Kg5 40. Bc6 Kf5 41. Kf3 Kg5 42. Bb7 Kf5 43. Bd5 Kg5 44. Nb6 Kf5 45. Nc4 Kg5 46. Ne3 Kf6 47. Kf4 Kg6 48. Nf5 Kf6 49. Bf7 Kxf7 1/2-1/2 [Event "?"] [Site "?"] [Date "2024.01.20"] [Round "?"] [White "Obsidian 10.0"] [Black "Player"] [Result "1/2-1/2"] [FEN "N7/3B4/8/8/8/8/5k2/7K w - - 0 1"] [SetUp "1"] 1. Kh2 Kf3 2. Nc7 Kf4 3. Bc6 Kg5 4. Kg3 Kg6 5. Kg4 Kg7 6. Kg5 Kh8 7. Bd5 Kg7 8. Bc4 Kh8 9. Kf6 Kh7 10. Ne6 Kh8 11. Ba2 Kh7 12. Ng5+ Kh8 13. Ne4 Kh7 14. Nc5 Kh8 15. Bd5 Kh7 16. Bb3 Kh8 17. Ne4 Kh7 18. Nf2 Kh8 19. Bc2 Kg8 20. Bd3 Kh8 21. Bg6 Kg8 22. Ne4 Kh8 23. Ng5 Kg8 24. Nf3 Kh8 25. Bd3 Kg8 26. Be4 Kh8 27. Nd4 Kg8 28. Bf3 Kh8 29. Bd1 Kg8 30. Be2 Kh8 31. Bb5 Kg8 32. Bc6 Kh8 33. Nb5 Kg8 34. Bd5+ Kh8 35. Nc7 Kh7 36. Bg2 Kh8 37. Ke5 Kg8 38. Ke6 Kh8 39. Be4 Kg8 40. Na6 Kh8 41. Kf6 Kg8 42. Bg2 Kh8 43. Bd5 Kh7 44. Bh1 Kh8 45. Nc7 Kg8 46. Ne8 Kh8 47. Nd6 Kg8 48. Ne4 Kh8 49. Nc3 Kg8 50. Nd5 Kh8 1/2-1/2 [/pgn]
Not using DTZ is however a different problem than the one Frank found, and which I was also seeing, although not all the time. In that case it could not even win a KQK endgame. I guess this is somehow related to the transposition table, because it does not happen if you set up a position where it went wrong in the game and analyze that position from scratch.

[Frank Quisinsky]

Hi Peter,

thank you Peter!
Shortly a question for you?

Could you look here ...
https://github.com/gab8192/Obsidian/com ... or=gab8192

The programmer wrote about "Fix TB scores reporting"
https://github.com/gab8192/Obsidian/com ... 155f448c82

I am not a programmer.
Maybe this fix the problem?

For me ...
I have to wait of an update and hope the programmer found the messages about the problem.
To have the version in my new tourney with such a bug isn't very nice.
So, if a fix is available I will start direct a new test.

Thank you again and ...

BTW:
Have a look in move-average win for your own.
Only Velvet can beat it very narrow.
So, maybe a bit animation for you ...



Have a nice weekend!

Best
Frank

[peter]

A search in the source code confirms that Obsidian does not probe DTZ tables, which explains why it needs to search deep enough to win in complicated tablebase positions. It did manage to win the KQKR position also on my computer when given more thinking time. However, consider the following position:
[d]N7/3B4/8/8/8/8/5k2/7K w - - 0 1 I was able to draw two out of two times with black, when Obsidian was using 12 threads and 10 seconds per move. The first time I played DTM-optimal moves and additionally tried to obstruct the white king whenever possible. The second time I just retreated immediately to h8 and stayed in that corner as long as possible.

[pgn] [Event "?"] [Site "?"] [Date "2024.01.20"] [Round "?"] [White "Obsidian 10.0"] [Black "Player"] [Result "1/2-1/2"] [FEN "N7/3B4/8/8/8/8/5k2/7K w - - 0 1"] [SetUp "1"] 1. Kh2 Kf3 2. Nb6 Kf4 3. Bc6 Kf5 4. Nc4 Kg4 5. Bd5 Kf4 6. Bb7 Kg4 7. Bd5 Kf4 8. Bc6 Kg4 9. Kg2 Kf4 10. Bf3 Kf5 11. Kh2 Kf4 12. Bd5 Kg4 13. Kg2 Kf4 14. Kf2 Kg4 15. Bf3+ Kf4 16. Bg2 Kg4 17. Be4 Kf4 18. Bb7 Kg4 19. Ba8 Kf4 20. Ke2 Kg4 21. Bb7 Kf4 22. Kf2 Kg4 23. Ke2 Kf4 24. Kd3 Kf5 25. Ba8 Kf4 26. Bh1 Kf5 27. Bf3 Kf4 28. Be4 Kg5 29. Ke2 Kf4 30. Bc2 Kg4 31. Be4 Kf4 32. Bd3 Kg5 33. Bh7 Kh6 34. Bc2 Kh5 35. Kd2 Kg4 36. Be4 Kf4 37. Bh7 Kg5 38. Ke3 Kh6 39. Be4 Kg5 40. Bc6 Kf5 41. Kf3 Kg5 42. Bb7 Kf5 43. Bd5 Kg5 44. Nb6 Kf5 45. Nc4 Kg5 46. Ne3 Kf6 47. Kf4 Kg6 48. Nf5 Kf6 49. Bf7 Kxf7 1/2-1/2 [Event "?"] [Site "?"] [Date "2024.01.20"] [Round "?"] [White "Obsidian 10.0"] [Black "Player"] [Result "1/2-1/2"] [FEN "N7/3B4/8/8/8/8/5k2/7K w - - 0 1"] [SetUp "1"] 1. Kh2 Kf3 2. Nc7 Kf4 3. Bc6 Kg5 4. Kg3 Kg6 5. Kg4 Kg7 6. Kg5 Kh8 7. Bd5 Kg7 8. Bc4 Kh8 9. Kf6 Kh7 10. Ne6 Kh8 11. Ba2 Kh7 12. Ng5+ Kh8 13. Ne4 Kh7 14. Nc5 Kh8 15. Bd5 Kh7 16. Bb3 Kh8 17. Ne4 Kh7 18. Nf2 Kh8 19. Bc2 Kg8 20. Bd3 Kh8 21. Bg6 Kg8 22. Ne4 Kh8 23. Ng5 Kg8 24. Nf3 Kh8 25. Bd3 Kg8 26. Be4 Kh8 27. Nd4 Kg8 28. Bf3 Kh8 29. Bd1 Kg8 30. Be2 Kh8 31. Bb5 Kg8 32. Bc6 Kh8 33. Nb5 Kg8 34. Bd5+ Kh8 35. Nc7 Kh7 36. Bg2 Kh8 37. Ke5 Kg8 38. Ke6 Kh8 39. Be4 Kg8 40. Na6 Kh8 41. Kf6 Kg8 42. Bg2 Kh8 43. Bd5 Kh7 44. Bh1 Kh8 45. Nc7 Kg8 46. Ne8 Kh8 47. Nd6 Kg8 48. Ne4 Kh8 49. Nc3 Kg8 50. Nd5 Kh8 1/2-1/2 [/pgn]
Not using DTZ is however a different problem than the one Frank found, and which I was also seeing, although not all the time. In that case it could not even win a KQK endgame. I guess this is somehow related to the transposition table, because it does not happen if you set up a position where it went wrong in the game and analyze that position from scratch.

I updated the file:
obsidian-bug.zip

https://www.amateurschach.de/download/obsidian-bug.zip (1.3Mb).
Inside the complete 850 games in *.pgn and the *.cbh database with the 56 collected bug-games.
Maybe more of such games are in database, looking very fast in the afternoon.

Gave the first one game in .cbh (by the way, the .pgn- version isn't of same order of games as .cbh is) Renegade- Obsidian a try and indeed single threaded with 5"/move Obsidian with 6men tbs didn't make the full point against best moves out of online- tbs, starting from 4men- position at 82nd move of the game:
[pgn] [Event "4 Minutes/Game + 2 Seconds/Move"] [Site "test-19, WASP-1"] [Date "2024.01.17"] [Round "3.14"] [White "Renegade 1.0.0 NN"] [Black "Obsidian 10.0 NN"] [ECO "C01"] [Result "1/2-1/2"] 1. e4 {book 0s} e6 {book 0s} 2. d4 {book 0s} d5 {book 0s} 3. exd5 {book 0s} exd5 {book 0s} 4. Bd3 {book 0s} Bd6 {book 0s} 5. Ne2 {book 0s} Nf6 {book 0s} 6. O-O {book 0s} O-O {book 0s} 7. Nbc3 {0.00/25 7s} Re8 {-0.11/25 10s} 8. Ng3 {-0.01/27 10s} Bg4 {0.00/22 8s} 9. Nce2 {+0.07/23 4s} c5 {+0.04/23 6s} 10. dxc5 {+0.01/26 6s} Bxc5 {+0.08/25 5s} 11. h3 {+0.11/26 5s} Bxe2 {+0.08/24 7s} 12. Bxe2 {+0.01/25 5s} Nc6 {+0.19/26 13s} 13. Bg5 {+0.03/25 6s} h6 {+0.18/28 11s} 14. Bxf6 {+0.35/26 4s} Qxf6 {+0.14/28 5s} 15. Qxd5 {0.00/28 7s} Bb6 {+0.13/26 5s} 16. Rab1 {+0.01/28 11s} Qh4 {+0.54/26 11s} 17. Qd3 {0.00/29 18s} g6 {+0.68/22 4s} 18. Rbd1 {0.00/29 6s} h5 {+0.67/25 6s} 19. Kh1 {-0.10/25 8s} Rad8 {+0.94/26 9s} 20. Qc3 {-0.25/27 6s} Rxd1 {+0.70/27 9s} 21. Bxd1 {-0.22/28 4s} Qa4 {+0.84/28 6s} 22. Bf3 {-0.30/26 8s} h4 {+0.84/26 4s} 23. Ne4 {-0.50/29 6s} Rxe4 {+1.18/29 5s} 24. Bxe4 {-0.51/28 3s} Qxe4 {+1.26/27 5s} 25. Re1 {-0.41/30 4s} Qd4 {+1.33/29 8s} 26. Qxd4 {-0.30/29 3s} Bxd4 {+1.25/29 5s} 27. Kg1 {-0.44/30 5s} Kf8 {+1.34/30 11s} 28. Kf1 {-0.53/29 6s} Be5 {+1.35/27 4s} 29. b4 {-0.52/31 9s} f6 {+1.26/28 7s} 30. b5 {-0.56/28 3s} Nd4 {+1.22/28 3s} 31. c4 {-0.44/30 5s} g5 {+1.14/30 8s} 32. Rd1 {-0.43/30 4s} Ke7 {+1.31/30 4s} 33. Rd3 {-0.41/29 2s} b6 {+1.09/31 14s} 34. Ra3 {-0.31/30 4s} Bb8 {+0.97/30 9s} 35. Rd3 {-0.40/28 2s} Ne6 {+1.21/27 3s} 36. g3 {-0.34/30 5s} hxg3 {+1.40/29 10s} 37. fxg3 {-0.38/33 10s} Ng7 {+1.23/28 4s} 38. Re3+ {-0.30/27 4s} Kd7 {+1.09/30 4s} 39. Rd3+ {-0.49/29 6s} Bd6 {+1.50/25 4s} 40. c5 {-0.40/29 9s} bxc5 {+1.49/24 3s} 41. Ra3 {-0.65/30 6s} Bb8 {+1.35/25 4s} 42. Ra6 {-0.51/28 3s} f5 {+1.25/26 5s } 43. Ke2 {-0.46/27 5s} c4 {+1.30/27 9s} 44. g4 {-0.48/29 4s} fxg4 {+1.39/30 7s} 45. hxg4 {-0.52/29 2s} Ne6 {+1.23/32 8s} 46. Ra4 {-0.53/30 3s} c3 {+1.55/26 2s} 47. Kd1 {-0.79/29 5s} Nf4 {+1.44/28 2s} 48. Kc2 {-0.73/26 2s} Nd5 {+1.50/29 2s} 49. Ra5 {-0.88/28 4s} Kd6 {+2.07/25 2s} 50. Kd3 {-0.73/29 4s} Bc7 {+2.14/25 2s} 51. Rxa7 {-0.96/29 4s} Kc5 {+2.48/27 2s} 52. a3 {-0.88/31 2s} Kxb5 {+2.75/27 3s} 53. Ra8 {-1.05/32 8s} Bf4 {+2.96/27 2s} 54. Rd8 {-1.29/31 8s} Kc6 {+3.07/26 2s} 55. Rc8+ {-1.34/30 3s} Kd7 {+3.17/26 2s} 56. Rc4 {-1.35/31 5s} Kd6 {+3.31/26 2s} 57. Re4 {-1.32/32 3s} Bd2 {+3.58/25 2s} 58. a4 {-1.15/27 3s} Nf4+ {+4.24/22 2s} 59. Kc2 {-1.50/27 2s} Kd5 {+5.04/24 3s} 60. Re7 {-1.73/28 11s} Kc5 {+5.90/23 2s} 61. Re5+ {-1.69/22 3s} Kd4 {+6.37/25 2s} 62. Rxg5 {-0.33/25 1s} Ng2 {+42.80/26 2s} 63. Rg8 {-0.45/26 3s} Ne3+ {+M278/44 2s} 64. Kb3 {-0.48/29 2s} c2 {+M290/46 4s} 65. Rd8+ {-0.48/29 2s} Ke4 {+M270/46 2s} 66. Re8+ {-0.48/30 5s} Kf4 {+M262/51 3s} 67. Rc8 {-0.55/27 2s} c1=Q {+M252/60 4s} 68. Rxc1 {-0.56/24 1s} Bxc1 {+M250/77 5s} 69. g5 {-0.57/27 2s} Kxg5 {+M248/73 2s} 70. a5 {-0.61/28 2s} Nf5 {+M264/70 3s} 71. Kc4 {-0.61/26 1s} Bd2 {+M254/80 5s} 72. a6 {-0.62/28 1s} Be3 {+124.36/41 2s} 73. Kd5 {-0.61/29 1s} Bf2 {+126.99/47 4s} 74. Kc6 {-0.61/28 4s} Nd4+ {+M294/56 3s} 75. Kb7 {-0.61/29 2s} Nb5 {+M268/69 2s} 76. Kc6 {-0.60/30 1s} Na7+ {+M266/76 1s} 77. Kd5 {-0.60/31 1s} Kf5 {+M260/81 6s} 78. Kd6 {-0.60/31 2s} Ke4 {+M258/77 2s} 79. Ke6 {-0.67/29 3s} Nc6 {+M252/82 9s} 80. Kf6 {-0.66/26 2s} Nb4 {+M250/74 1s} 81. a7 {-0.67/26 2s} Bxa7 {+M248/65 1s} 82. Ke7 {-0.67/22 1s} (82. Ke6 {35s} Bb8 {-4.85/39 9s} 83. Kd7 {21s} Kd5 {-5.05/43 7s} 84. Kc8 {3s} Bf4 {-143.10/42 8s} 85. Kb7 {2s} Kc5 {-142.53/52 8s} 86. Ka8 {3s} Nd3 {-114.12/47 7s} 87. Ka7 {2s} Nb4 {-103.03/46 9s} 88. Kb7 {2s} Bd6 {-102.05/49 9s} 89. Ka8 {2s} Kb5 {-71.05/49 9s} 90. Kb7 {2s} Nd3 {-69.97/49 7s} 91. Ka7 {2s} Kc6 {-68.72/45 9s} 92. Ka8 {2s} Ne5 {-68.54/47 7s} 93. Ka7 {38s} Nd7 {-4.22/35 4s} 94. Ka8 {3s} Nc5 {-4.22/24 4s} 95. Ka7 {2s} Nb7 {-4.22/37 3s} 96. Ka6 {2s} Bc7 {-4.22/22 4s} 97. Ka7 {2s} Bd6 {-4.18/22 4s} 98. Ka6 {2s} Bh2 {-4.18/21 4s} 99. Ka7 {3s} Be5 {-4.22/37 4s} 100. Ka6 {3s} Nc5+ {-4.35/39 4s} 101. Ka7 {1s} Nd7 {-135.98/41 4s} 102. Ka8 {2s} Bd6 {-135.78/42 4s} 103. Ka7 {2s} Nb6 {-91.39/38 4s} 104. Ka6 {2s} Na4 {-101.15/40 3s} 105. Ka7 {1s} Nb2 {-57.86/37 4s} 106. Ka8 {2s} Bc7 {-57.94/40 4s} 107. Ka7 {2s} Be5 {-76.36/37 4s} 108. Ka8 {3s} Na4 {-57.94/37 4s} 109. Ka7 {2s} Nc3 {-91.16/37 4s} 110. Ka8 {2s} Ne2 {-57.38/35 4s} 111. Ka7 {2s} Bd6 {-49.52/36 4s} 112. Ka8 {2s} Nd4 {-49.37/36 4s} 113. Ka7 {2s} Nb3 {-49.24/37 4s} 114. Ka6 {2s} Bf4 {-48.80/33 4s} 115. Ka7 {1s} Nc5 {-48.81/34 4s} 116. Ka8 {1s} Na4 {-48.38/34 4s} 117. Ka7 {1s} Nc3 {-48.51/31 4s} 118. Ka8 {2s} Be3 {-48.38/27 4s} 119. Kb8 {1s} Bg5 {-48.27/23 4s} 120. Ka8 {2s} Na4 {-3.57/26 4s} 121. Ka7 {2s} Bf4 {-3.25/26 4s} 122. Ka6 {2s} Nb6 {-1.95/24 4s} 123. Ka7 {2s} Nc4 {0.00/38 4s} 124. Ka6 {1s} Bd6 {0.00/41 3s} 125. Ka7 {1s} Nb2 {0.00/50 3s}) 82... Bf2 {+92.85/36 5s} 83. Kf6 {-0.72/26 2s} Na6 {+51.34/37 1s} 84. Ke6 {-0.72/29 2s} Bd4 {+68.88/39 4s} 85. Kd6 {-0.72/29 2s} Be3 {+92.45/40 1s} 86. Kc6 {-0.59/19 1s} Nc5 {+51.16/38 0s} 87. Kb5 {-0.71/26 1s} Kd5 {+92.70/38 1s} 88. Ka5 {-0.71/28 2s} Kd6 {+52.55/39 3s} 89. Kb4 {-0.71/26 1s} Kd5 {+72.65/40 2s} 90. Ka5 {0.00/128 0s} Bd2+ {+72.35/42 3s} 91. Kb6 {-0.71/29 3s} Bc3 {+55.33/30 2s} 92. Kb5 {-0.71/29 1s} Nb3 {+62.90/38 1s} 93. Ka6 {-0.70/29 2s} Nc5+ {+62.90/36 2s} 94. Kb5 {0.00/128 0s} Nb3 {+89.78/37 0s} 95. Ka6 {0.00/128 0s} Kc6 {+69.77/40 1s} 96. Ka7 {0.00/1 0s} Nc5 {+59.70/41 3s} 97. Ka8 {-0.70/30 1s} Be5 {+59.17/34 1s} 98. Ka7 {0.00/1 0s} Bd4 {+62.10/40 2s} 99. Ka8 {-0.69/29 1s} Ne6 {+62.26/42 2s} 100. Kb8 {0.00/1 0s} Bb6 {+61.95/29 0s} 101. Kc8 {-0.69/30 1s} Nc5 {+70.82/40 1s} 102. Kb8 {0.00/1 0s} Nb3 {+62.09/42 3s} 103. Kc8 {-0.69/30 2s} Kc5 {+62.40/41 1s} 104. Kb8 {-0.68/31 2s} Ba5 {+62.03/41 1s} 105. Kb7 {-0.68/31 2s} Kd5 {+61.87/39 1s} 106. Ka6 {-0.68/30 2s} Bd8 {+61.86/37 1s} 107. Kb5 {-0.68/30 1s} Be7 {+61.66/39 2s} 108. Kb6 {-0.68/31 2s} Nd4 {+61.74/39 2s} 109. Kb7 {-0.68/32 2s} Bd8 {+61.74/41 2s} 110. Ka8 {-0.68/32 1s} Ba5 {+61.61/27 1s} 111. Kb8 {-0.67/32 4s} Nb3 {+61.74/36 3s} 112. Ka7 {-0.67/32 1s} Ke5 {+59.86/34 1s} 113. Kb8 {-0.67/32 3s} Nd4 {+59.70/32 1s} 114. Ka8 {-0.67/33 2s} Bb4 {+61.45/30 3s} 115. Kb7 {-0.67/33 4s} Bd6 {+59.70/29 2s} 116. Kb6 {-0.67/32 3s} Ne6 {+50.25/22 2s} 117. Kb7 {-0.67/31 5s} Kd5 {+48.55/23 0s} 118. Kc8 {-0.67/29 2s} Bh2 {+43.17/22 2s} 119. Kb7 {-0.67/28 3s} Bg3 {+43.48/23 2s} 120. Ka8 {-0.66/24 2s} Bf4 {+29.70/20 2s} 121. Kb7 {-0.66/23 3s} Bd2 {+29.60/19 2s} 122. Kb6 {-0.66/21 7s} Be3+ {+5.15/21 0s} 123. Kb7 {-0.66/19 3s} Bg5 {+29.70/19 3s} 124. Kb6 {-0.48/17 2s} Bf6 {+5.86/20 2s} 125. Kb7 {0.00/16 1s} Nc5+ {+4.00/16 2s} 126. Kb8 {0.00/29 1s} Nb3 {+3.78/19 2s} 127. Kc7 {0.00/43 1s} Ke5 {0.00/34 1s} 128. Kb7 {0.00/103 1s} Kd4 {0.00/40 0s} 129. Kc8 {0.00/128 0s} Be5 {0.00/40 0s} 130. Kb7 {0.00/128 0s} Nc5+ {0.00/40 0s} 131. Kc8 {0.00/128 0s} Bc7 {0.00/40 0s} 1/2-1/2 [/pgn]
Regard side-line starting at nr. 82.
So I yet could reproduce missing dtz- interpretation of the engine in this one example too, regards

[Frank Quisinsky]

Hi Peter,

sorry for my "wirrwarr" in *.zip file.
Now the 56 games as *.pgn only are in the obsidian-bug.zip file.

https://www.amateurschach.de/download/obsidian-bug.zip (96.099 bytes)

Best
Frank

[petero2]

Hi Peter,

thank you Peter!
Shortly a question for you?

Could you look here ...
https://github.com/gab8192/Obsidian/com ... or=gab8192

The programmer wrote about "Fix TB scores reporting"
https://github.com/gab8192/Obsidian/com ... 155f448c82

I am not a programmer.
Maybe this fix the problem?

Hi Frank,

It does fix one problem where mate scores and TB scores were sort of mixed up and sometimes caused the search to not make progress.

It does not however fix the missing DTZ problem. This problem is perhaps not common in games though. I think in most cases KQ vs KR can be solved by search without tablebases, and KBN vs K is quite rare.

It also does not fix the "shuffling instead of mating problem" that sometimes happens, even when not using tablebases. Two examples:
[pgn] [Event "?"] [Site "?"] [Date "2024.01.20"] [Round "?"] [White "Obsidian dev-10.03"] [Black "Player"] [Result "1/2-1/2"] [FEN "8/2k4B/4K3/4N3/8/8/8/8 w - - 6 4"] [SetUp "1"] 4. Nd7 Kb7 5. Bd3 Kc6 6. Be2 Kc7 7. Bf3 Kd8 8. Be4 Kc7 9. Ke7 Kc8 10. Kd6 Kd8 11. Ke6 Kc7 12. Ke7 Kc8 13. Kd6 Kd8 14. Bd3 Ke8 15. Bg6+ Kd8 16. Nc5 Kc8 17. Nd7 Kd8 18. Nc5 Kc8 19. Nd7 Kd8 20. Be4 Ke8 21. Ke6 Kd8 22. Kd6 Ke8 23. Ke6 Kd8 24. Bh7 Kc7 25. Be4 Kd8 26. Bh7 Kc7 27. Be4 Kd8 28. Bg6 Kc7 29. Bh7 Kb7 30. Bd3 Kc6 31. Be2 Kc7 32. Bf3 Kd8 33. Ba8 Kc7 34. Ke7 Kc8 35. Bf3 Kc7 36. Bd5 Kc8 37. Kd6 Kd8 38. Bf7 Kc8 39. Bd5 Kd8 40. Bf7 Kc8 41. Bc4 Kd8 42. Bd3 Ke8 43. Bc4 Kd8 44. Bd5 Ke8 45. Ke6 Kd8 46. Kd6 Ke8 47. Ke6 Kd8 48. Kf7 Kc8 49. Ke7 Kc7 50. Bb7 Kxb7 1/2-1/2 [Event "?"] [Site "?"] [Date "2024.01.20"] [Round "?"] [White "Obsidian dev-10.03"] [Black "Player"] [Result "1/2-1/2"] [FEN "8/2k4B/4K3/4N3/8/8/8/8 w - - 0 1"] [SetUp "1"] 1. Nd7 Kb7 2. Bd3 Kc6 3. Bf1 Kc7 4. Bg2 Kd8 5. Bc6 Ke8 6. Ba4 Kd8 7. Bc6 Ke8 8. Ba4 Kd8 9. Bb5 Ke8 10. Bc6 Kd8 11. Ne5 Kc7 12. Kd5 Kb6 13. Kd6 Ka5 14. Kc5 Ka6 15. Nf7 Ka5 16. Nd6 Ka6 17. Nc8 Ka5 18. Nd6 Ka6 19. Nc8 Ka5 20. Nb6 Ka6 21. Nc4 Ka7 22. Kd6 Kb8 23. Kd7 Ka7 24. Kc7 Ka6 25. Ba4 Ka7 26. Bc6 Ka6 27. Ba4 Ka7 28. Bd7 Ka6 29. Bc6 Ka7 30. Bd7 Ka6 31. Bc6 Ka7 32. Ba4 Ka6 33. Kd6 Kb7 34. Kd7 Kb8 35. Bc6 Ka7 36. Bf3 Ka6 37. Bc6 Ka7 38. Bf3 Ka6 39. Bc6 Ka7 40. Ba4 Kb8 41. Bb5 Kb7 42. Ba4 Kb8 43. Bb5 Kb7 44. Kd6 Kc8 45. Bc6 Kd8 46. Ba4 Kc8 47. Bb5 Kd8 48. Ne3 Kc8 49. Ke7 Kb7 50. Nd5 Ka7 1/2-1/2 [/pgn]
This bug seems very sensitive to exact initial conditions though. In those games the engine was freshly started before the game, using 256MB hash, one thread and no TBs. Note that the first game starts with a non-zero half-move clock. Also I expect this to behave differently on 10.0 vs dev-10.03, but trying with some different hash sizes may make the problem possible to reproduce.

[Guenther]

Bug reporting is very important + interesting (if done properly), also publically - not only in github and I did this myself a lot too, but it should be done in own threads, now this 'new releases thread' is a total mess already after three weeks :(

Just to explain again (and again) I started those threads some years ago for future maintainers of something like my uci/xb chronology
in case I would lose motivation, or won't be able to collect the data anymore. If informations are kept tight together, research won't need
a lot of time again.

Also questions about new releases, or matches/games deserve own new threads in the right subforum with the name of the program in the header :)

Maybe moderation can split parts away into different threads...

Günther

[JoAnnP38]

Pedantic 0.6.2 -- a primarily defect correction release intended to keep Pedantic off the bottom of Div 5 and sent packing to Div 6.

+34 +/-8 Elo @ 20+0.2 TC from self-play.

[mar]

Cheng 4.44 is out:
https://github.com/kmar/cheng4/releases/tag/4.44

now with avx2 binaries - the preferred ones to use
see release notes for more details

60 + 0.6 seconds:
cheng4_43 vs cheng4_dev: 355 - 1173 - 1800 [0.377] 3328
-87.2 +/- 7.9

expect 30-40 elo against other programs

have fun