So I put this draw through the computer analyzer here, and it came back with everything clean but two inaccuracies (neither made a big difference).  Can you spot anything the computer can't.

no i can't 

I too don't understand 16... Kc7; you probably should've played there 16... Rhe8, so if White takes the d5 pawn you get counterplay with ... Re2.  For his part, White should've played 20 Kd2 with a comfortably winning endgame, instead of pushing his pawns on the queenside and creating weaknesses.  (Computers by the way are notoriously unreliable at evaluating endgames.)

yeah, i agree, it also depends on the strength of the analysis.  But I didn't really see anything that the computer didn't notice. white had the pawn advantage and lost it though at the end.  That was the only thing I noticed.

I have a feeling that the 2000 rated computer would have had a draw as well then.  Some of my other moves may have contributed to losing the pawn.

Actually, I realized (at work) why the computer suggested 16... Kc7.  It is to avoid 17 c4 dc 18 Bxb7+ (in fact, it mentions the move in the variation it gives after 16... f5).