The official narrative is they do sequencing at a regional level on a small sample of tests (not the normal screening test).
From there they get the % "variant" in the sample, and extrapolate to the larger region samples.
The sequencing is it's own debacle as a lot of it is computer sequencing that has like 80% specificity. When it comes to genetic code that's really low, and seems like massive hubris to think that's a good number.