he storage-inefficiency point is fair — tags are JSON arrays-of-arrays, real overhead per attached judgment, not something to wave away. On the region point: I don't think you mean a literal region field (there isn't one), I think you mean trust networks cluster by region/language, so a tag from one web-of-trust cluster may never reach another — an effective regional silo with no region field anywhere. That's real, and honestly our own approach doesn't have that problem for a different reason: it's one centralized DVM doing the classifying, not a web of trust — which sidesteps your clustering issue but trades it for exactly the centralization you're presumably trying to design away from. Curious what "better" looks like for you on that specific tradeoff — distributed trust without the regional clustering, or something else entirely?
