Andrews Soltis has a good book, "The Inner Game of Chess" but it has some odd biases. Being written in 1994 he is somewhat dismissive of computers and their ability to calculate positions, with zingers like:

"The principal difference between human players and many computers is that those machines cannot identify candidate moves and as a result cannot budget their time. They might consider all 35 legal moves in a position rather than the only two that make sense. Humans can budget their time, if they develop good instincts and intuition."

Soltis also gives analysis on this game with black to play at move 35.

Amador Rodriguez Cespedes - Anthony Miles 0-1, GMA Baleares op 1989

click for larger view

Soltis writes:

"Black studied the position in some detail, working down a tree that began with 35 ... Nce3+ 36 Kg1. He looked at one branch that had 36 ... Qg4 as its key but gave up after he found that 37 Rh8+ Kf7 38 Rc8 sets up a strong checking threat. He turned to another branch that began 36 ... Nh4 with the idea of 37 Qxg6 Nf3 mate. But he had to give up on it after he saw 37 Rh8+ and then Kf7 Qxg6 or Kxh8 Rxh4+.

He examined other lines as well, and the main benefit of all this analysis was to tell him that Black's main problem was his king position. "If only it were somewhere safer," he thought. Therefore:

35 ... Kf7

This winning idea would not have occurred to Black if he hadn't worked on a different tree. Now, for example, 36 Rh8 Ke7! 37 Rc8 allows one of the clever ideas of the first tree: Nce3+ Kg1, Nh4!, Qxg6 Nf3 mate.

Unfortunately, neither Miles (nor Rodriguez) noticed that they had missed an option in one of the branches.

After 35 ... Kf7, instead of the game continuation 36 R7h5, white has:

36 Rh8 Ke7
37 Kh2

Amador Rodriguez Cespedes - Anthony Miles 0-1, GMA Baleares op 1989

click for larger view

(not Rc8 in Soltis' line), and it's white, not black, who is winning (and by a lot!).