TCEC - Deciding a drawn result

[MikeGL]

So I got a bit of a free time and browsed latest TCEC games played last Jan 17, 2021
by the top 2 contenders at superfinal Lc0 and SF,

[pgn]
[Event "TCEC S20 Superfinal"]
[Site "?"]
[Date ""]
[Round "?"]
[White "?"]
[Black "?"]
[Result "0.5-0.5"]
[Termination "unterminated"]
[PlyCount "87"]
[WhiteType "Engine"]
[BlackType "Engine"]

1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 g6 6. Be3 Bg7 7. f3 O-O 8.
Qd2 Nc6 9. Bc4 Bd7 10. O-O-O Rc8 11. Bb3 Ne5 12. Kb1 Re8 13. h4 h5 14. g4
hxg4 15. h5 Nxh5 16. Bh6 Kh7 17. Bxg7 Kxg7 18. fxg4 Bxg4 19. Nf5+ Bxf5 20.
exf5 Rh8 21. fxg6 Nxg6 22. Qd5 e6 23. Qd4+ Qf6 24. Qxa7 Ng3 25. Rxh8 Rxh8
26. Qxb7 d5 27. a4 Qf4 28. Qb6 Nf5 29. Nb5 Qe3 30. Qc7 Qe5 31. Qd7 Qe2 32.
Rc1 Kf6 33. Qb7 Qe5 34. Ka2 Qf4 35. Rd1 Nge7 36. Qa6 Ne3 37. Ra1 N7f5 38.
Rc1 Qb4 39. Qc6 Qd2 40. Ra1 Rh2 41. Na3 Rh8 42. Qb6 Qd4 43. Qc6 Qd2 44. Qb6
[/pgn]

Who exactly decides a draw on the above final position when it is still unclear because _NO_ 3-fold repetition was announced yet.
Most engines have tendencies to repeat up to twice, but would avoid a 3-fold.

The above endgame still looks interesting, and I was so upset when the game was not finished.
This is one of those drawn games at TCEC where the position is still unclear but was automatically considered a draw when SF shows 0.00
We all know there are so many 0.00 positions evald by A/B but where actually won by NN-MTCS engines .

You can also check all TCEC drawn games between these two top engines and you'd noticed there are still some advantages in most games but was stopped and
considered a draw because some pairing (or tourney?) algo already called it a draw. How irritating.

[Guenther]

So I got a bit of a free time and browsed latest TCEC games played last Jan 17, 2021
by the top 2 contenders at superfinal Lc0 and SF,

[pgn]
[Event "TCEC S20 Superfinal"]
[Site "?"]
[Date ""]
[Round "?"]
[White "?"]
[Black "?"]
[Result "0.5-0.5"]
[Termination "unterminated"]
[PlyCount "87"]
[WhiteType "Engine"]
[BlackType "Engine"]

1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 g6 6. Be3 Bg7 7. f3 O-O 8.
Qd2 Nc6 9. Bc4 Bd7 10. O-O-O Rc8 11. Bb3 Ne5 12. Kb1 Re8 13. h4 h5 14. g4
hxg4 15. h5 Nxh5 16. Bh6 Kh7 17. Bxg7 Kxg7 18. fxg4 Bxg4 19. Nf5+ Bxf5 20.
exf5 Rh8 21. fxg6 Nxg6 22. Qd5 e6 23. Qd4+ Qf6 24. Qxa7 Ng3 25. Rxh8 Rxh8
26. Qxb7 d5 27. a4 Qf4 28. Qb6 Nf5 29. Nb5 Qe3 30. Qc7 Qe5 31. Qd7 Qe2 32.
Rc1 Kf6 33. Qb7 Qe5 34. Ka2 Qf4 35. Rd1 Nge7 36. Qa6 Ne3 37. Ra1 N7f5 38.
Rc1 Qb4 39. Qc6 Qd2 40. Ra1 Rh2 41. Na3 Rh8 42. Qb6 Qd4 43. Qc6 Qd2 44. Qb6
[/pgn]

Who exactly decides a draw on the above final position when it is still unclear because _NO_ 3-fold repetition was announced yet.
Most engines have tendencies to repeat up to twice, but would avoid a 3-fold.

The above endgame still looks interesting, and I was so upset when the game was not finished.
This is one of those drawn games at TCEC where the position is still unclear but was automatically considered a draw when SF shows 0.00
We all know there are so many 0.00 positions evald by A/B but where actually won by NN-MTCS engines .

You can also check all TCEC drawn games between these two top engines and you'd noticed there are still some advantages in most games but was stopped and
considered a draw because some pairing (or tourney?) algo already called it a draw. How irritating.

Code: Select all

11. Game adjudication

  1.  A game can be drawn by the normal 3-fold repetition rule or the 50-move rule.
  2.  However, a game can also be drawn at move 35 or later if the eval from both playing engines are within +0.15 to -0.15 pawns for the last 5 moves, or 10 plies.
      If there is a pawn advance, or a capture of any kind, this special draw rule will reset and start over.
      On the GUI interface, this rule is shown as "TCEC draw rule" with a number indicating how many plies there are left until it becomes official.
  3.  Cutechess will automatically adjudicate endgame positions with 6-men Syzygy tablebases.
  4.  For the record, let it be mentioned that from this TCEC Season onwards, the so-called winrule of former Seasons has been dropped.

It would be more visible if you had copied the real pgn, with evals... it seems rule 2 was triggered
(BTW you also removed the termination tag which says why it was ended)

Code: Select all

[Termination "adjudication"]

[Guenther]

Converted the real tcec pgn output to optimized pgn.
As expected rule 2 was triggered by the last 10 plies fullfilling the instructions.

[pgn][Event "TCEC Season 20 - Superfinal"]
[Site "https://tcec-chess.com"]
[Date "2021.01.31"]
[Round "94.1"]
[White "LCZero 0.27.0d-Tilps-dje-magic_JH.94-100"]
[Black "Stockfish 20210113"]
[Result "1/2-1/2"]
[BlackElo "3601"]
[ECO "B78"]
[GameDuration "03:02:04"]
[GameEndTime "2021-01-31T12:08:56.167 UTC"]
[GameStartTime "2021-01-31T09:06:51.868 UTC"]
[Opening "Sicilian"]
[PlyCount "87"]
[Termination "adjudication"]
[TerminationDetails "TCEC draw rule"]
[TimeControl "7200+10"]
[Variation "dragon, Yugoslav attack, 10.O-O-O"]
[WhiteElo "3586"]

1. e4 {book} c5 {book}
2. Nf3 {book} d6 {book}
3. d4 {book} cxd4 {book}
4. Nxd4 {book} Nf6 {book}
5. Nc3 {book} g6 {book}
6. Be3 {book} Bg7 {book}
7. f3 {book} O-O {book}
8. Qd2 {book} Nc6 {book}
9. Bc4 {book} Bd7 {book}
10. O-O-O {book} Rc8 {book}
11. Bb3 {book} Ne5 {book}
12. Kb1 {book} Re8 {0.99/58 6675.498}
13. h4 {0.38/35 7124.325} h5 {0.68/59 6582.686}
14. g4 {0.34/40 7028.994} hxg4 {0.75/57 6506.807}
15. h5 {0.33/40 6940.936} Nxh5 {0.51/61 6421.806}
16. Bh6 {0.33/39 6866.787} Kh7 {0.50/59 6337.068}
17. Bxg7 {0.32/38 6775.293} Kxg7 {0.45/60 6185.918}
18. fxg4 {0.32/37 6691.044} Bxg4 {0.45/68 6063.388}
19. Nf5+ {0.32/36 6607.917} Bxf5 {0.45/60 5984.835}
20. exf5 {0.32/35 6527.051} Rh8 {0.45/72 5904.109}
21. fxg6 {0.32/32 6535.227} Nxg6 {0.45/68 5437.323}
22. Qd5 {0.32/31 6542.755} e6 {0.37/64 5364.431}
23. Qd4+ {0.32/32 6158.272} Qf6 {0.37/66 5268.885}
24. Qxa7 {0.32/32 6021.875} Ng3 {0.26/68 4424.295}
25. Rxh8 {0.32/29 6029.255} Rxh8 {0.26/75 4256.147}
26. Qxb7 {0.32/30 6036.934} d5 {0.26/76 4098.394}
27. a4 {0.31/28 5770.454} Qf4 {0.26/78 4022.417}
28. Qb6 {0.30/26 4954.218} Nf5 {0.26/77 3941.971}
29. Nb5 {0.30/27 4961.806} Qe3 {0.26/75 3824.863}
30. Qc7 {0.30/26 4956.465} Qe5 {0.26/76 3748.570}
31. Qd7 {0.30/22 4272.374} Qe2 {0.26/74 3685.446}
32. Rc1 {0.23/21 3928.285} Kf6 {0.26/74 3627.259}
33. Qb7 {0.27/19 3684.739} Qe5 {0.26/64 3550.972}
34. Ka2 {0.29/17 3425.692} Qf4 {0.26/64 3441.870}
35. Rd1 {0.28/18 3281.313} Nge7 {0.10/73 2584.516}
36. Qa6 {0.28/19 3287.420} Ne3 {0.00/71 2545.274}
37. Ra1 {0.27/19 3122.722} N7f5 {0.00/71 2500.434}
38. Rc1 {0.18/19 2851.879} Qb4 {0.00/74 2372.856}
39. Qc6 {0.17/22 2857.997} Qd2 {0.00/70 2346.722}
40. Ra1 {0.13/20 2630.219} Rh2 {0.00/70 2303.417}
41. Na3 {0.13/5 2636.809} Rh8 {0.00/78 2251.604}
42. Qb6 {0.10/21 2332.469} Qd4 {0.00/67 2158.291}
43. Qc6 {0.10/20 2161.712} Qd2 {0.00/75 2117.714}
44. Qb6 {0.14/19 2069.250}
1/2-1/2[/pgn]

Code: Select all

                         Qd2 {0.00/70 2346.722}
40. Ra1 {0.13/20 2630.219} Rh2 {0.00/70 2303.417}
41. Na3 {0.13/5 2636.809} Rh8 {0.00/78 2251.604}
42. Qb6 {0.10/21 2332.469} Qd4 {0.00/67 2158.291}
43. Qc6 {0.10/20 2161.712} Qd2 {0.00/75 2117.714}
44. Qb6 {0.14/19 2069.250}

[MikeGL]

So I got a bit of a free time and browsed latest TCEC games played last Jan 17, 2021
by the top 2 contenders at superfinal Lc0 and SF,

[pgn]
[Event "TCEC S20 Superfinal"]
[Site "?"]
[Date ""]
[Round "?"]
[White "?"]
[Black "?"]
[Result "0.5-0.5"]
[Termination "unterminated"]
[PlyCount "87"]
[WhiteType "Engine"]
[BlackType "Engine"]

1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 g6 6. Be3 Bg7 7. f3 O-O 8.
Qd2 Nc6 9. Bc4 Bd7 10. O-O-O Rc8 11. Bb3 Ne5 12. Kb1 Re8 13. h4 h5 14. g4
hxg4 15. h5 Nxh5 16. Bh6 Kh7 17. Bxg7 Kxg7 18. fxg4 Bxg4 19. Nf5+ Bxf5 20.
exf5 Rh8 21. fxg6 Nxg6 22. Qd5 e6 23. Qd4+ Qf6 24. Qxa7 Ng3 25. Rxh8 Rxh8
26. Qxb7 d5 27. a4 Qf4 28. Qb6 Nf5 29. Nb5 Qe3 30. Qc7 Qe5 31. Qd7 Qe2 32.
Rc1 Kf6 33. Qb7 Qe5 34. Ka2 Qf4 35. Rd1 Nge7 36. Qa6 Ne3 37. Ra1 N7f5 38.
Rc1 Qb4 39. Qc6 Qd2 40. Ra1 Rh2 41. Na3 Rh8 42. Qb6 Qd4 43. Qc6 Qd2 44. Qb6
[/pgn]

Who exactly decides a draw on the above final position when it is still unclear because _NO_ 3-fold repetition was announced yet.
Most engines have tendencies to repeat up to twice, but would avoid a 3-fold.

The above endgame still looks interesting, and I was so upset when the game was not finished.
This is one of those drawn games at TCEC where the position is still unclear but was automatically considered a draw when SF shows 0.00
We all know there are so many 0.00 positions evald by A/B but where actually won by NN-MTCS engines .

You can also check all TCEC drawn games between these two top engines and you'd noticed there are still some advantages in most games but was stopped and
considered a draw because some pairing (or tourney?) algo already called it a draw. How irritating.

Code: Select all

11. Game adjudication

  1.  A game can be drawn by the normal 3-fold repetition rule or the 50-move rule.
  2.  However, a game can also be drawn at move 35 or later if the eval from both playing engines are within +0.15 to -0.15 pawns for the last 5 moves, or 10 plies.
      If there is a pawn advance, or a capture of any kind, this special draw rule will reset and start over.
      On the GUI interface, this rule is shown as "TCEC draw rule" with a number indicating how many plies there are left until it becomes official.
  3.  Cutechess will automatically adjudicate endgame positions with 6-men Syzygy tablebases.
  4.  For the record, let it be mentioned that from this TCEC Season onwards, the so-called winrule of former Seasons has been dropped.

It would be more visible if you had copied the real pgn, with evals... it seems rule 2 was triggered
(BTW you also removed the termination tag which says why it was ended)

Code: Select all

[Termination "adjudication"]

Thanks a lot for the info.
No PGN tags, nor eval from both engines on my above post because I just did a
copy/paste from chessbomb website https://www.chessbomb.com/arena/-/2020-tcec-20
I don't have a login account yet, so I was unable to download the complete PGN details.

My PC browser have some strange properties. It cannot open the official tcec website due to some unknown reason.
Just displays a blank screen, even a vpm won't help. So I think it is a browser issue.

[Guenther]

My PC browser have some strange properties. It cannot open the official tcec website due to some unknown reason.
Just displays a blank screen, even a vpm won't help. So I think it is a browser issue.

Try downloading from the direct link
https://tcec-chess.com/archive/json/TCE ... al.pgn.zip

I can add a cleaned up pgn too as I did once for an other SuFi.

[Uri Blass]

So I got a bit of a free time and browsed latest TCEC games played last Jan 17, 2021
by the top 2 contenders at superfinal Lc0 and SF,

[pgn]
[Event "TCEC S20 Superfinal"]
[Site "?"]
[Date ""]
[Round "?"]
[White "?"]
[Black "?"]
[Result "0.5-0.5"]
[Termination "unterminated"]
[PlyCount "87"]
[WhiteType "Engine"]
[BlackType "Engine"]

1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 g6 6. Be3 Bg7 7. f3 O-O 8.
Qd2 Nc6 9. Bc4 Bd7 10. O-O-O Rc8 11. Bb3 Ne5 12. Kb1 Re8 13. h4 h5 14. g4
hxg4 15. h5 Nxh5 16. Bh6 Kh7 17. Bxg7 Kxg7 18. fxg4 Bxg4 19. Nf5+ Bxf5 20.
exf5 Rh8 21. fxg6 Nxg6 22. Qd5 e6 23. Qd4+ Qf6 24. Qxa7 Ng3 25. Rxh8 Rxh8
26. Qxb7 d5 27. a4 Qf4 28. Qb6 Nf5 29. Nb5 Qe3 30. Qc7 Qe5 31. Qd7 Qe2 32.
Rc1 Kf6 33. Qb7 Qe5 34. Ka2 Qf4 35. Rd1 Nge7 36. Qa6 Ne3 37. Ra1 N7f5 38.
Rc1 Qb4 39. Qc6 Qd2 40. Ra1 Rh2 41. Na3 Rh8 42. Qb6 Qd4 43. Qc6 Qd2 44. Qb6
[/pgn]

Who exactly decides a draw on the above final position when it is still unclear because _NO_ 3-fold repetition was announced yet.
Most engines have tendencies to repeat up to twice, but would avoid a 3-fold.

The above endgame still looks interesting, and I was so upset when the game was not finished.
This is one of those drawn games at TCEC where the position is still unclear but was automatically considered a draw when SF shows 0.00
We all know there are so many 0.00 positions evald by A/B but where actually won by NN-MTCS engines .

You can also check all TCEC drawn games between these two top engines and you'd noticed there are still some advantages in most games but was stopped and
considered a draw because some pairing (or tourney?) algo already called it a draw. How irritating.

The reason it is considered a rule is because of rule number 2.
https://wiki.chessdom.org/TCEC_Season_2 ... judication

"However, a game can also be drawn at move 35 or later if the eval from both playing engines are within +0.15 to -0.15 pawns for the last 5 moves, or 10 plies. If there is a pawn advance, or a capture of any kind, this special draw rule will reset and start over. On the GUI interface, this rule is shown as "TCEC draw rule" with a number indicating how many plies there are left until it becomes official."

Note that I do not like this rule that make the competition unfair and I think that it is better if programmers ask TCEC not to participate in their tournaments unless they make the tournaments fair.

Using the evaluation of the engines that play to decide if to adjudicate the game is clearly unfair and I do not think that 0.15 evaluation means a draw(even 0.00 does not mean a draw because the players can do mistakes).

[Alayan]

The proportion of wrongly adjudicated games (i.e;, games that would have had a decisive result if continued) is very low, especially between the strongest engines.

Even if the position still looks "alive" to the human chess player's eye, if Stockfish and Leela both gives it a very drawish eval, they have no chance of beating each other from it.

Continuing such games is a pointless waste of time, and arguably the TCEC adjudication rules are too lax because the pawn move/capture reset routinely results in dozens of minutes wasted playing out a dead drawn (for these engines) endgame.

[MikeB]

Interesting final position, most humans would continue to play it I believe. For these two engines, it was a draw.