Let's say that a new trait has been found, or made in a lab and you want to determin whether it is beneficial for a population (that is what you propose).
Oke, well, let's test it.

How shall we proceed? Well, there are various ways, we could introduce this new trait to either a lab environment with other individuals who do not have this trait and see who survives, or we can "let it loose" (either a single organism, or a small population of organisms we let reproduce in the lab).

Next, what can we measure. Well, we can measure if that trait is still present in the test population after X number of generations, let math do it's work to determine how much generations we need and how many individuals we need to test in order to determine the frequency with a reasonable amount of certainty.

There are several scenario's possible, but lets just pick 3 possible outcomes.
A) the trait has disappeared
B) the trait does still occur, but at a low frequency, about the same frequency it was there at the start
C) the trait occurs in (almost) all individuals

In case A we conclude: the trait was not beneficial, when B we conclude that it was (near) neutral and when C we conclude that it was beneficial.

Fine, but that only works IF natural selection is true in the tautological form. So we have not tested or proven (or disproven) natural selection.