Yeah, I would advise adding the root for sure; a lot of cognates may not be necessary if we approach this whole thing with the intent of plugging dictionaries together.
Rather than working on LIV heavily, I may switch back to converting the Mallory and Adams data I already have; that contains a fair amount of cognate data itself. Preserving provenance is important, so if you include cognates, I would suggest having a way to indicate which text, and page made the claim.