I hadn't heard about this testing and it is indeed a test that shows that NS does not have to be a tautology.

But the book I'm reading anticipated this line of argumentation. He accepts the fact that some individuals and some traits have a better chance to survive and reproduce and thereby chaning the gene pool.
The guppy experiment is very interesting, but it shows something else: specific selection.

I'll repeat his conclusion here and then explain a bit:
"Special Definition - These are measurable, explanatory, testable, non-tautologous and true for a particular case. Problems: They are false for the general case; they do not unify nature; they are many disjointed conflicting theories masquerading as a single unified theory."

What he means is that in one case it can be shown that trait X is very beneficial, while in another case the same trait can be very unbeneficial.
When applied to the guppies: there were 4 scenario's that were tested and two results. For next generations it can now be predicted what NS will do in their specific situation concerning a single trait.
But the problem is, even in the guppy case, that it only works in a controlled environment.

In one discussion here the question of the stability of the fitness terrain came up. Now it would go to far as to say that it is totally random, but it is never fully static either. That means that nature can select for one type of guppies in one season, but the next there will be predators, two years later there will be a storm and the ocean floor will change it's color or structure and so on and so forth.

But when one tries to combine all these specific examples into a single principle or explain them all by use of a single 'law of nature', then we are lost again and have no other choice, but to define NS in a tautologous form.