To tablebase or Not to tablebase

[AdminX]

The following position is from a game I played on Playchess versus a player named Karpovvv.

[d]8/2rk4/R7/7p/P2K4/5bP1/8/8 b - - 0 59

Rybka 3 on my quad likes 59. ... Kc8 for black. As black in the game I played Rc4+ after 54 seconds of thinking time, because the win looked easier to me. Now I suppose both moves are winning, but isn't Rc4 more direct when trying to finish the game? Or can this be a case of tablebases clouding the win. I noticed that Stockfish which does not use tablebases see's it right away.


Analysis by Rybka 3:

59...Be2 60.Rh6 Kc8 61.a5
-/+ (-1.39) Depth: 6 00:00:00 1kN
-+ (-1.41) Depth: 7 00:00:00 1kN

59...Kc8
-+ (-1.42) Depth: 7 00:00:00 2kN
-+ (-3.33 !) Depth: 26 00:11:49 194mN, tb=38560


Analysis by Stockfish 1.3 JA:

59...Rc1 60.a5 Rd1+ 61.Ke3 Bc6 62.Rb6 Rg1 63.Kf4 Kd6 64.Rb8 Rf1+ 65.Ke3
-+ (-2.94) Depth: 6 00:00:01 8kN
-+ (-2.90) Depth: 9 00:00:02 67kN

59...Be2 60.Rh6 Rc4+ 61.Ke3 Bg4 62.a5 Rc6 63.Rh7+ Kc8 64.Rg7 Rc3+ 65.Kd4 Ra3 66.Ra7 Kb8 67.Ra6 Rxg3 68.Rb6+ Kc7 69.Rg6 h4 70.a6 Bd7 71.Rh6 h3 72.Ke5 Re3+ 73.Kf4
-+ (-3.11) Depth: 9 00:00:02 94kN
-+ (-4.07) Depth: 20 00:00:16 31006kN

59...Rc4+ 60.Ke3 Bg4 61.Kd2 Kc7 62.Kd3 Bd1 63.a5 Ra4 64.Kc3 Kb7 65.Rb6+ Ka7 66.Rd6 Bg4 67.Rd3 Ka6 68.Kb3 Rxa5 69.Kb4 Kb6 70.Rd6+ Kc7 71.Rd3 Re5 72.Kc4 Be6+ 73.Kd4 Rd5+ 74.Ke3 Rxd3+ 75.Kxd3 Kd6
-+ (-4.43) Depth: 20 00:00:27 79224kN
-+ (-4.64) Depth: 26 00:02:20 573mN

[Mike S.]

I think the reason is that engines don't expect 60.Kxc4, and so, the difference to other continuations is smaller.

Bright 0.4a with 3+4+5 piece bitbases:

Analysis by bright-0.4a:

59...Bc6
-+ (-2.66) Depth: 1/1 00:00:00
(...)
59...Rb7 60.Kc5 Rb3 61.a5 Bg4 62.Rh6 Rxg3 63.Rh7+ Ke6 64.a6 Ra3 65.a7
-+ (-2.35) Depth: 11/23 00:00:01 1199kN
(...)
59...Be2 60.Rh6 Rc4+ 61.Ke3 Bg4 62.a5 Rc3+ 63.Kf2 Rf3+ 64.Kg2 Ra3 65.a6 Ra2+ 66.Kf1 Kc7 67.Ke1 Bf3 68.Rf6 Be4 69.Re6
-+ (-3.00) Depth: 16/35 00:00:22 25473kN
59...Rc4+ 60.Ke3 Rc3+ 61.Kd4 Ra3 62.Kc5 Be2 63.Ra7+ Kc8 64.Kc6 Rc3+ 65.Kb6 Rxg3 66.Rh7 Rg6+ 67.Kc5 Rg5+ 68.Kc6 Bf3+ 69.Kb6 Be4 70.Rh8+ Kd7 71.a5
-+ (-3.16) Depth: 16/36 00:00:43 67682kN
59...Rc4+ 60.Ke3 Rc3+ 61.Kd4 Ra3 62.Kc5 Be2 63.Ra7+ Kc8 64.Kc6 Rc3+ 65.Kb6 Rxg3 66.Rh7 Rg6+ 67.Kc5 Rg5+ 68.Kc6 Bf3+ 69.Kb6 Be4 70.Rh8+ Kd7 71.a5 Kd6
-+ (-3.24) Depth: 17/36 00:00:51 83463kN
59...Rc4+ 60.Ke3 Rc3+ 61.Kd4 Ra3 62.Kc5 Be2 63.Ra7+ Kc8 64.Kc6 Bf3+ 65.Kd6 Rd3+ 66.Ke7 Bg4 67.a5 Kb8 68.Ra6 Rxg3 69.Rh6 Ra3 70.a6 Ka7
-+ (-3.35) Depth: 18/38 00:01:23 143mN

I don't use many 5-piece Nalimovs but all R+? vs. R. With frequent tbs. access, R232a prefers Rc4+ at ply 13, but at ply 17 Be2... There seem to be a couple of moves within a narrow evaluation margin.

Analysis by Rybka 2.3.2a 32-bit, after depth 19:

1. -+ (-2.91): 59...Be2 60.Ra5 Rc4+ 61.Ke3 Bg4 62.Kd3 Rc1 63.Kd4 Rg1 64.Kd5 Kc7 65.Kc5 Rxg3 66.Ra7+
2. -+ (-2.88): 59...Rc4+ 60.Ke3 Rc3+ 61.Kd4 Ra3 62.Kc5 Be2 63.Ra7+ Kc8 64.Re7 Bd1 65.Re8+ Kb7 66.Re7+
3. -+ (-2.88): 59...Rc8 60.Ke3 Rc3+ 61.Kd4 Ra3 62.Kc5 Be2 63.Ra7+ Kc8 64.Re7 Bd1 65.Re8+ Kb7 66.Re7+