Combination for Positional Advantage; When Does the Computer See It?

Continuing with Fischer - Keres, Zurich 1959, the diagram shows a position where a Keres inaccuracy allowed Fischer to execute a combination for positional advantage. Keres played 20...Nxd4 21.N3xd4 Rxc1, when both Fischer and Kasparov assigned '?' to Black's last move.

Fischer: 'Overlooking White's follow-up. After 21...exd4 22.Rxc8 Qxc8 23.Qxd4 Nc5 White has precisely nothing.' • Kasparov continued the variation and agreed: '24.Bd2 Qe6 25.f3 Qe5, Black would have solved all his problems.'

The problem with 21...Rxc1 is that after 22.Bxc1 exd4, it allows 23.Nh6+! gxh6 24.Qg4+ Kh8 25.Qxd7, when Fischer noted (and Kasparov agreed) that 'Black's extra Pawn is meaningless in view of the mangled Pawn formation.'

Zurich 1959
Keres, Paul

Fischer, Robert
(After 20.Bd3-b1)
[FEN "1qr1rbk1/1b1n1ppp/p1np4/1p2pNB1/3PP3/5N1P/PP3PP1/1BRQR1K1 b - - 0 20"]

I didn't let my chess software run on any position for more than a couple of minutes, but it only evaluates the position as 'slightly better' for White after 25.Qxd7. Starting with 21...Rxc1, which was its preferred move, it evaluated the subsequent positions as 'slightly better' for Black.