MoKy wrote on 02/18/15 at 23:38:23:

a late reply, but how is this a win (given the 50 move rule)?

after both 77.Kd4 Bg5 78.h7 Bf6+ 79.Kc5 Bxh7 and 77.Kd4 Bf8 78.h7 Bg7+ 79.Kc5 Bxh7 the following site gives 51 moves to conversion, so a draw:

http://chess.jaet.org/endings/(FEN codes are 8/7b/5b2/2K5/5k2/8/3N4/8 w - - 0 1 and 8/6bb/8/2K5/5k2/8/3N4/8 w - - 0 1)

another line is 77.Kd4 Bh7 78.Kd5 Bf8 79.Nc4 Bg8+ 80.Kd4 Bxh6 and again depth to conversion is 51

in the game at move 100 white has two options which both lose in 43 according to Nalimov tablebases: 100.Nf4 and 100.Nf2

but when you look at the chess.jaet.org site, you'll see that Nf4 is one move further away from conversion (i.e. the capture of the knight) than Nf2

so did white look at that site, or did he work all this out by himself, or was he just lucky to have chosen the right move?

edit: I think I have the answer: he was just lucky!

because at move 104 white chose the wrong move, he should have played 104.Ne2, while 104.Nh3 is losing because it's one depth less to conversion

at move 110 white again played a subobtimal move (because of blindly following Nalimov): he played 110.Nh3 with depth to mate 33, but depth to conversion 19, while 110.Ng4 has depth to mate 32, but depth to conversion 20

so at this point black is 2 moves 'ahead' of the optimal path (which has 51 moves between 2 captures, so draw) but black immediately returns the favor by playing 110...Bd7 (DTC 19) instead of 110...Be8 (DTC 18)

and black blunders again at move 113 where he should have played 113...Bg4 or 113...Ke5, but Nalimov says 113...Bd6 or 113...Kd3 so he plays the latter

so now the game is drawn again, and white was lucky to choose the right move at move 114 because 114.Ng1 (which is equally optimal in depth to mate) is two moves closer to conversion

but it gets even more exciting (who would have thought just looking at this?), white plays the losing move 116.Nd2+ while 116.Ng1 should have been played; and again 117.Kg2 is subobtimal (when looking at depth to conversion), 117.Nf1 was the right move

black again immediately returns the favor by playing 117...Bg4 (DTM 26, DTC 12) iso 117...Bb6 (DTM 27, DTC 11), but white also misses 118.Kf2 (one move further from capture), so black is still 2 moves 'ahead'

but now the most horrible blunder of all: 118...Bf3+?? (DTC 13) iso 118...Be6 (DTC 10), which means the (optimal) move count between captures goes from 49 to 52

120...Ke3 (DTM 23, DTC 12) is also subobtimal, 120...Be7 (DTM 24, DTC 11) was better, but white also misses 126.Kf1 (DTM 15, DTC 7) and plays 126.Kg1 (DTM 17, DTC 6)

and finally black plays 128...Be6 (DTM 15, DTC 5) instead of 128...Kf2 (DTM 16, DTC 3)

conclusion: black could have won easily but I admit that if you don't know something like

http://chess.jaet.org/endings/ exists, it looks horribly complicated to work this out on your own