| DrGridlock: <Mateo> corrected version with the 35 f6 line does give a better computer eval for White, but I'm not sure the line is "winning." Using computers in evaluating endgames can be tricky, because there can arise situations with a material advantage for one side (producing a large computer eval for that side) which simply cannot be converted into a win. From this position:
Erich Eliskases - Paul Keres
click for larger viewPlay could proceed in this way:
Black checks White's king, which marches up to f6 to support the pawn. Black gets the h-pawn, and then places his rook on the 8th rank to prevent the f-pawn from promoting. 35 ... Rd4
36 Kh3 Rd3+
37 Kg4 Rd4+
38 Kg3 Rd5
39 Kf4 Rd4+
40 Ke5 Rxh4
41 f7 Rh8
42 Kf6
Erich Eliskases - Paul Keres
click for larger viewNow black puts his rook on b8, simultaneously supporting his b-pawn, and preventing the f-pawn from advancing. Black queens the b-pawn for which black trades his rook. Black now has a rook and pawn against White's two advanced pawns. 42 ... Rb8
43 Kxg6 b3
44 Ra1 Kd3
45 Rg1 b2
46 Kf5 b1Q
47 Rxb1 Rxb1
48 g6 Rg1
49 Kf6 Kf1+
50 Ke7 Re1
51 Kf8
Erich Eliskases - Paul Keres
click for larger viewEval here still strongly favors white (over +2). Now black advances his c-pawn while White queens his g-pawn. 51 ... c4
52 g7 c3
53 g8Q c2
54 Qg6+ Kd2
Erich Eliskases - Paul Keres
click for larger viewComputer evals of this position are interesting. They'll consistently give White's material advantage (queen and pawn versus rook and pawn) an edge (over +2 for white from Komodo), but there is simply no way for White to convert this to a win. Run a computer on this position, and you'll just find White's queen dancing around delivering checks. There's no way to prevent black from advancing and promoting his c-pawn, and trading his rook for White's f-pawn. | 
Kibitzing
