It can be tested, just not predicted; however, I'd wager most people would say the goal is a description rather than a prediction of the process. That is to say, not to identify which traits are most helpful to their possessors and document their growing prevalence as evidence of natural selection, but to document helpful traits enhancing survival and thereby predominating in future generations as evidence of natural selection. I think what we're dealing with is a truism rather than a tautology, and while the two are very similar, there are critical differences: A truism is a statement of undeniable fact, like "2+2=4"; a tautology is an equivalency masquerading as a comparison, like "2+2 resembles 4".

Natural selection, as I understand it, is the phenomenon of traits enhancing survival predominating over time as they become an increasingly large part of the gene pool, forcing out less useful traits. Organisms with traits enhancing survival are, by definition, more likely to survive and pass on those traits, and those without them are less likely; that's a truism, not a tautology. Identifying such traits specifically is immaterial to verifying the occurrence of that process; in fact, the attempt can actually prejudice it because traits that might initially seem more beneficial can prove otherwise in practice. It's difficult to dispute the occurrence of natural selection as defined above, because it's what we would logically expect and all evidence supports it (again, it's important to note that's not automatically proof, just very compelling evidence). An independent opposing force would almost be required to PREVENT it.