I noticed that whenever I had to resort to trial and error or to set up a forcing chain in the bilocation/bivalue graphs so far I ALWAYS chose the wrong number when there was a binary choice, with only one exception (out of a dozen guesses).
First I thought "bad luck!", a couple of puzzles later "Murphy's Law!", but now I tend to think that this is a systematic effect, even though the "statistics" is not overwhelming...
I don't mean to imply that there is some systematics in a 50:50 effect - of course there is none! But I believe that the choice in reality is not 50:50.
Let me explain why: I do not choose at random, but rather I take the number which "stands out a mile"; so far there always has been a unique choice, because one of the two numbers typically collapses more cells than the other - one of the two choices typically is more "rigid", in the sense that it has stronger implications on other cells in the sudoku.
And this is probably the reason why this number leads to a contradiction: BECAUSE it has stronger implications on the remaining cells in the sudoku.
Of course, this will not always be the case, nor is this anything resembling a "proof" that this will always work that way, but it seems plausible enough to me to explain why most of the time the "convenient" choice leads to a contradiction.
Has anybody made similar experiences with guessing/choosing the initial number for a forcing chain?