Even after many (small ones, that slowly added up, I've now looked this over with a computer) inaccurices, apparently 33...Re8! is still very good for black, up +1, just trying to consolidate the position with ...Kb7, a move I somehow missed. Of course I thought that endgame would be very good for me, and in fact it still says black is slightly better throughout the endgame. It turns out 44...Rxb2 does draw, and ...Rc7 loses, which is quite ironic, because the whole point for me playing rc7 instead of grabbing the pawn was damage control, preventing Rh7. I was worried about 45 Rh7 but after ...Kd8 black is fine, and if white tries to advance his pawns black can play ...Rd2, when the tactic I feared Rxf8+ of course isn't dangerous because I'm only losing an extra piece, and we both queen and it's a draw.

Interesting game. I would resign right away after losing a piece like that, my feeling of the position, and what engine evaluations say is that white is pretty lost in there. But he hung on very well, made it very annoying for black. 

My opponent had a 45 minute rating of about 1700 and mine about 2000, but they're probably not very reliable because I had only played 4 45 minute games when I had that rating and most likely he hasn't played many either because the idea of a 45 minute rating is pretty new. In fact my rating went down 100 points just from that loss.

Yeah I know. But I remember old games when I used to do the opposite and just stop caring about pawns figuring I was up a piece but what I ended up doing was just giving my opponent more chances to get back in when they have some pawns for it, so I'm trying to find the balance. At the time I didn't think my position would improve much by madly giving away pawns. I was concerned about the possibility of white playing exf6 at any moment, that's why I played ...fxe5.

But if I was still better after fxe5 then at what point did the game turn around? I don't know if fritz can be trusted in that kind of position after ...fxe5 because like me it didn't see any clear path for white and probably just concluded black was better because he was a piece up. Computers are even more dogmatic. After as far as ...fxe5 I still can't see any clear improvement.

What you played was still okay. I put the game in the computer and it said you were winning until you played fxe5 and even then you still had an advantage just no longer winning. I think when youre up a piece you shouldnt worry so much about giving back a few pawns you need to actually use your extra piece.

Yeah I had a feeling it was a computer line. I had looked at ...Qb6 but thought I had Qa8, forgetting my knight would also go. In the psychological state I was in there was no way I would find that or think such an idea (moving a piece twice) was necessary to get the win (at least an easy one). I just found it really hard to take white's chances seriously until it was too late. I knew my development was slow but just thought I would consolidate with my king being safe. Even if they blunder a piece in the process it took a few moves to pick it up and in that time white already has compensation, even if he didn't plan it and that always annoys me.

So yeah I'm not surprised a win was there but it was just a matter of finding it, and realizing that I still had to play accurately.

If nxb5 then just qb6 which pretty much forces a queen trade and you have two bishops and a knight against two knights you'll eventually be able to pick off all the pawns you lost with the extra piece.  Computer check it, fritz gives between a 2 and 3 point advantage in the lines with 15... nf6, which is winning.

Bringing the knight to d7 is another way to give back two pawns. After 16 Nxd4 Nd7 17 Nxb5 Bc5 leads to a similar situation. This line is probably one of the better ones because now white can't castle. White wants to play d4 with tempi (then I think the space would give him a very decent position) but that could be hard if the knight moves to b6, like on 18 Qa4 0-0 19 d4? Nb6, or 18 Qa2 Nb6. White could still try to play for e5 and Ne4-d6 though, or even Ke2, Rd1 and d4, and hope b4 is weak some day. But I don't think black is clearly winning. Again I wasn't in the mood to give back a second pawn because I thought it was going to give white "unnecessary drawing chances", when in fact I had to be very careful.

15... Nf6 is very good, then you can bring the knight to d7 and bishop to c5 and castle and then black is clearly winning.

Really interesting game. I enjoyed it a lot. I think I have a similar kind of materialistic problem as well. If you're naturally a kind of positional player I think it's pretty common. It happened in this recent game: http://www.chess.com/forum/view/game-analysis/interesting-middlegame-enough-compensation-for-the-pawn that on move 15 I just thought I'd won a pawn, but had obviously misjudged the position after the exchange and without apparently making any obvious mistakes after that my position just seemed to deteriorate until he "wussed out" on move 22. Both games I think show the importance of calculating one more move than you originally think you need to, and taking seriously the dynamic factors in positions.