quoting Morrígan, Countess:

Cool; no idea what is the appropriate thing to do in that case; just go with the first element? That seems to be insufficiently a principled decision.

For nōn, that's probably the best choice, since nōn syntactically and semantically patterns with Sanskrit na, OCS ne, OIrish ní and the like, and not with words meaning "one". I agree, though, that this seems a very touch-and-go rule that won't apply to everything. (We could just decide to live with this...semantics are fuzzy, after all.)

If we're using a master list of roots to construct this, maybe it's better to break each word down into its constituent parts and turn such words into disambiguation pages of a sort? I know travisb recommended using a real database, not ad-libbing a spreadsheet. I don't know how to do databases, but I can certainly try to learn.