22.i5 might have been a losing move. i3 looks much better, for example

|22.i3 23.i4 24.h5 25.k5 26.k2 27.o3 28.m3 29.q12 30.r5

The line I missed was |28.f5 29.j2 30.h6 31.j6

White had a relatively easy win with |29.i16 instead of L15.

White still could have won with |31.o18 instead of M17.

If |43.e21 44.g19 45.h18 46.i15.

If |61.s16 62.r15 63.r14 64.o12

In part you can chalk this win up to David Bush's reputation. At first I reconsidered the obvious |29.i16 move solely for that reason -- the theory that David would never have bothered to even make that move if the standard |29.i16 response would have led to a conventional win! In looking at |29.i16 further, however, I became worried about black's possibilities after |29.i16 30.j21 31.k17. I did not worry too much about this since I thought the move that white actually made, 29.l15, would lead to an easy win anyway -- which it would have, as David points out, with |31.o18 instead of 31.m17 -- except that I thought 31.m17 was a stronger move (oops!) ;-)

What I had worried about with |31.o18 instead of M17 was something like |31.o18 32.p20 33.p21 34.n19 35.q19 36.o17 37.p16 38.n15 39.q14 40.o12, which looked like a black win. In retrospect |31.o18 32.p20 33.m17 would have solved that problem.

LATE GAME LOWER LEFT QUADRANT
Where white chose 47.f15, the other option would have been |47.f17 48.i15, but Black could win with |47.f17 48.i15 49.e15 50.f12 51.f11 52.e14

LATE GAME LOWER RIGHT QUADRANT
David comments: If |61.s16 62.r15 63.r14 64.o12

I would have taken that option if I had thought black would respond that way, because it looked like 65.v5 66.u8 67.q6 68.q3 69.s5 70.v3 71.u3 72.t2 73.s2 might lead to a white win from that setup. The move that worried me, and deterred me from talking |61.s16, was 62.r13. From there, I could not see a way for white to break through to the center group of white links for a win.

|31.o18 32.p20 33.m17 34.n21 35.l19 36.k21 would have won for black the same as in the game. BUT

|31.o18 32.p20 33.p21 34.n19 35.m17 was winning for white.

After |61.s16 62.r13 63.r14 64.q16 65.s12 66.o12 it looks pretty much the same as the game.

|61.s16 62.r15 63.r14 64.o12 65.v5 66.u8 67.q6 68.q3 69.s5 70.t4 71.u3 72.u6 73.u7 74.s7 looks like an easy win for black.

Uh sorry, I meant |61.s16 62.r13 63.r14 64.q16 65.s12 66.o12 looks pretty much the same as the variation I gave.

And |29.i16 30.j21 31.j18 is horrifyingly ugly. I was relieved when you played L15, even though it was still winning, at least I wouldn't lose ugly.

20.h11 looks pretty good.

After |20.h11 21.q12 how does black stop white on the right side?

20.h11 would have been followed by 21.q12 22.o12 23.r14 24.q18 25.p15 26.o17 27.n16 28.m16 29.l15 30.k17 31.g17, which looks reasonable for white, or less likely either 21.q12 22.r12 23.s13 24.p11 25.r11 26.r15 27.o11 or perhaps 21.q12 22.r12 23.s13 24.p11 25.r11 26.q13 27.q14 28.p15 29.p16, both of which look OK for white.

.... and here I had been hoping -- at least at first -- that David might not have spotted the potential of 21.q12 ;-)

After |20.h11 21.q12, what about 22.s13?

Thanks Alan :-)