Stockfish misevaluations:

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Raphexon
Posts: 93
Joined: Sun Mar 17, 2019 11:00 am
Full name: Henk Drost

Re: Stockfish misevaluations:

Post by Raphexon » Mon Apr 22, 2019 5:49 pm

SF dev is still +5.38 at 80/88. :oops:

JohnWoe
Posts: 99
Joined: Sat Mar 02, 2013 10:31 pm

Re: Stockfish misevaluations:

Post by JohnWoe » Mon Apr 22, 2019 5:51 pm

My engine which is just a fast calculator. There's 0 fortress code. It pushes b5 both times. But not immediately. After Bb5 it's a draw.

40/60 game against Fairy:
https://lichess.org/GXQHobRW

40/180 game against Fairy:
https://lichess.org/uMgnTwnq

Analysis.
exclude: none best +tail
dep score nodes time (not shown: tbhits knps seldep)
15 +3,15 351,4M 1:25.60 Kg1
14 +3,15 142,0M 0:35.11 Kg1
13 +3,15 42,7M 0:10.70 Kg1
12 +3,15 16,5M 0:04.26 Kg1
11 +3,15 6,18M 0:01.62 Kg1
10 +3,15 1,86M 0:00.50 Kf1
9 +3,17 844610 0:00.23 Kf1
8 +3,15 308717 0:00.09 Kf1
7 +3,17 78624 0:00.03 Kf1
6 +3,17 34877 0:00.02 Kf1
5 +3,17 8296 0:00.00 Kg1
4 +3,17 3158 0:00.00 Kg1
3 +3,17 1695 0:00.00 Kg1
2 +3,21 368 0:00.00 Kg1
1 +3,19 164 0:00.00 Kg1
0 #

Raphexon
Posts: 93
Joined: Sun Mar 17, 2019 11:00 am
Full name: Henk Drost

Re: Stockfish misevaluations:

Post by Raphexon » Mon Apr 22, 2019 7:01 pm

SF dev has been stuck at 81/88 for an hour now.
I think it's sensing the draw.

Almost stuck for 90 minutes now.
I think the eval may shift to 0 at 82.

Raphexon
Posts: 93
Joined: Sun Mar 17, 2019 11:00 am
Full name: Henk Drost

Re: Stockfish misevaluations:

Post by Raphexon » Mon Apr 22, 2019 8:10 pm

Nevermind still +5.38 at depth 82.

User avatar
Eelco de Groot
Posts: 4159
Joined: Sun Mar 12, 2006 1:40 am
Location: Groningen

Re: Stockfish misevaluations:

Post by Eelco de Groot » Mon Apr 22, 2019 8:22 pm

You could patch the eval of Stockfish a little in this particular position by scaling the stronger side down for having all pawns locked (after giving b5). The 'lantern' formation on the right is a well known 'indirectly' blocked formation. But I think it is only cosmetical. Kaissa gives 3.71. If you scale that down by 50% it would still be 1.85 high.

Code: Select all


4b3/5k2/p5p1/P2p1p1p/1PpPrP1P/2P2QP1/5K2/8 w - -

Engine: Kaissa IV (512 MB)
by T. Romstad, M. Costalba, J. Kiiski, G.

28/52  0:01   +2.76    1.b5 Bxb5 2.Qd1 Re8 3.Qc1 Ke6 4.Ke3 Rc8 
                       5.Qb1 Kd7 6.Kd2 Re8 7.Qc1 Bc6 8.Qa3 Rb8 
                       9.Qc5 Rb2+ 10.Kc1 Rg2 11.Qa7+ Kd6 
                       12.Qxa6 Rxg3 13.Kd2 Rg2+ 14.Ke1 (3.304.617) 2214 

29/54  0:01   +2.92    1.b5 Bxb5 2.Qd1 Re8 3.Qc1 Ke6 4.Ke3 Rb8 
                       5.Kd2 Kd7 6.Qb2 Rg8 7.Qb4 Rc8 8.Qa3 Ke6 
                       9.Qc1 Kd6 10.Qb1 Kd7 11.g4 hxg4 
                       12.h5 Re8 13.hxg6 Ke6 14.Qe1+ (3.398.776) 2218 

30/50  0:01   +4.28    1.b5 Bxb5 2.Qh1 Bc6 3.Qb1 Re7 4.Qb6 Bb7 
                       5.Qa7 Rd7 6.Qb8 Bc6 7.Qc8 Bb5 8.Qa8 Kg7 
                       9.Ke1 Re7+ 10.Kd2 Rd7 11.Qe8 Kh7 
                       12.Kd1 Rb7 13.Qe6 Ba4+ 14.Ke1 (3.538.082) 2222 

31/95  0:02   +2.88    1.b5 Bxb5 2.Qh1 Ke7 3.Qc1 Re6 4.Qa3+ Rd6 
                       5.Qc5 Bd7 6.Qa7 Kf6 7.Qb8 Ke7 8.Qh8 Be6 
                       9.Qg7+ Bf7 10.Qe5+ Be6 11.Ke1 Rd7 
                       12.Qg7+ Bf7 13.Kd2 Rd8 14.Qe5+ (5.845.525) 2235 

32/84  0:02   +2.98    1.b5 Bxb5 2.Qh1 Ke7 3.Qc1 Re6 4.Qa3+ Rd6 
                       5.Qb2 Kd7 6.Qa2 Re6 7.Qa1 Re8 8.Qa3 Rc8 
                       9.Qb4 Ke6 10.Ke3 Kd7 11.Kd2 Ke6 
                       12.Qb1 Re8 13.Qc1 Kd7 14.Qb2 (6.123.939) 2239 

33/47  0:02   +3.23    1.b5 Bxb5 2.Qh1 Ke7 3.Qc1 Re6 4.Qa3+ Rd6 
                       5.Qb2 Rc6 6.Qd2 Kd7 7.Qc1 Rd6 8.Kf3 Rc6 
                       9.Qa3 Rc8 10.Qb4 Ke6 11.Ke3 Kd7 
                       12.Kd2 Ke6 13.Qb1 Re8 14.Qc1 (6.239.126) 2231 

34/51  0:02   +3.05    1.b5 Bxb5 2.Qh1 Ke7 3.Qc1 Re6 4.Qa3+ Rd6 
                       5.Qc5 Bd7 6.Qa7 Re6 7.Qb7 Rd6 8.Qb2 Bb5 
                       9.Qc2 Re6 10.Qa2 Rd6 11.Kf3 Kf7 
                       12.Qb1 Re6 13.Qb2 Re1 14.Qa3 (6.724.439) 2245 

35/31  0:03   +3.15++  1.b5 (6.739.462) 2244 

35/40  0:03   +3.24++  1.b5 (6.746.374) 2245 

35/40  0:03   +3.37++  1.b5 (6.759.836) 2245 

35/61  0:03   +3.55++  1.b5 (7.027.213) 2253 

35/61  0:03   +3.78++  1.b5 (7.098.549) 2254 

35/61  0:03   +3.44    1.b5 Bxb5 2.Qh1 Ke7 3.Qc1 Re6 4.Qa3+ Rd6 
                       5.Ke2 Be8 6.Kd2 Ke6 7.Qc5 Bc6 8.Kc1 Bd7 
                       9.Qa3 Ke7 10.Kc2 Ke6 11.Kb2 Bb5 
                       12.Qc5 Ke7 13.Ka3 Bd7 14.Kb4 (7.148.217) 2252 

36/55  0:03   +3.53++  1.b5 (7.213.738) 2251 

36/55  0:03   +3.62++  1.b5 (7.265.266) 2251 

36/55  0:03   +3.34--  1.b5 Bxb5 (7.462.356) 2255 

36/55  0:03   +3.55++  1.b5 (7.485.887) 2254 

36/74  0:03   +3.16--  1.b5 Bxb5 (8.824.232) 2281 

36/74  0:03   +3.47++  1.b5 (8.841.675) 2281 

36/84  0:04   +3.82++  1.b5 (9.751.143) 2282 

36/84  0:04   +3.68    1.b5 Bxb5 2.Qh1 Bd7 3.Qb1 Re6 4.Qb7 Rd6 
                       5.Qc7 Ke7 6.Qc5 Ba4 7.Ke1 Bd7 8.Qb4 Be6 
                       9.Kd2 Bd7 10.Qb7 Ke6 11.Qc7 Ke7 
                       12.Qa7 Rc6 13.Kc1 Kd8 14.Kb2 (9.829.050) 2280 

37/43  0:04   +3.77++  1.b5 (9.866.512) 2278 

37/57  0:04   +3.58--  1.b5 Bxb5 (10.070.091) 2280 

37/57  0:04   +3.72++  1.b5 (10.123.125) 2278 

37/81  0:04   +3.90++  1.b5 (10.427.327) 2280 

37/82  0:04   +4.12++  1.b5 (11.010.589) 2278 

37/82  0:05   +4.41++  1.b5 (11.500.614) 2276 

37/82  0:05   +3.45--  1.b5 Bxb5 (11.953.633) 2277 

37/83  0:05   +4.10++  1.b5 (12.880.748) 2280 

37/129 0:06   +3.74    1.b5 Bxb5 2.Qh1 Ke7 3.Qc1 Re6 4.Qa3+ Rd6 
                       5.Qb4 Be8 6.Ke3 Bd7 7.Qc5 Be6 8.Qc7+ Rd7 
                       9.Qb6 Rd6 10.Qc5 Bd7 11.Qc7 Rc6 
                       12.Qb8 Rf6 13.Kf3 Rd6 14.Kf2 (14.363.697) 2273 

38/38  0:06   +3.83++  1.b5 (14.504.164) 2273 

38/40  0:06   +3.92++  1.b5 (14.553.054) 2272 

38/41  0:06   +4.06++  1.b5 (14.590.008) 2271 

38/71  0:06   +4.23++  1.b5 (15.023.089) 2273 

38/71  0:06   +3.64--  1.b5 Bxb5 (15.151.748) 2274 

38/71  0:06   +3.35--  1.b5 Bxb5 (15.258.174) 2274 

38/71  0:06   +3.00--  1.b5 Bxb5 (15.444.045) 2274 

38/71  0:06   +2.58--  1.b5 Bxb5 (15.832.350) 2273 

38/71  0:07   +2.52    1.b5 Bxb5 2.Qd1 Kf6 3.Qc2 Be8 4.Qb2 Re6 
                       5.Qa3 Kg7 6.Qc5 Bf7 7.Qc7 Kg8 8.Qd7 Kg7 
                       9.Qd8 Re4 10.Qb8 Re6 11.Kf3 Be8 
                       12.Qc7+ Bf7 13.Qc8 Rd6 14.Qb7 (17.601.876) 2274 

39/15  0:08   +2.61++  1.b5 (18.827.185) 2278 

39/30  0:08   +2.71++  1.b5 (18.859.487) 2278 

39/32  0:08   +2.84++  1.b5 (18.873.853) 2278 

39/32  0:08   +3.02++  1.b5 (18.914.739) 2277 

39/38  0:08   +3.25++  1.b5 (18.964.446) 2277 

39/48  0:08   +3.54++  1.b5 (19.055.007) 2276 

39/59  0:08   +3.90++  1.b5 (19.210.716) 2275 

39/116 0:09   +3.35    1.b5 Bxb5 2.Qh1 Ke8 3.Qc1 Re6 4.Qa3 Rc6 
                       5.Qb4 Kf7 6.Kf3 Ke8 7.Kg2 Kf7 8.Kf2 Ke8 
                       9.Kf3 Kf7 10.Qb1 Rc7 11.Qb2 Rc8 
                       12.Qc1 Re8 13.Kf2 Ke6 14.Qb1 (20.563.448) 2256 

40/21  0:09   +3.45++  1.b5 (20.639.782) 2255 

40/50  0:09   +3.54++  1.b5 (20.836.367) 2256 

40/50  0:09   +3.67++  1.b5 (20.979.930) 2256 

40/50  0:09   +3.85++  1.b5 (21.231.117) 2255 

40/79  0:09   +4.08++  1.b5 (21.626.323) 2253 

40/79  0:10   +4.37++  1.b5 (22.910.695) 2255 

40/79  0:10   +4.32    1.b5 Bxb5 2.Qh1 Ke8 3.Qc1 Re6 4.Qa3 Rc6 
                       5.Qb4 Kf7 6.Kf1 Re6 7.Qc5 Re3 
                       8.Qxd5+ Kg7 9.Kf2 Rxc3 10.Qe5+ Kf7 
                       11.d5 Rc2+ 12.Kf3 Rd2 13.Qe6+ Kg7 
                       14.Qe7+ (23.482.741) 2254 

41/52  0:10   +4.23--  1.b5 Bxb5 (23.616.796) 2254 

41/52  0:10   +4.13--  1.b5 Bxb5 (23.943.385) 2253 

41/52  0:10   +4.00--  1.b5 Bxb5 (24.207.870) 2253 

41/52  0:10   +3.82--  1.b5 Bxb5 (24.666.925) 2255 

41/52  0:10   +3.97++  1.b5 (24.776.597) 2254 

41/52  0:11   +4.27++  1.b5 (25.123.765) 2254 

41/94  0:11   +3.59--  1.b5 Bxb5 (25.346.308) 2254 

41/94  0:11   +4.11++  1.b5 (26.084.948) 2253 

41/95  0:11   +4.38    1.b5 Bxb5 2.Qh1 Kg7 3.Qb1 Re6 4.Qb2 Be8 
                       5.Kg2 Bf7 6.Qb7 Rd6 7.Qc7 Re6 8.Qa7 Rd6 
                       9.Kh3 Kf6 10.Qb8 Ke6 11.Qb7 Be8 
                       12.Qg7 Bf7 13.Qe5+ Kd7 14.Kh2 (26.806.304) 2251 

42/53  0:12   +4.28--  1.b5 Bxb5 (27.581.600) 2254 

42/53  0:12   +4.19--  1.b5 Bxb5 (27.663.400) 2254 

42/53  0:12   +4.06--  1.b5 Bxb5 (27.860.664) 2254 

42/53  0:12   +3.88--  1.b5 Bxb5 (28.181.018) 2257 

42/53  0:12   +4.02++  1.b5 (28.239.486) 2257 

42/60  0:12   +4.32++  1.b5 (29.250.273) 2262 

42/72  0:13   +3.88    1.b5 Bxb5 2.Qh1 Ke8 3.Qc1 Re6 4.Qa1 Kf7 
                       5.Qb2 Rf6 6.Qa3 Rc6 7.Qb4 Ke8 8.Kg1 Kf7 
                       9.Kf1 Ke8 10.Kf2 Kf7 11.Qa3 Ke8 
                       12.Ke3 Kf7 13.Kd2 Ke8 14.Qb4 (31.107.630) 2269 

43/55  0:13   +3.98++  1.b5 (31.254.376) 2269 

43/66  0:15   +3.79    1.b5 Bxb5 2.Qh1 Ke8 3.Qc1 Re6 4.Qb2 Kf7 
                       5.Qb4 Rc6 6.Kg1 Ke8 7.Qa3 Kf7 8.Kf1 Ke8 
                       9.Kg2 Kf7 10.Kg1 Ke8 11.Qb4 Kf7 
                       12.Kf2 Ke8 13.Ke2 Kf7 14.Kd1 (35.425.486) 2312 

44/43  0:15   +3.89++  1.b5 (35.685.872) 2312 

44/43  0:15   +3.98++  1.b5 (35.803.374) 2312 

44/68  0:15   +4.00    1.b5 Bxb5 2.Qh1 Ke8 3.Qc1 Re6 4.Qb2 Kf7 
                       5.Qa3 Rc6 6.Kf3 Ke8 7.Ke2 Re6+ 8.Kd2 Rc6 
                       9.Kc2 Kf7 10.Kc1 Ke8 11.Qb4 Kf7 
                       12.Kd2 Ke8 13.Ke1 Kf7 14.Kf1 (36.269.949) 2316 

45/52  0:16   +3.91--  1.b5 Bxb5 (38.855.678) 2317 

45/52  0:16   +4.00++  1.b5 (38.998.719) 2318 

45/63  0:17   +4.13++  1.b5 (39.713.671) 2320 

45/63  0:17   +4.00    1.b5 Bxb5 2.Qh1 Ke8 3.Qc1 Re6 4.Qb2 Kf7 
                       5.Qa3 Rc6 6.Qb4 Ke8 7.Kg1 Kf7 8.Kg2 Ke8 
                       9.Kf2 Kf7 10.Qa3 Ke8 11.Ke3 Re6+ 
                       12.Kd2 Rc6 13.Kc2 Kf7 14.Kc1 (39.771.653) 2319 

46/23  0:18   +3.91--  1.b5 Bxb5 (42.763.930) 2323 

46/53  0:19   +3.82--  1.b5 Bxb5 (46.463.159) 2323 

46/53  0:20   +3.91++  1.b5 (46.742.654) 2323 

46/53  0:20   +4.09++  1.b5 (46.906.809) 2323 

46/72  0:20   +3.91    1.b5 Bxb5 2.Qh1 Ke8 3.Qb1 Re6 4.Qb4 Rc6 
                       5.Kf3 Kf7 6.Qa3 Ke8 7.Kg2 Kf7 8.Kf2 Ke8 
                       9.Kg1 Kf7 10.Kg2 Ke8 11.Kh2 Kf7 
                       12.Qb4 Re6 13.Qc5 Re3 14.Qxd5+ (47.426.647) 2323 

47/27  0:22   +3.82--  1.b5 Bxb5 (52.664.713) 2325 

47/27  0:22   +3.91++  1.b5 (52.887.056) 2326 

47/77  0:23   +3.82    1.b5 Bxb5 2.Qh1 Ke8 3.Qb1 Re6 4.Qb4 Rc6 
                       5.Ke1 Kf7 6.Kd2 Ke8 7.Kc2 Kf7 8.Kc1 Ke8 
                       9.Qa3 Kf7 10.Kd2 Ke8 11.Ke1 Kf7 
                       12.Qb4 Ke8 13.Kf2 Kf7 14.Kg1 (53.653.756) 2328 

48/44  0:23   +3.91++  1.b5 (54.025.633) 2328 

48/62  0:23   +3.82    1.b5 Bxb5 2.Qh1 Ke8 3.Qb1 Re6 4.Qb4 Rc6 
                       5.Ke1 Kf7 6.Kd2 Ke8 7.Kc2 Kf7 8.Kc1 Ke8 
                       9.Qa3 Kf7 10.Kb2 Ke8 11.Kc2 Kf7 
                       12.Kc1 Ke8 13.Kd1 Kf7 14.Ke2 (54.953.994) 2331 

49/44  0:23   +3.91++  1.b5 (55.688.181) 2331 

49/72  0:26   +3.84    1.b5 Bxb5 2.Qh1 Ke8 3.Qb1 Re6 4.Qb4 Rc6 
                       5.Ke1 Re6+ 6.Kd1 Rc6 7.Kd2 Kf7 8.Ke2 Re6+ 
                       9.Kf2 Rc6 10.Kg2 Ke8 11.Qa3 Kf7 
                       12.Kh3 Ke8 13.Kh2 Kf7 14.Kg2 (61.216.766) 2343 

50/76  0:28   +3.75--  1.b5 Bxb5 (67.484.496) 2347 

50/76  0:28   +3.84++  1.b5 (67.952.091) 2347 

50/76  0:30   +3.82    1.b5 Bxb5 2.Qh1 Ke8 3.Qb1 Re6 4.Qb4 Rc6 
                       5.Ke1 Re6+ 6.Kd2 Rc6 7.Kc2 Kf7 8.Kc1 Ke8 
                       9.Kd2 Kf7 10.Ke2 Re6+ 11.Kf2 Rc6 
                       12.Qa3 Ke8 13.Kf3 Kf7 14.Ke2 (72.007.171) 2350 

51/86  0:35   +3.82    1.b5 Bxb5 2.Qh1 Ke8 3.Qb1 Re6 4.Qb4 Rc6 
                       5.Ke1 Re6+ 6.Kd2 Rc6 7.Kc2 Kf7 8.Kc1 Ke8 
                       9.Kd2 Kf7 10.Ke2 Re6+ 11.Kf1 Rc6 
                       12.Qa3 Ke8 13.Ke2 Kf7 14.Kf2 (80.678.486) 2304 

52/86  0:39   +3.82    1.b5 Bxb5 2.Qh1 Ke8 3.Qb1 Re6 4.Qb4 Rc6 
                       5.Ke1 Re6+ 6.Kf1 Rc6 7.Qa3 Kf7 8.Ke2 Re6+ 
                       9.Kf3 Rc6 10.Ke3 Ke8 11.Kd2 Kf7 
                       12.Kc1 Ke8 13.Kd1 Kf7 14.Kd2 (89.668.553) 2261 
.
.
.
88/124 27:41  +3.71    1.b5 Bxb5 2.Qh1 Ke8 3.Qb1 Re6 4.Kf1 Kf7 
                       5.Qb4 Rc6 6.Qa3 Ke8 7.Kf2 Kf7 8.Kg1 Ke8 
                       9.Qb4 Kf7 10.Kh2 Ke8 11.Qb1 Kf7 
                       12.Qe1 Re6 13.Qd2 Bc6 14.Qb2 (3.933.697.350) 2367 

89/116 32:25  +3.71    1.b5 Bxb5 2.Qh1 Ke8 3.Qb1 Re6 4.Kf1 Kf7 
                       5.Qb4 Rc6 6.Qa3 Ke8 7.Kf2 Kf7 8.Kg1 Ke8 
                       9.Qb4 Kf7 10.Kh2 Ke8 11.Qb1 Kf7 
                       12.Qe1 Re6 13.Qd2 Bc6 14.Qb2 (4.595.926.034) 2361 

90/129 36:35  +3.71    1.b5 Bxb5 2.Qh1 Ke8 3.Qb1 Re6 4.Kf1 Kf7 
                       5.Qb4 Rc6 6.Qa3 Ke8 7.Kf2 Kf7 8.Kg1 Ke8 
                       9.Qb4 Kf7 10.Kh2 Ke8 11.Qb1 Kf7 
                       12.Qe1 Re6 13.Qd2 Bc6 14.Qb2 (5.172.651.894) 2356 

91/98  43:46  +3.80++  1.b5 (6.186.645.090) 2355 

91/106 44:18  +3.71    1.b5 Bxb5 2.Qh1 Ke8 3.Qb1 Re6 4.Kf1 Kf7 
                       5.Qb4 Rc6 6.Qa3 Ke8 7.Kf2 Kf7 8.Kg2 Ke8 
                       9.Kh3 Kf7 10.Qa1 Kf6 11.Kg2 Ke7 
                       12.Qe1+ Re6 13.Qb1 Ke8 14.Qb4 (6.260.603.893) 2354 

92/116 45:27  +3.71    1.b5 Bxb5 2.Qh1 Ke8 3.Qb1 Re6 4.Kf1 Kf7 
                       5.Qb4 Rc6 6.Qa3 Ke8 7.Kf2 Kf7 8.Kg2 Ke8 
                       9.Kh3 Kf7 10.Qa1 Kf6 11.Kg2 Ke7 
                       12.Qe1+ Re6 13.Qb1 Ke8 14.Qb4 (6.421.145.686) 2354 

93/117 47:31  +3.71    1.b5 Bxb5 2.Qh1 Ke8 3.Qb1 Re6 4.Kf1 Kf7 
                       5.Qb4 Rc6 6.Qa3 Ke8 7.Kf2 Kf7 8.Kg2 Ke8 
                       9.Kh3 Kf7 10.Qa1 Kf6 11.Kg2 Ke7 
                       12.Qe1+ Re6 13.Qb1 Ke8 14.Qb4 (6.715.009.002) 2354 

94/118 50:46  +3.71    1.b5 Bxb5 2.Qh1 Ke8 3.Qb1 Re6 4.Kf1 Kf7 
                       5.Qb4 Rc6 6.Qa3 Ke8 7.Kf2 Kf7 8.Kg2 Ke8 
                       9.Kh3 Kf7 10.Qa1 Kf6 11.Kg2 Ke7 
                       12.Qe1+ Re6 13.Qb1 Ke8 14.Qb4 (7.181.179.434) 2357 

95/109 58:22  +3.71    1.b5 Bxb5 2.Qh1 Ke8 3.Qb1 Re6 4.Kf1 Kf7 
                       5.Qb4 Rc6 6.Qa3 Ke8 7.Kf2 Kf7 8.Kg2 Ke8 
                       9.Kh3 Kf7 10.Qa1 Kf6 11.Kg2 Ke7 
                       12.Qe1+ Re6 13.Qb1 Ke8 14.Qb4 (8.271.545.116) 2361 

96/127 65:05  +3.71    1.b5 Bxb5 2.Qh1 Ke8 3.Qb1 Re6 4.Kf1 Kf7 
                       5.Qb4 Rc6 6.Qa3 Ke8 7.Kf2 Kf7 8.Kg2 Ke8 
                       9.Kh3 Kf7 10.Qa1 Kf6 11.Kg2 Ke7 
                       12.Qe1+ Re6 13.Qb1 Ke8 14.Qb4 (9.248.081.923) 2367 
Debugging is twice as hard as writing the code in the first
place. Therefore, if you write the code as cleverly as possible, you
are, by definition, not smart enough to debug it.
-- Brian W. Kernighan

Raphexon
Posts: 93
Joined: Sun Mar 17, 2019 11:00 am
Full name: Henk Drost

Re: Stockfish misevaluations:

Post by Raphexon » Tue Apr 23, 2019 5:52 am

Woke up and saw that even at depth 87 Stockfish Dev still thinks it's a winning position.
At work now so it will run another 8 hours before I can check again.

User avatar
hgm
Posts: 23718
Joined: Fri Mar 10, 2006 9:06 am
Location: Amsterdam
Full name: H G Muller
Contact:

Re: Stockfish misevaluations:

Post by hgm » Tue Apr 23, 2019 7:44 am

Eelco de Groot wrote:
Mon Apr 22, 2019 8:22 pm
You could patch the eval of Stockfish a little in this particular position by scaling the stronger side down for having all pawns locked (after giving b5). The 'lantern' formation on the right is a well known 'indirectly' blocked formation.
You could also put the eval at 0 (or scale it down by a factor 8 or 16) for all positions with this Pawn structure. Or whenever the ply counter gets above 10.

This could be a good method to quickly detect no-progress situations: start analyzing with a 10-move rule, to quickly force searching of positions after a Pawn push and conclude that they are no good, and then, as the depth increases, slowly relax the ply limit, to allow it to hunt for opportunities to advance with better preparation to positions after the sac that are not yet in the hash table with bad scores. In other words, iterating the reversible ply limit might give better results than iterating the total depth.

peter
Posts: 1782
Joined: Sat Feb 16, 2008 6:38 am
Full name: Peter Martan

Re: Stockfish misevaluations:

Post by peter » Tue Apr 23, 2019 8:20 am

Hi Eelco!
Eelco de Groot wrote:
Mon Apr 22, 2019 8:22 pm
Kaissa gives 3.71. If you scale that down by 50% it would still be 1.85 high.
Junior at least gives a not-winning eval:

4b3/5k2/p5p1/P2p1p1p/1PpPrP1P/2P2QP1/5K2/8 w - - 0 1

Analysis by Deep Junior 13.3:

Code: Select all

8.Qh1 Bb5 9.Qg2 Ke6 10.Qf1 Kd6 
  +-  (2.12)   Depth: 9   00:00:00  4kN
8.Qh1 Bb5 9.Qg2 Ke6 10.Qf1 Kd6 11.Kf3 Re8 12.Qg2 
  +-  (1.80)   Depth: 12   00:00:00  57kN
8.Qd1 Bb5 9.Qh1 Ke7 10.Kf3 Kd6 11.Qd1 Re8 
  +-  (1.81)   Depth: 12   00:00:00  104kN
...
8.Qd1 Bb5 9.Qh1 Ke7 10.Kf3 Kd7 11.Qg1 Re7 12.Qf1 Kd6 13.Kf2 Re8 14.Qg2 Re7 15.Qh1 Re6 16.Qh2 Re4 17.Qh3 
  +/-  (1.04)   Depth: 24   00:00:01  14928kN
8.b5 Bxb5 9.Qh1 Ke8 10.Qa1 Re6 11.Qc1 Kf7 12.Qh1 Re4 13.Qa1 Re6 14.Qb1 Ke8 15.Qb2 Kf7 
  +/-  (1.06)   Depth: 24   00:00:02  28925kN
...
8.b5 Bxb5 9.Qh1 Ke8 10.Qa1 Re6 11.Qb2 Kf7 12.Qb1 Ke8 13.Qa1 Kf7 14.Qc1 Ba4 
  +/-  (1.02)   Depth: 31   00:00:28  651MN
8.Qh1 Bb5 9.Qd1 Ke7 10.Qf3 Kd6 11.Qg2 Re6 12.Qh2 Kd7 13.Qg1 
  +/-  (1.03)   Depth: 31   00:01:01  1475MN
...
8.Qh1 Bb5 9.Qd1 Ke7 10.Qf3 Kd6 11.Qg2 Re6 12.Qh2 Re4 13.Qh1 Ke6 14.Kf3 Ke7 15.Qg1 Re6 
  +/-  (1.03)   Depth: 34   00:06:50  10950MN
24 threads of 12x3GHz, 4G Hash, no tbs.

As for Houdini 6.03, lowering 50 moves boundary to 20 gives eval of +/=, boundary 10 of =.
Last edited by peter on Tue Apr 23, 2019 8:43 am, edited 1 time in total.
Peter.

Paloma
Posts: 848
Joined: Thu Dec 25, 2008 8:07 pm

Re: Stockfish misevaluations:

Post by Paloma » Tue Apr 23, 2019 8:32 am

Raphexon wrote:
Tue Apr 23, 2019 5:52 am
Woke up and saw that even at depth 87 Stockfish Dev still thinks it's a winning position.
At work now so it will run another 8 hours before I can check again.
Believe me, it will still at +5.38

User avatar
hgm
Posts: 23718
Joined: Fri Mar 10, 2006 9:06 am
Location: Amsterdam
Full name: H G Muller
Contact:

Re: Stockfish misevaluations:

Post by hgm » Tue Apr 23, 2019 9:41 am

It could be useful to look at the position at the end of the PV, to see what exactly it considers +5.38, and whether this is an obvious mis-evaluation.

Post Reply