Originally posted by Stansi
The position was a draw after 79.Rb7 (but Black needed to reply Nd3+). It was still a win after 70...Nf8. Between these moves it's unclear, as FinalGen isn't able to solve these positions fully (just says White wins or Draw). Maybe someone with access to the Lomonosov tablebases can tell you the status of these positions.
Here's an optimal line from the di ...[text shortened]... f8 75.Rxh6 and we're down to six pieces, so you can use the Nalimov tablebases from here onward.
Yes, after making the move Rb7 I did have the feeling of "maybe I shouldn't have done that", which is never a pleasant sensation. Looking at Crafty's annotation of the game the assessment drops from +1.85 to +0.25 with that move. It also does at 78. Re7, apart from that the assessment bounces between +1.85 and +1.6 so I think that based on the evidence I drifted into a draw twice, but otherwise it was won.
Judging by what it says on their website, FinalGen's really for pawn endings, it can only cope with 1 piece per side, so if there is no winning line not involving a pawn promotion leaving two pieces for white (say) it cannot calculate the position. It also, quite sensibly, has some internal checks that prevent it from giving a complete solution if the position is too complex. It'll calculate distance to conversion rather than distance to mate. But since the whole sequence takes less than 50 moves, 15 until the pawn is taken and another 25 to checkmate, it's not really an issue.
The line you've given from the position at move 61 is:
and according to the 6 man tablebases on the Shredder site it's a win in 25 more moves. The correct plan looks to be to force the black king away from the h-pawn using the rook along the g-file forcing it to choose between defending the f-pawn and then push the knight around until the h-pawn can be taken. I was fixated on controlling the 7th rank in the game.
Great - thanks very much for the help!