How obnoxious stem variation is likely to be retained?

As I mentioned some time ago, I'm working, if fitfully, on pielang (working
name: Wa). The sound-changes I've instituted, this far at least, are not very
radical, but sometimes they conspire to make closely related forms diverge
wildly. The case inspiring this post is this: PIE *kerd "heart" quite regularly
yields Wa /Sard/, and the PIE genitive *krdos equally regularly yields Wa
/karado/.

Naturally, I'm considering whether I should replace the later form with
analogical /Sardo/. I do quite like irregular paradigms with historical
explanations, but I worry whether this case would stretch believability. What
say ye sages? :)

                     Andreas

Replies