How “who could” is measured & caveats
Who could break it uses two signals, always labelled so they are never confused. The strong one (capability-adjacency) is the product-space density of each country to the refined product with the leader removed — latent capability to stand the stage up, not merely present output. It exists only for the 8 clean ore→refined HS pairs. For the rest we fall back to the current alternative-supplier share — a weaker proxy (who already produces, which is not who could scale). Who is building is announced public projects (Lynas, MP Materials, Iluka, Rio Tinto Rincón, Umicore…) plus USGS forward capacity to 2029; it is curated, not exhaustive, so a blank is a sourcing gap — shown blank, not read as “no one”.
Caveats. Refined-share leaders are export-based, so a giant that refines huge volumes but consumes them at home is understated — cobalt shows Finland, not China, for exactly this reason (China’s refined cobalt stays in China). Where the USGS Outlook measures refining concentration directly (gallium, magnesium, titanium…) we use its share instead, which repairs the shared-code dilution (gallium reads its true ~87%, not the 811292-diluted figure). The moat type comes from the trade feedstock signature + the HS-code provenance flags; the share from USGS where available else the capability layer. Built by build_breakout.py from capability.json, mine_refine.json, usgs_outlook.json, pipeline.json and scenario.json. See also the reallocation stress test (can the rest actually cover a cut, in tonnage — and does removal just shift the chokepoint?), the leverage map (how exposed is each importing country), Who actually refines, the product-space map, and shock scenarios.