It is White to play and draw and the pawns on h7 and g2 clearly hold the key. Since 1. h8=Q g1=Q is clearly hopeless (2. Qd8+ is met by Qd4+) White must start with a bishop move to cover g1. We can quickly reject 1. Bb6, because 1 . . . Rb4+ wins the bishop. (Actually, we should look a little more deeply to discover that after 2. Kc3 Rxb6 3. h8=Q Black plays Rc6+] and promotes his pawn the next move to avoid any trouble caused by Qd8+.)
So the answer must begin with 1. Bh2] and now Black has to play 1 . . . Rb4+ 2. Kc3 g1=Q] 3. Bxg1 Rb8 to stop White's pawn. Black now threatens Rh8 and Rxh7, and there seems little White can do about it. The only slim chance appears to be 4. Bd4 Rh8 (4 . . . f5]? is met most simply by 5. Kd3]) 5. Bxf6 Rxh7 6. h5] (the fifth and sixth moves may be played in reverse order). At least White has managed to bury the rook, but is it enough?
Play continues 6 . . . Ke2 7. Kd4 Kf3 8. Ke5 Kg4 and now the beautiful final touch 9. Ke4] Kxh5 10. Kf5] Black must move his rook and lose it, leaving a draw position.
Herbstman must have started his work in this final position of mutual zugzwang: White to play loses; Black to play draws. By moving the kings and adding a white pawn on h5, he set up the final dance, and the early moves were just camouflage. A beautiful conception, set elegantly and economically, all staring from the paradoxical idea that a lone bishop might draw against rook and two pawns.