Note: This is an analysis of the position at move 34. Obviously, White's play in the beginning of the endgame was poor and he should never have entered such trouble.
Black's advantage is clear and long-lived because he has the better minor piece, a knight vs a bad bishop with a locked center. Whether this advantage is enough for a win requires analysis, of course.
Black has a great stronghold on e4, and your reflex to place the knight there seems very natural, but... this is rather the right plan in the middlegame, not in the endgame !
In a minor piece endgame, the best piece to place on a central stronghold is not the knight anymore (as in the middlegame, or in most multi-pieces endgames with rooks), but the king ! As long as it remains safe, the king on the stronghold is centralized, attacks the pawns around it, and often threatens to invade the enemy camp.
This principle is very often applied in isolani endgames. Imagine Black has an isolated pawn on d5. If we still have rooks, White will gladly place his knight on d4, blocking the pawn and controling a bunch of important squares, and try to generate play down the c-file. But what happens if rooks are exchanged and Black has no control of d4 ? The winning plan usually becomes Nd4-e2, Kd3-d4, Ne2-c3, and Pd5 will soon fall. The king, not the knight, makes the decisive threats from d4.
This applies to your game. Just place the bK on e4 and you will see that White is doomed to passive defense of Pe3 and of the entry squares d3 and f3. That's why the right plan on move 34 is to try and gain access to e4 for your king.
34...Kf5 threatens 35...g4+ followed by ...Ke4, so 35.h3 seems forced. However, you then make use of your doubled g-pawn to open the way with 35...g4+ 36.hg4 Ng4 and there are three plans to expel the wK from f3:
- bring your second g-pawn to g4. Then you can reach a position with Ke4, Pg4, Nf5 vs Ke2, Bf2 that is actually a zugzwang with White to play. How to give him the tempo, however, is not obvious. Most probably you will have to go for e6-e5, dxe5, Kxe5, Kd3, and try to gain access to the fourth rank again. Some winning chances, but my feeling is that White will hold if there is no refinement before ; I will stop the analysis here for now.
With an animated diagram:
[Title "jasonhaller-RewanDemontay"]
[FEN "6k1/1pn3p1/p1p1p1p1/2Pp4/PP1P4/4P3/1B4PP/5K2"]
[startply "67"]
1. d4 d5 2. Nf3 Nc6 3. c4 e6 4. Nc3 Bb4 5. Qc2 Nf6 6. e3 O-O 7. a3 Bxc3+ 8. Qxc3 Bd7 9. Bd3 Qe7 10. c5 a6 11. b4 Na7 12. O-O Bb5 13. Re1 Bxd3 14. Qxd3 c6 15. Nd2 Nb5 16. a4 Nc7 17. Bb2 Rfe8 18. Nf3 Rab8 19. Ne5 Nd7 20. f4 Nxe5 21. fxe5 f5 22. exf6 Qxf6 23. Rf1 Qg6 24. Qxg6 hxg6 25. Rae1 Rf8 26. Rxf8+ Rxf8 27. Rf1 Rxf1+ 28. Kxf1 Kf7 29. Kf2 Kf6 30. Kf3 Ne8 31. Kf4 g5+ 32. Kg4 Kg6 33. g3 Nf6+ 34. Kf3 Kf5 35.h3 g4+ 36.hxg4+ Nxg4 37.Bc1 Nf6 38.Bd2 g5 39.Bc3 g4+ 40.Ke2 Ke4 41.Be1 Ng8 42.Bf2 Ne7 43.Bg1 Nf5 44.Bf2 e5 45.dxe5 Kxe5 46.Kd3
- try to check with the knight, keeping Pg7 on place. If you can get a similar position with Ke4, Nf5, you are certainly winning because you have spared yourself tempo-moves with the Pg7. It means that White has to prevent such plan by playing g4 himself when his king still stand on f3, but then the Pg4 himself will become a weakness.
If White waits passively:
[Title "jasonhaller-RewanDemontay"]
[FEN "6k1/1pn3p1/p1p1p1p1/2Pp4/PP1P4/4P3/1B4PP/5K2"]
[startply "67"]
1. d4 d5 2. Nf3 Nc6 3. c4 e6 4. Nc3 Bb4 5. Qc2 Nf6 6. e3 O-O 7. a3 Bxc3+ 8. Qxc3 Bd7 9. Bd3 Qe7 10. c5 a6 11. b4 Na7 12. O-O Bb5 13. Re1 Bxd3 14. Qxd3 c6 15. Nd2 Nb5 16. a4 Nc7 17. Bb2 Rfe8 18. Nf3 Rab8 19. Ne5 Nd7 20. f4 Nxe5 21. fxe5 f5 22. exf6 Qxf6 23. Rf1 Qg6 24. Qxg6 hxg6 25. Rae1 Rf8 26. Rxf8+ Rxf8 27. Rf1 Rxf1+ 28. Kxf1 Kf7 29. Kf2 Kf6 30. Kf3 Ne8 31. Kf4 g5+ 32. Kg4 Kg6 33. g3 Nf6+ 34. Kf3 Kf5 35.h3 g4+ 36.hxg4+ Nxg4 37.Bc1 Nf6 38.Bd2 Nh7 39.Bc3 Ng5+ 40.Ke2 Ke4 41.Be1 Nf7 42.Bf2 Nh6 43.Bg1 Nf5 44.Bf2 g6 45.a5 g5 46.g4 Nh6
then Black wins a pawn and the game.
But if White reacts with g4:
[Title "jasonhaller-RewanDemontay"]
[FEN "6k1/1pn3p1/p1p1p1p1/2Pp4/PP1P4/4P3/1B4PP/5K2"]
[startply "67"]
1. d4 d5 2. Nf3 Nc6 3. c4 e6 4. Nc3 Bb4 5. Qc2 Nf6 6. e3 O-O 7. a3 Bxc3+ 8. Qxc3 Bd7 9. Bd3 Qe7 10. c5 a6 11. b4 Na7 12. O-O Bb5 13. Re1 Bxd3 14. Qxd3 c6 15. Nd2 Nb5 16. a4 Nc7 17. Bb2 Rfe8 18. Nf3 Rab8 19. Ne5 Nd7 20. f4 Nxe5 21. fxe5 f5 22. exf6 Qxf6 23. Rf1 Qg6 24. Qxg6 hxg6 25. Rae1 Rf8 26. Rxf8+ Rxf8 27. Rf1 Rxf1+ 28. Kxf1 Kf7 29. Kf2 Kf6 30. Kf3 Ne8 31. Kf4 g5+ 32. Kg4 Kg6 33. g3 Nf6+ 34. Kf3 Kf5 35.h3 g4+ 36.hxg4+ Nxg4 37.Bc1 Nh6 38.Bd2 Nf7 39.g4+ Kg5 40.Kg3 e5 41. dxe5 Nxe5 42.e4+ Kf6 43.exd5 cxd5 44. b5
White has some counterplay and should draw.
- So the most dangerous plan is probably to gain even more space with e6-e5-e4, place the king on g4, and target Pg3.
It looks promising:
[Title "jasonhaller-RewanDemontay"]
[FEN "6k1/1pn3p1/p1p1p1p1/2Pp4/PP1P4/4P3/1B4PP/5K2"]
[startply "67"]
1. d4 d5 2. Nf3 Nc6 3. c4 e6 4. Nc3 Bb4 5. Qc2 Nf6 6. e3 O-O 7. a3 Bxc3+ 8. Qxc3 Bd7 9. Bd3 Qe7 10. c5 a6 11. b4 Na7 12. O-O Bb5 13. Re1 Bxd3 14. Qxd3 c6 15. Nd2 Nb5 16. a4 Nc7 17. Bb2 Rfe8 18. Nf3 Rab8 19. Ne5 Nd7 20. f4 Nxe5 21. fxe5 f5 22. exf6 Qxf6 23. Rf1 Qg6 24. Qxg6 hxg6 25. Rae1 Rf8 26. Rxf8+ Rxf8 27. Rf1 Rxf1+ 28. Kxf1 Kf7 29. Kf2 Kf6 30. Kf3 Ne8 31. Kf4 g5+ 32. Kg4 Kg6 33. g3 Nf6+ 34. Kf3 Kf5 35.h3 g4+ 36.hxg4+ Nxg4 37.Bc1 e5 38. Bd2 {38. dxe5 Nxe5+} e4+ 39.Kg2 Nh6 40.Kh3 {40.Be1 Kg4 41. Bf2 Nf5} Nf7 41. Bc3 Ng5+ 42.Kh4 Nf3+ 43.Kh3 Kg5 {Ng1}
Black's winning chances are very serious. This third plan would be my favorite choice.
Computer analysis will refine this and probably prove me wrong more than once, but it also spoils all the fun of searching for the truth...