After black's 20th move, he has moved his knights 6 times and they stand on squares they could have reached in 2 moves, and in fact after all this hopping they're right back on the same squares they were on after black's 6th move.

SF sees all sorts of interesting things throughout, and so do Najdorf and Bronstein, but I'll focus on the game starting here:

click for larger view

...where both sides are breaking through on "their" side of the board in classic King's Indian fashion.

Bronstein thinks Kotov's 34.Ng4 is best, but Najdorf thinks it's time to pull the queen back for defense with 34.Qc4. SF agrees with Najdorf. After the pretty much forced 34....Bxg4 35.Bxh6 Rfc8 36.Qxb7 Rcb8 37.a6!

click for larger view

Bronstein and Najdorf give the identical drawing variation starting with 37....Rxb7 38.ab Rb8 40.Ra8 Qd6 41.Rc8! Rxc8 42.Rxc8 Bxc8 43.b8/Q. Bronstein says it's what he should have played, Najdorf thinks Bronstein's 37....g5 is correct, because Bronstein thinks he worse and Najdorf thinks he's better! Again the engine sides with Najdorf.

It all comes down to how to evaluate the position after Bronstein's beautiful suggestion (which Najdorf misses) after 37....g5, namely 38.Rcb1!, and if 38....Rxb7 39.ab Rb8 40.Ra8 Qd8 41.Bxg5! White wins, just as Bronstein says. But (unlike Bronstein), the engine thinks Black is slightly-better-than-OK after 38....Re8 39.a7, and, instead of Bronstein's 39....Kh7 (which gets Black into deep trouble after 40.Bxg5!! Qxg5 41.Qxf7+ Qg7 42.Qxg7+ Kxg7 43.Rb8+ Kg6 44.Rab1 and the pawn on a7 is worth far more than the piece), the engine finds 39....Qxh6! 40.Qxa8! Rxa8 41.Rb8+ Qf8 42.Rxf8+ Kxf8 43.f3 Bc8 and Black is a little better, though a draw is likely.

After Kotov's 38.Qxb8+?, however, White is losing. But the game is still very hard to win. A key moment comes after 43.Rbb7.

click for larger view

Najdorf thinks 43....Kg6 is a mistake, Bronstein doesn't mention it, but they both miss the engine's 43....Qg6!!, which wins after 44.f3 Bxf3! 45.Nxf3 Qxe4 46.Rxf7 Qe3+ 47.Kf1 Nxg3+ 48.Kg2 Nh5. I think this is unfair and no human can be expected to see this, even after an adjournment. But a bigger miss by Bronstein, which Najdorf points out, comes on move 45: ...Nxg3! 46.Rxc8 Qf4! with all sorts of forks and mates in the offing.

Apart from noting that 47.Rxf7 can be met by ...Bd7, Najdorf gives the rest of the game without comment, but as Bronstein was aware, there was a lot going on, and he missed several chances to win. But he also missed some of White's chances to draw, because the engine finds some -- again, very unfair -- fortresses. I'll mention just one.

Bronstein quite naturally thinks that Black is winning after 55....Kg5 56.Rcg7+ Kf4 57.Ng2+ Kxe4 58.Rxh6 Qf3+, but the engine coolly spits out 59.Kg1 g3 60.Rxg3!! Qxg3 61.Rxd6 Qc3 62.Rg6 Kxd5 63.Kh2 e4 64.Rg3

click for larger view

...and that pawn ain't never getting anywhere.