CCRL Testing (@Testers)

[RubiChess]

You write "Setup: 10s+10ms Hash 16MB (cutechess-cli) (1xThread both)" for your test.
This is totally different from the time controls used at CCRL which uses time-per-40moves and far longer times than your 10+0.01 even in blitz. Btw. Do you really use 10ms (=+0.01) increment??
I wouldn't be surprised if one engine performs badly on TC x+y (with very small x,y) compared to TC 40/blitz just by some buggy time management code.

Regards, Andreas

[Guenther]

You write "Setup: 10s+10ms Hash 16MB (cutechess-cli) (1xThread both)" for your test.
This is totally different from the time controls used at CCRL which uses time-per-40moves and far longer times than your 10+0.01 even in blitz. Btw. Do you really use 10ms (=+0.01) increment??
I wouldn't be surprised if one engine performs badly on TC x+y (with very small x,y) compared to TC 40/blitz just by some buggy time management code.

Regards, Andreas

I can add that there were at least two patches concerning TM between Xiphos 0.2 and 0.3, resulting in around 50 elo already together,
but with a more 'normal' inc for ultra bullet tested. I also think 10ms inc doesn't make much sense, as it is lower than the default clock
resolution of WIN.

https://github.com/milostatarevic/xipho ... bcf9496bc6
https://github.com/milostatarevic/xipho ... a6011c7bdf

[MikeB]

Hello all,

I am currently doing some testing with my engine, playing against various opponents in the 3000 to 3100 elo range.
It's not the first time I've noticed strange elo gaps in the ccrl list. So I'm curious if this can be explained in a reasonable way.

Here is an example from the Blitz list:

demolito_20181029-cygwin-64-popcnt.exe
CCRL (rating 3017) 1300 games: 376 wins, 438 losses, 486 draws (37.4%), score: 47.6% (+-16).

Xiphos 0.2 SSE by Milos Tatarevic
CCRL (rating 3063): 1967 games: 629 wins, 496 losses, 842 draws (42.8%), score: 53.4% (+-13)

Demolito did much better against my engine, significantly better. I thought it might be because Demolito is very good with my engine. To exclude this I started a small match between Xiphos and Demolito and stopped after a little more than 2000 games.

Setup: 10s+10ms Hash 16MB (cutechess-cli) (1xThread both)
Score of Demolito vs Xiphos2: 945 - 506 - 681 [0.603] 2132
ELO difference: 72.58 +/- 12.29

Gap: 3063-3017 + 72.58 = 118.58

Although the list says that Xiphos is almost 50 Elo stronger, Xiphos is rather 70 Elo weaker.
The difference of this magnitude cannot be explained by the time mode or by the number of games.

At first glance, this looks like a systematic error.
So, can someone give me a plausible explanation ?

Thanks a lot.

Just my $.02 , although for testing certain changes , 10 second games with 10ms increment is fine (what I call call "micro bullet"), results can vary greatly between two unique engines at 10 sec game with 10 ms increments versus a more standard bullet time control of 2 min with 1 second increment. I have personally found older engines typically do not do well at micro bullet tc. Even Komodo performs far worse vs Stockfish with micro bullet tc. I am not dismissing your results, but I find it not that unusual with micro bullet tc results may differ from CCRL, especially in a head to head match. Opening books, exe performance on various CPUs, exe's all can also make a difference. Typical testing requires one to adjust the time for the differences in CPU, I know stockfish testing does that, we used to it for crafty, CCRL does that etc. - simply more evidence that times controls do matter.

[Desperado]

You write "Setup: 10s+10ms Hash 16MB (cutechess-cli) (1xThread both)" for your test.
This is totally different from the time controls used at CCRL which uses time-per-40moves and far longer times than your 10+0.01 even in blitz. Btw. Do you really use 10ms (=+0.01) increment??
I wouldn't be surprised if one engine performs badly on TC x+y (with very small x,y) compared to TC 40/blitz just by some buggy time management code.

Regards, Andreas

Ok, the time controls are very different. My experience never showed a scaling issue more than 40 Elo (30-50 Elo) switching from
hyperfast to bullet games. (And that is already an extreme). If there is a scaling reason it can not be the only reason.

Both engines handle the 10s time control easily and i did not see any time losses.

Well, i am still collecting ideas and i think i can check my observations with two tests (maybe someone else can assist).

1. Playing a bullet match, something like 1m + 1s 2000 games to verify my observations with a time limit closer to CCRL.
2. Playing a gauntlet for both engines with the same engine pool of 8-10 engines always including both engines. (Maybe 2000 games in total per gauntlet)

[Desperado]

You write "Setup: 10s+10ms Hash 16MB (cutechess-cli) (1xThread both)" for your test.
This is totally different from the time controls used at CCRL which uses time-per-40moves and far longer times than your 10+0.01 even in blitz. Btw. Do you really use 10ms (=+0.01) increment??
I wouldn't be surprised if one engine performs badly on TC x+y (with very small x,y) compared to TC 40/blitz just by some buggy time management code.

Regards, Andreas

I can add that there were at least two patches concerning TM between Xiphos 0.2 and 0.3, resulting in around 50 elo already together,
but with a more 'normal' inc for ultra bullet tested. I also think 10ms inc doesn't make much sense, as it is lower than the default clock
resolution of WIN.

https://github.com/milostatarevic/xipho ... bcf9496bc6
https://github.com/milostatarevic/xipho ... a6011c7bdf

The clock resolution is not an issue for Demolito nor for my engine (maybe for Xihpos. At least there aren't time losses and having
a game with 40-80 moves it sums up to half a second easily). It can be handled even on windows.

BTW, with 10ms fixed movetime it is possible to reach depths of 8,9,10 or deeper in endgames today, which would
be enough to test eval (parameter) changes.

[Desperado]

Hello all,

I am currently doing some testing with my engine, playing against various opponents in the 3000 to 3100 elo range.
It's not the first time I've noticed strange elo gaps in the ccrl list. So I'm curious if this can be explained in a reasonable way.

Here is an example from the Blitz list:

demolito_20181029-cygwin-64-popcnt.exe
CCRL (rating 3017) 1300 games: 376 wins, 438 losses, 486 draws (37.4%), score: 47.6% (+-16).

Xiphos 0.2 SSE by Milos Tatarevic
CCRL (rating 3063): 1967 games: 629 wins, 496 losses, 842 draws (42.8%), score: 53.4% (+-13)

Demolito did much better against my engine, significantly better. I thought it might be because Demolito is very good with my engine. To exclude this I started a small match between Xiphos and Demolito and stopped after a little more than 2000 games.

Setup: 10s+10ms Hash 16MB (cutechess-cli) (1xThread both)
Score of Demolito vs Xiphos2: 945 - 506 - 681 [0.603] 2132
ELO difference: 72.58 +/- 12.29

Gap: 3063-3017 + 72.58 = 118.58

Although the list says that Xiphos is almost 50 Elo stronger, Xiphos is rather 70 Elo weaker.
The difference of this magnitude cannot be explained by the time mode or by the number of games.

At first glance, this looks like a systematic error.
So, can someone give me a plausible explanation ?

Thanks a lot.

Just my $.02 , although for testing certain changes , 10 second games with 10ms increment is fine (what I call call "micro bullet"), results can vary greatly between two unique engines at 10 sec game with 10 ms increments versus a more standard bullet time control of 2 min with 1 second increment. I have personally found older engines typically do not do well at micro bullet tc. Even Komodo performs far worse vs Stockfish with micro bullet tc. I am not dismissing your results, but I find it not that unusual with micro bullet tc results may differ from CCRL, especially in a head to head match. Opening books, exe performance on various CPUs, exe's all can also make a difference. Typical testing requires one to adjust the time for the differences in CPU, I know stockfish testing does that, we used to it for crafty, CCRL does that etc. - simply more evidence that times controls do matter.

Hi Mike,

i think times have changed. Imho today 10s + inc is a normal time. My guess is that most engines in the Top 50 test with time controls 8s,10s,20s and take snapshots of longer time controls. Longer time controls are still better but not practical. Scaling issues can examined very well, going
from 4s,8s,...60s games. In principle bullet games 1m+sth is the longest time control i ever used for testing.

That said, i consider hyperfast today below 4s (1s,2s,4s), with increments (0.01s to 0.1s):
I mentioned in the post before, that my engine is able to handle 10ms movetime games (but i don't use it).
I thought of it when testing/tuning eval parameters.

[RubiChess]

Both engines handle the 10s time control easily and i did not see any time losses.

But maybe one engines uses 9.9 seconds in very few moves and than running on increment not losing on time but barely leaving the "pruning at shallow depth" area (if you really used 10ms increment) which will result in blunders.

[Graham Banks]

The CEGT rating lists show a similar rating difference between Xiphos 0.2 and Demolito 2018-10-29 as the CCRL rating lists.

[Desperado]

Both engines handle the 10s time control easily and i did not see any time losses.

But maybe one engines uses 9.9 seconds in very few moves and than running on increment not losing on time but barely leaving the "pruning at shallow depth" area (if you really used 10ms increment) which will result in blunders.

Well, that is possible. I started a match with 60s+100ms right now. The goal is to see a significant change in the match result.
If the binaries are correct, i assume it is the case, we should be far away from 70 Elo advantage of Demolito. I save the games as pgn for further
investigation. Maybe Xiphos 0.2 does not handle the time control 10s very well and there is only a problem with the engine related to time control.
This would be important to know, because the engine is in my test pool.

[Guenther]

Both engines handle the 10s time control easily and i did not see any time losses.

But maybe one engines uses 9.9 seconds in very few moves and than running on increment not losing on time but barely leaving the "pruning at shallow depth" area (if you really used 10ms increment) which will result in blunders.

Well, that is possible. I started a match with 60s+100ms right now. The goal is to see a significant change in the match result.
If the binaries are correct, i assume it is the case, we should be far away from 70 Elo advantage of Demolito. I save the games as pgn for further
investigation. Maybe Xiphos 0.2 does not handle the time control 10s very well and there is only a problem with the engine related to time control.
This would be important to know, because the engine is in my test pool.

I did a very quick test already to see, if there is an anomaly with Xiphos time management in sth like 10+0.01 and actually there was not (see data below - stripped off most except timings), both nearly used the same amount of time most of the time.

(Edit: Xiphos 0.22 is practically the same for this test - it just has an smp and a fen reading fix, both 0.2 and 0.22 tests are cumulated in CCRL)

Code: Select all

[White "Demolito_20181029-64"]
[Black "Xiphos_022-64"]
[TimeControl "10+0.01"]

1. f4 {book} d5 {book}
2. b3 {book} Nf6 {book}
3. Nf3 {book} Bg4 {book}
4. e3 0.45 Nc6 0.41
5. Bb2 0.39 e6 0.39
6. Be2 0.35 Be7 0.37
7. O-O 0.36 O-O 0.36
8. Nc3 0.38 d4 0.34
9. Nxd4 0.31 Nxd4 0.33
10. Bxg4 0.30 Nxc2 0.32
11. Qxc2 0.29 Nxg4 0.31
12. Ne4 0.30 c6 0.29
13. h3 0.29 Nf6 0.28
14. Bxf6 0.28 Bxf6 0.27
15. Nxf6+ 0.25 Qxf6 0.26
16. d4 0.24 Rfd8 0.25
17. Rac1 0.25 Rd5 0.24
18. Rf2 0.28 Rad8 0.23
19. Re2 0.25 Qe7 0.22
20. Kh2 0.22 Qb4 0.21
21. Qc4 0.23 Qxc4 0.21
22. bxc4 0.18 Ra5 0.20
23. c5 0.17 Rb5 0.19
24. g4 0.16 Kf8 0.18
25. Kg3 0.16 Ke7 0.18
26. h4 0.17 a5 0.17
27. h5 0.18 a4 0.16
28. Rd2 0.17 Ra8 0.16
29. h6 0.18 g6 0.16
30. g5 0.16 f5 0.14
31. Kf2 0.12 a3 0.14
32. Ke2 0.11 Rb2 0.14
33. Rxb2 0.12 axb2 0.13
34. Rb1 0.10 Rxa2 0.13
35. Kd3 0.12 Kd7 0.12
36. Kc2 0.098 Kc7 0.12
37. Rxb2 0.093 Rxb2+ 0.11
38. Kxb2 0.089 b5 0.11
39. cxb6+ 0.092 Kxb6 0.10
40. Kb3 0.088 Kb5 0.10
41. Kc3 0.083 Kb6 0.098
42. Kc4 0.097 Kc7 0.094
43. Kd3 0.083 Kb6 0.090
44. Ke2 0.078 Kb5 0.088
45. Kd2 0.071 Kc4 0.084
46. Kc2 0.067 c5 0.080
47. dxc5 0.073 Kxc5 0.078
48. Kd3 0.067 e5 0.076
49. Kc3 0.073 e4 0.074
50. Kb3 0.056 Kb5 0.073
51. Kc3 0.053 Kc5 0.068
52. Kb3 0.050 Kb5 0.066
53. Kc3 0.048 Kc5 0.063
{Draw by 3-fold repetition}
1/2-1/2

[White "Xiphos_022-64"]
[Black "Demolito_20181029-64"]
[TimeControl "10+0.01"]

1. f4 {book} d5 {book}
2. b3 {book} Nf6 {book}
3. Nf3 {book} Bg4 {book}
4. e3 0.41 e6 0.41
5. Be2 0.39 Nbd7 0.37
6. O-O 0.37 Be7 0.35
7. Nc3 0.36 O-O 0.34
8. Bb2 0.35 c6 0.36
9. a4 0.33 Re8 0.36
10. d4 0.32 Bf5 0.35
11. Ne5 0.31 Ne4 0.30
12. Nxe4 0.29 Bxe4 0.28
13. Bd3 0.28 f5 0.29
14. Bxe4 0.27 dxe4 0.26
15. Nxd7 0.26 Qxd7 0.27
16. a5 0.25 Red8 0.28
17. c4 0.24 c5 0.29
18. Qe2 0.23 Rac8 0.28
19. Rfd1 0.22 Qc6 0.25
20. dxc5 0.21 Qxc5 0.22
21. Bd4 0.21 Qb4 0.22
22. Qc2 0.20 Bf6 0.21
23. Bxf6 0.19 gxf6 0.21
24. a6 0.18 bxa6 0.20
25. Rxa6 0.18 Rd3 0.19
26. Ra4 0.17 Qd6 0.17
27. Rxd3 0.16 exd3 0.18
28. Qd2 0.16 Rc7 0.17
29. Ra1 0.15 Rd7 0.16
30. b4 0.15 Qc6 0.17
31. c5 0.14 Kf7 0.16
32. Kf2 0.14 e5 0.16
33. g3 0.13 Kg7 0.17
34. fxe5 0.13 fxe5 0.13
35. Qa2 0.12 Kf8 0.14
36. h4 0.12 a6 0.13
37. Qd2 0.11 a5 0.13
38. bxa5 0.11 Qxc5 0.12
39. Rc1 0.10 Qd6 0.12
40. Qc3 0.10 Kg7 0.11
41. a6 0.096 Qxa6 0.10
42. Qxe5+ 0.093 Qf6 0.093
43. Qb5 0.091 Qe6 0.088
44. Ra1 0.087 d2 0.089
45. Rd1 0.085 Kg6 0.094
46. Ke2 0.081 Rd5 0.083
47. Qc4 0.077 h6 0.083
48. Qc2 0.076 Qd6 0.079
49. Qc4 0.072 Kh5 0.073
50. Kf2 0.069 Qd7 0.068
51. Ke2 0.068 Qe6 0.073
52. Kf2 0.065 Qd6 0.064
53. Qe2+ 0.063 Kg6 0.060
54. Qc4 0.062 Qd7 0.062
55. Ke2 0.058 Rd6 0.059
56. Qc5 0.057 h5 0.060
57. Qc4 0.056 Kh7 0.051
58. Kf2 0.053 Kg7 0.057
59. Qf4 0.053 Kh7 0.046
60. Qg5 0.049 Qf7 0.039
61. Qf4 0.049 Qd7 0.041
62. Qg5 0.046 Qf7 0.034
63. Qf4 0.046 Qd7 0.034
{Draw by 3-fold repetition}
1/2-1/2

[White "Demolito_20181029-64"]
[Black "Xiphos_022-64"]
[TimeControl "10+0.01"]

1. Nf3 {book} g6 {book}
2. g3 {book} c5 {book}
3. d4 {book} cxd4 {book}
4. Qxd4 0.44 Nf6 0.41
5. Nc3 0.40 Nc6 0.39
6. Qa4 0.38 Bg7 0.37
7. Bg2 0.36 O-O 0.36
8. O-O 0.38 d5 0.34
9. Bg5 0.36 h6 0.33
10. Bxf6 0.35 Bxf6 0.32
11. Rad1 0.33 d4 0.31
12. e3 0.35 e5 0.29
13. exd4 0.33 exd4 0.28
14. Ne4 0.32 Bg7 0.27
15. Rfe1 0.30 Qb6 0.26
16. Nd6 0.29 Be6 0.25
17. Nxb7 0.27 d3 0.24
18. Nd6 0.24 dxc2 0.23
19. Qxc2 0.23 Bxa2 0.22
20. Nc4 0.24 Qc5 0.21
21. Rc1 0.23 Rae8 0.21
22. Rxe8 0.20 Rxe8 0.20
23. Bf1 0.19 Rb8 0.19
24. Qd2 0.20 Qf5 0.18
25. Qe3 0.18 Rd8 0.18
26. Qa3 0.21 Bxc4 0.17
27. Bxc4 0.19 Bxb2 0.17
28. Qxb2 0.17 Qxf3 0.16
29. Qb7 0.15 Ne5 0.15
30. Qxf3 0.13 Nxf3+ 0.15
31. Kg2 0.13 Nd2 0.14
32. Ba6 0.12 Rd6 0.14
33. Ra1 0.12 Kg7 0.13
34. Bc8 0.11 Rc6 0.13
35. Bb7 0.12 Rc7 0.12
36. Bf3 0.10 Nb3 0.12
37. Ra6 0.10 Nc5 0.11
38. Ra1 0.11 f5 0.11
39. h4 0.095 Kf6 0.10
40. Bd5 0.097 Ke5 0.10
41. Bc4 0.098 g5 0.096
42. hxg5 0.091 hxg5 0.093
43. Rd1 0.097 g4 0.090
44. Ra1 0.088 Kd4 0.088
45. Ba2 0.085 Kc3 0.085
46. Bg8 0.093 a6 0.081
47. Rc1+ 0.084 Kb2 0.078
48. Rh1 0.076 a5 0.076
49. Bh7 0.060 Rf7 0.073
50. f3 0.074 a4 0.071
51. fxg4 0.065 fxg4 0.067
52. Rb1+ 0.070 Ka2 0.066
53. Bg6 0.061 Rg7 0.063
54. Bf5 0.060 Rg5 0.062
55. Bc2 0.055 Rd5 0.060
56. Kf1 0.065 Nb3 0.057
57. Ke1 0.053 Rc5 0.055
58. Bg6 0.050 Rc6 0.053
59. Bd3 0.050 a3 0.051
60. Kf2 0.044 Rc1 0.051
61. Rxc1 0.044 Nxc1 0.048
62. Bc4+ 0.043 Kb2 0.048
63. Be6 0.040 Nb3 0.047
64. Bxg4 0.043 a2 0.046
65. Be6 0.039 a1=Q 0.042
66. Bxb3 0.035 Kxb3 0.043
67. Kf3 0.034 Qd4 0.040
68. Kg2 0.036 Qd1 0.039
69. Kf2 0.030 Kc2 0.037
70. Ke3 0.035 Qg4 0.038
71. Kf2 Kd2 0.035
72. Kg1 Qe2 0.034
73. Kh1 Ke3 0.033
74. Kg1 Kf3 0.015
75. Kh1 Qg2# 0.005
{Black mates}
0-1

[White "Xiphos_022-64"]
[Black "Demolito_20181029-64"]
[TimeControl "10+0.01"]

1. Nf3 {book} g6 {book}
2. g3 {book} c5 {book}
3. d4 {book} cxd4 {book}
4. c3 0.40 dxc3 0.41
5. Nxc3 0.39 Bg7 0.40
6. e4 0.37 Nf6 0.41
7. Qa4 0.36 O-O 0.39
8. e5 0.34 Ne8 0.35
9. Be3 0.33 d6 0.37
10. O-O-O 0.32 Nc6 0.38
11. Qb3 0.31 Bg4 0.33
12. Bg2 0.29 Qc8 0.32
13. h3 0.28 Be6 0.30
14. Qa4 0.27 dxe5 0.31
15. Ng5 0.26 Bd7 0.28
16. Qh4 0.25 h6 0.29
17. Nge4 0.24 g5 0.25
18. Qh5 0.23 Nd4 0.25
19. Kb1 0.22 Bf5 0.23
20. Rc1 0.21 Qd7 0.25
21. Qd1 0.21 Rc8 0.23
22. g4 0.20 Bg6 0.22
23. h4 0.19 gxh4 0.22
24. Rxh4 0.18 Nd6 0.22
25. Qd3 0.18 Rfd8 0.21
26. Rd1 0.17 Qe6 0.21
27. g5 0.16 N6f5 0.19
28. Bh3 0.16 h5 0.19
29. Bxf5 0.15 Qxf5 0.16
30. a4 0.15 f6 0.15
31. a5 0.14 fxg5 0.14
32. Bxg5 0.14 Nf3 0.13
33. Qxd8+ 0.13 Rxd8 0.13
34. Rxd8+ 0.13 Kf7 0.12
35. Ka1 0.12 Nxg5 0.11
36. Nxg5+ 0.12 Qxg5 0.12
37. Rh1 0.11 h4 0.11
38. Ra8 0.11 Qf4 0.11
39. Rd8 0.10 Bf5 0.10
40. Rdd1 0.10 h3 0.099
41. f3 0.097 h2 0.094
42. Ne4 0.093 Kg6 0.090
43. Rdf1 0.090 Bxe4 0.092
44. fxe4 0.088 Qd2 0.084
45. Rc1 0.083 Bh6 0.078
46. a6 0.082 bxa6 0.090
47. Rc6+ 0.080 Kf7 0.074
48. Ka2 0.077 Qg2 0.068
49. Rxh2 0.074 Qxh2 0.061
50. Rc4 0.071 Be3 0.060
51. Rc3 0.068 Bd4 0.056
52. Rb3 0.066 Ke6 0.054
53. Ka3 0.064 Qe2 0.049
54. Rb8 0.061 Qc4 0.054
55. Rd8 0.060 Bc5+ 0.050
56. b4 0.058 Qxb4+ 0.056
57. Ka2 Qa5+ 0.049
58. Kb3 Qxd8 0.042
59. Kc2 Qd4 0.045
60. Kc1 Qc3+ 0.040
61. Kd1 Qd3+ 0.041
62. Ke1 Qc2 0.037
63. Kf1 Qf2# 0.034
{Black mates}
0-1

Then I did a quick test with an mps time control, you know that CCRL does not use inc, but this was already mentioned. Here there was some big difference especially between moves 35-40 or so, when Demolito used much less time and what I noticed already before, the big depth difference between both might hit on Demolito in some circumstances and with some higher tc? [Just my working theory for now ;-)]

Example:

[Event "Test@CuteCGdv"]
[Site "RWBC-CAPPUCCINO Win7U64 Q8200 2.33Ghz + Nvidia GT 710"]
[Date "2021.03.25"]
[Round "1"]
[White "Xiphos_022-64"]
[Black "Demolito_20181029-64"]
[Result "1-0"]
[ECO "D10"]
[GameDuration "00:01:48"]
[GameEndTime "2021-03-25T13:48:55.711 Mitteleuropäische Zeit"]
[GameStartTime "2021-03-25T13:47:07.558 Mitteleuropäische Zeit"]
[Opening "QGD Slav defense"]
[PlyCount "143"]
[TimeControl "40/30"]

1. d4 {book} d5 {book}
2. c4 {book} c6 {book}
3. Bf4 {book} dxc4 {book}
4. e4 {+0.05/16 1.19} b5 {-0.19/13 1.11}
5. Nc3 {-0.03/18 1.15} h6 {-0.05/13 1.11}
6. a3 {+0.06/17 1.10} Nf6 {+0.05/13 1.07}
7. Be2 {+0.19/17 1.06} Nbd7 {+0.15/13 1.07}
8. Nf3 {+0.16/17 1.01} Bb7 {+0.21/13 1.07}
9. O-O {+0.33/16 0.97} e6 {+0.22/13 1.03}
10. Qc1 {-0.13/16 0.94} Be7 {+0.39/14 0.96}
11. a4 {-0.37/14 0.90} a6 {+0.67/12 1.11}
12. h3 {-0.65/15 0.86} O-O {+0.70/13 0.98}
13. Bh2 {-0.58/17 0.83} Qb6 {+0.54/11 1.15}
14. Rd1 {-0.39/14 0.79} a5 {+0.65/12 1.06}
15. e5 {-0.15/16 0.76} Nd5 {+0.73/13 0.93}
16. Ne4 {-0.29/15 0.73} Rfd8 {+0.80/14 0.86}
17. Nd6 {-0.20/13 0.73} Nf8 {+0.73/14 0.87}
18. Qd2 {-0.06/15 0.73} Nh7 {+0.75/12 0.95}
19. Qc2 {0.00/14 0.73} Nb4 {+0.58/12 0.95}
20. Qc1 {0.00/16 0.73} f5 {+0.66/12 0.95}
21. h4 {+0.09/14 0.73} Nd5 {+0.71/14 0.76}
22. Re1 {-0.10/14 0.73} Nf8 {+0.69/13 0.83}
23. Qd2 {-0.34/15 0.73} Nd7 {+0.72/13 0.83}
24. Bd1 {-0.07/12 0.73} Qa6 {+0.89/12 0.83}
25. Be2 {-0.38/15 0.73} Bxd6 {+0.88/14 0.71}
26. exd6 {-0.31/18 0.73} N7f6 {+0.91/14 0.72}
27. Bf1 {-0.31/17 0.73} Bc8 {+1.09/12 0.77}
28. Ne5 {-0.66/15 0.73} Ne4 {+1.04/13 0.71}
29. Rxe4 {-0.19/16 0.73} fxe4 {+1.34/12 0.77}
30. Qe2 {-0.27/17 0.73} Nf6 {+1.02/11 0.73}
31. Ng4 {-0.25/18 0.73} Nxg4 {+0.82/12 0.67}
32. Qxg4 {-0.44/18 0.73} e5 {+0.87/12 0.67}
33. Qxe4 {-0.16/16 0.74} exd4 {+0.54/13 0.61}
34. axb5 {0.00/16 0.74} cxb5 {+0.50/13 0.52}
35. Qxd4 {0.00/18 0.73} Ra7 {+0.51/12 0.56}
36. Be2 {0.00/18 0.73} a4 {+0.48/11 0.59}
37. Bf3 {+0.21/15 0.74} Bb7 {+0.25/12 0.53}
38. Bg4 {0.00/21 0.74} Bc8 {+0.04/12 0.35}
39. Bf3 {0.00/23 0.74} Bf5 {+0.23/11 0.33}
40. Qe5 {0.00/14 0.67} Bd3 {+0.36/10 0.23}
41. Qc5 {-0.23/15 1.20} Kh7 {+0.29/12 1.11}
42. Be5 {-0.44/17 1.15} Rad7 {+0.16/13 1.03}
43. Rd1 {-0.44/18 1.10} Rf7 {+0.19/12 1.11}
44. Qd5 {0.00/19 1.06} Rdf8 {+0.45/13 0.99}
45. d7 {0.00/21 1.02} Qb6 {-0.10/13 0.99}
46. Qc6 {0.00/20 0.98} Qd8 {-0.10/15 0.83}
47. Rxd3 {+2.38/21 0.94} cxd3 {-2.46/14 0.90}
48. Be4+ {+9.14/26 0.90} Kg8 {-6.53/16 0.79}
49. Qg6 {+10.55/24 0.87} Rf6 {-9.69/16 0.77}
50. Qh7+ {+10.84/25 0.83} Kf7 {-9.82/16 0.77}
51. Bd5+ {+10.94/27 0.80} Ke7 {-9.81/16 0.81}
52. Qxg7+ {+10.91/24 0.77} R6f7 {-10.34/16 0.75}
53. Bf6+ {+11.31/22 0.74} Kd6 {-10.64/16 0.76}
54. Bxf7 {+10.93/23 0.71} d2 {-10.86/16 0.77}
55. Qg3+ {+11.63/21 0.68} Kc6 {-11.10/16 0.74}
56. Qf3+ {+12.11/21 0.65} Kxd7 {-11.26/15 0.79}
57. Bxd8 {+12.60/20 0.65} Rxd8 {-11.62/14 0.85}
58. Qd3+ {+12.71/21 0.65} Kc7 {-12.11/14 0.81}
59. Bd5 {+13.08/18 0.65} Rc8 {-12.26/13 0.88}
60. Qxd2 {+14.61/18 0.65} Kb6 {-12.92/13 0.88}
61. Qxh6+ {+16.79/19 0.65} Ka5 {-14.46/15 0.71}
62. Qd2+ {+17.31/22 0.65} Kb6 {-15.07/14 0.77}
63. h5 {+18.56/20 0.65} Rc7 {-15.60/15 0.73}
64. Bf3 {+19.89/22 0.65} Rc4 {-17.48/14 0.73}
65. h6 {+29.03/19 0.65} a3 {-24.18/15 0.69}
66. bxa3 {+235.69/17 0.65} Kc5 {-27.73/14 0.75}
67. h7 {+M11/17 0.65} Rd4 {-M12/15 0.64}
68. Qe3 {+M7/22 0.65} Kd6 {-M8/15 0.65}
69. Qxd4+ {+M5/41 0.65} Ke7 {-M6/15 0.66}
70. h8=Q {+M3/99 0.49} b4 {-M4/15 0.60}
71. axb4 {+M1/99 0.009} Ke6 {-M2/16 0.58}
72. Qhf6# {0.00/99 0.007}
{White mates}
1-0