NS is the process through which evolution acts. The process itself is governed by things like biological fitness. Fitness is determined by the genotype and phenotype of an individual to produce viable offspring. NS requires a selection pressure to “do its job”.

I think what you are having an issue with (based on reading other posts) is that it is impossible for us to know which trait will be beneficial to a given species in the wild before it happens. In order to test something like this we would have to have a static environment which is hard to come by. However some have managed it. Have you ever heard of the guppy experiment? That experiment shows how selection pressure causes natural selection for a certain trait and thus can alter the gene pool (and cause micro-evolution). In this environment you can then hypothesize about what traits will enhance ones fitness given the selection pressure. Thus you have tested NS.

Edit: I wanted to add that NS is not random like most people think it is. The random portion about NS is finding out which individuals, if any, have the "right" combo of genes for a given selection pressure. Thus if a selection pressure is great enough or radical enough some species will not have the "right" genes and go extinct. NS is very specific in selecting the animals in a species with certain traits.