Notorious f+h Pawns

"ALL rook (and pawn) endings are drawn" is an aphorism often attributed variously to Tarrasch or Tartakower.
It sounds much more like the latter, but in any event there is a lot of truth in it, sometimes even when one player is two pawns up.
The ending of rook, f-pawn and h-pawn versus rook "has been notorious ever since the famous game Marshall-Rubinstein, San Sebastian 1911," according to Reuben Fine in Basic Chess Endings.
I reached such an ending last month in the second round of the Highlands Open in Trebic, part of the Czech Tour.
Here is the position after White's 51.Rxa3:

Black to play and draw in Spanton (1858) - Svante Norlander (1583)

Because this position can be analysed perfectly thanks to the Nalimov endgame tablebases, I shall use the John Nunn endgame convention to punctuate the moves.
Here is a subbed down version of the convention (a fuller explanation is available on Wikipedia):
!: The only move which maintains the current evaluation of a win or a draw. An ! is used no matter how trivial the move in question, unless it is the only legal move.
!!: A particularly difficult-to-find ! move.
?: A move which negatively affects the evaluation of the position. ie turns a win into a draw or a draw into a loss.
??: An obviously bad ? move.
51...Rd1+
Seven other moves maintain the draw, including 51...Rc1 and 51...Kd7 - evidence that the drawing margin in this ending is wide indeed.
52.Ke3 Rh1 53.Ra4 Ke5 54.h5 Kf6 55.h6 Re1+
No fewer than 10 moves maintain the draw here, but they do not include 55...Rxh6?, which becomes a lost pawn ending after 56.Ra6+ Kg7 57.Rxh6 Kxh6 58.Kf4 (or Ke4 or Kd4, or even Kd3).
56.Kf4 Re8?
Only one rook move draws, 57...Re7, and so does the natural-looking 57...Kg6 and even 57...Kf7.
57.Ra6+
57.Ra7 also wins.
57...Kf7

White to play and find the only move that keeps the position a win

58.Kf5!
That wasn't so difficult. Clearly White's king and f-pawn need to advance.
58...Re1

White again to play and find the only move that keeps the position a win

59.f4?
I needed to find 59.Ra8! The point is that Black's king has no useful move, while 59...Rh1 can be simply met by 60.h7 (60.Kg5 also wins).
59...Rh1 60.Ra7+ Kg8!
A fairly obvious choice, but a sign that the drawing margin has significantly narrowed.
61.Kg6 Rg1+! 62.Kf6 Kh8?
Black had to play 62...Rh1! He may have been worried by the line 63.h7+ Rxh7?? 64.Ra8mate. But 63...Kh8! draws.

White has two winning moves in this position -  I didn't find either

63.Rg7?
White wins with the understandable 63.f5 and the less obvious 63.Re7.
63...Ra1?
63.Rf1 and 63.Rh1 draw.
64.Rg6?
Again, 64.f5 and 64.Re7 win.
64...Ra6+ 65.Kg5 Ra1 66.Re6 Rg1+! 67.Kf6 Kh7 68.Kf7 Rf1 69.Rf6 Rg1
69...Rxf4?? is the only move that loses in this position!
70.f5 Rf1! 71.Kf8

My plan over the last few moves has been to sacrifice the h-pawn in return for reaching a winning Lucena position, but Black has three moves that foil it

71...Rg1?
Black had to keep his rook on the f file. After the text, I get to execute my plan.
72.Ra6 Rf1
Stockfish8 reckons 72...Rg8+ draws, but Nalimov (and Komodo9) show White winning after either legal king move.
73.f6! Kxh6 74.f7+ Kh7 75.Ra4 76.Rh4+ Kg6 77.Kg8 1-0