🚀 go-pugleaf

Disclaimer:  The following expresses a desire for an explanation, NOT a
flame against the comparative method or even against its results.

Something that's always puzzled me:  how do historical linguists decide
what form should be the standard reconstructed form?  Consider the following
two cases:

In the reconstruction of Proto-Indo-European, we find that Germanic languages
have /f/ where other IE languages have /p/ in a large number of words.
Confidently, we reconstruct IE /p/ and postulate a sound-change that affected
Germanic only.  Well and good.

But in the reconstruction of Proto-Romance, we find that Sard has /k/ where
other Romance languages have /s/ in a large number of words.  In this case,
the reconstruction is /k/, and the postulation is that a sound-change
affected >all< the Romance languages except Sard.  Apparent methodological
contradiction.

I realize, of course, that classical Latin provides a check on Proto-Romance.
Nevertheless, I suppose there must be other such cases that I have not
heard of.  What general principle determines what is the Right Thing?