Wash-Building Needs External Value Anchors, Not Graph Signals

Wash-building is not Sybil: it uses real identities, so graph-internal signals cannot separate it from genuine collaboration. No classifier beats ~0.60 accuracy. Wash-resistance requires an external value anchor. Solution: contribution-denominated bonds, refunded on external use, with coordination-proportional exposure. This forces conditional, non-capital costs. GitHub: WGlynn/noesis.