The argument here is in effect that no vernacular variant of any language could possibly embody a mistake. Which is true for some definitions of mistake, and false for others. The problem with the definitions that make it true is that there's a slippery slope from dialects to idiolects. How many people have to speak a dialect for it to count as a dialect? You can't have any threshold, or you get weird consequences in…
You can't have any threshold, or you get weird consequences in hypothetical cases, like all the speakers of a dialect dying off except one. So if you decide that no vernacular can embody mistakes, you're bound to conclude that nothing any individual does can be a mistake, so long as they do it consistently. But lots of disciplines involving living things have problems like this. You can't "define" species. A species…
You seem to think I'm naively making some argument that depends on the words breaking down at boundary conditions. Actually I'm not. I believe a language spoken by just two people (or even one) is just as real a language as one spoken by millions-- that the difference in the way they're treated is a matter of politics, not the intrinsic qualities of the language.