Genetic drift is not usually considered 'evolving' because the variants are selectively neutral. That's why finding selected variants isn't as simple as 'SNP X is 0.00001% more common in generation 2 than generation 1, case closed!' - you are looking for variants which are becoming more/less frequent than would be plausible under simple drift. Just like an experiment where you know the effect will never be exactly 0 because of chance imbalances and random noise.