14…Na4?! 14…g5! is better according to Stockfish. Thought about it, but didn’t spend much time evaluating the difference.
15.Rxd7? 15…NxRd7 is equal (it covers the critical b6 square) after 16.BxBe7 NxNc3, 17.BxRd8 (Black is giving back the exchange) RxB, 18.Rc1 Ne4, 19.Rd1 Nf6, and Black is back in time to defend. Sara pointed this out to me after the game, and I never even bothered to calculate it OTB.
19.g4?! Not a good move, but it served it’s purpose in giving her king luft and distracting me by it in my time-pressure.
21…Nxg4?? I felt this move was anti-positional, but I knew that either Ne5 or h3 would defend it, and it seemed it may be now or never to take it, but Black needs to grab more space with 21…b5, 22.Ne5 Rcc8, 23.h3? Ne4, 24.Bh2 Ne4-c5, 25.Bd1 f6 (hitting the Ne5, which has nowhere useful to go, and Black is winning. I needed to grab position rather than material.
26….Re8?? My blitz instinct was to correctly play 26…Rd2, but I neither saw White’s next move, nor the point of 26…Rd2, which only works after 27.Nd4 Nxe3! (and White is only up +1, amazing), 28.fxNe3 Rxe3, 29.Bf2 Re4. I can’t even begin to explain how this position works outside of the reduced pawn count – A rook and three pawns for a bishop and two knights.
I said after the game that taking the g4 pawn cost me the game, and I was right. After overnight analysis, the computer now says that it is +2 for White, to have a bishop and rook as White vs rook and three pawns as Black.
26… Rd2 27. Nd4 Nxe3 28. fxe3 Rxe3 29. Bf2 Re4 30. Rf1 Bc5 31. Nd7 Rg4+ 32. Kh1 Rxf2 33. Rxf2 Bxd4 34. Nf6+ Bxf6 35. Rxf6
At the time, I didn’t think that my clock was deciding the game, but it seems that it was in time-pressure that my game deteriorated yet again. Also, when I did use my time I didn’t work hard enough in analyzing continuations – didn’t bother to calculate many of them, only the ones I wanted to, and how I wanted to.