nabla:
It's simple:
Take for example the line after After 1.d4 d5:
2.c4 games played = 30, Percentage wins = 93%
2.f4 games played = 6, Percentage wins = 50%

It is not clear at all that 2.c4 is the better move here. You will say but why? It has won for white in the 93% of the games! A huge difference over 2.f4 that wins on only 50%. But this may be completely deceiving.

Consider for example that after 2.c4 and for simplicity's sake, that there are 2 responses to this:
2...X1 that has been played 26 times with a devastating score of 95% in favour of white and 2...X2 that has been played 4 times with a bad score for white of 25%.
2...X1 was the move all people played some time ago, until the new move 2...X2 discovered and been played with a good success for black.
That means we possibly have a refutation to 2.c4 since 2...X2 brings good results for black! Although statistics say 2.c4 has a good %, since many games were played with the bad response for black 2...X1.

The bad thing is the refutation may be deep in the openings-data tree or there may be another refutation to the 2...X2 move later in the tree so 2.c4 is good after all! To solve all these a complete examination of the whole opening-tree should be done starting from the leaves of the tree and going up all the time until we end to the starting opening moves. In that way going backwards(that's the meaning of backsolving) we find the best value(+-,-+, ++-, --+,etc) of each node.....