The layer's first finding was that the chokepoint is rarely the mine. But that labels stages after the fact. The sharper question is why a given stage concentrates, because the answer predicts it from the process and tells you the fix. Stages concentrate for a small number of distinct reasons: a continuous-hot process that physically cannot be small; a by-product that cannot scale on its own; a built capability that takes years to qualify; a deposit that cannot be moved; a conflict-linked supply; a policy; or a handful of fragile, irreplaceable facilities with no buffer (this last mechanism added by an expert review of the taxonomy). A further group barely concentrates at all. Name the mechanism, and the map becomes a to-do list — and some chains are a genuine blend of two, shown as such.
Click a mechanism to filter; click a column to sort; expand a mechanism below for what it would take to move it. Each chain links to its full page. "Share at the choke" is the concentration at the binding stage (not always the mine).
| Chain ↕ | End product ↕ | Binding stage ↕ | Why it concentrates ↕ | The physics ↕ | Who holds it ↕ | Share ↕ | Export-controlled ↕ |
|---|
The two cases that set the rule
The taxonomy has more than two regimes, and two chains prove it. Indium refining (~70% one country) is done electrolytically at modest temperature — not a continuous-hot process — so thermodynamics is not what concentrates it; it is a by-product of zinc, bounded by the zinc it rides on. Single-crystal turbine blades (aerospace) are hot but batch, cast one at a time; their concentration is decades of metallurgical qualification, not power-siting. Both would look identical on a "built vs geological" chart, yet they have different causes and different fixes — which is the whole reason to sort by mechanism. The thermal-continuity rule itself — a continuous-hot stage cannot be small, a switchable cold one can — is one of these eight mechanisms; generalising it, and the two tests above, are the atlas's own analysis.
What the map is for
The mechanism is the decision. A thermodynamic chokepoint (aluminium, magnesium, ferrochrome, steel, ammonia) is a build-a-smelter-at-clean-power problem — capital and energy policy can move it. A by-product (gallium, germanium, indium, helium) can't be willed into existence; you incentivise recovery at existing host refineries and recycle. A built capability (chip fabs, rare-earth separation, battery cells, turbine blades) takes years of funding and qualification but is not fixed by nature. A geological one (phosphate, PGMs, niobium) is the only kind you cannot out-build — there, substitution, recycling and stockpiles are the whole toolkit. Governance (tantalum, tin) is a traceability and sourcing problem; policy (nickel) can reverse in either direction, fast. A fragility chokepoint (medical isotopes) is a too-few-ageing-facilities problem with no buffer — build redundancy, if a decay clock even allows it. And the diffuse chains (wind, grids, heat pumps, copper's lead-time) are the switchable-and-cold ones that never strongly concentrated — the rule working in reverse.
A research synthesis of the Value Chains layer. Read each chain for its sources, confidence tags and boundaries. A share shown with a dotted underline is an estimate, not a figure quoted from a primary source — click it to see how it was derived and its limits. Plain shares are quoted from primary data (mostly USGS).
A + second mechanism on a row marks a genuine blend; a ? marks a split flagged for deeper research (how much of the concentration is each mechanism). Both came out of an expert review of the taxonomy — see the changelog.
How to weigh this taxonomy. It is a classification, not a statistical model — each chain’s mechanism is a judgment about its process, originally coded by one person. We since ran a partial version of the check the standard demands: a second, automated re-coding of all 58 stages — given only each chain’s product, binding stage, holder and share plus the eight definitions, with the atlas’s own label and its physics rationale hidden — agreed on 86% (Cohen’s κ = 0.83). Two honest limits a second reviewer pressed: the re-coder is an automated classifier, not an independent human expert, so this measures recoding consistency (and may share priors with the first pass), not external validation — a high κ here should not be read as confirmation that the mechanism labels are right, only that they are reproducible; we no longer call it “almost perfect inter-rater reliability”; and the stage wording, though stripped of the physics tell, may still carry cues, so the blinding is imperfect. What the exercise does show cleanly is where the scheme is soft: the eight disagreements fall mainly on chains already shown here as blends (beryllium capability↔geological, cobalt by-product↔capability, antimony capability↔policy), i.e. a forced single label understates agreement exactly on the cases the atlas already marks as two-mechanism (reproducible: build_taxonomy_agreement.py). Outside re-codings are welcome via the review invitation. It complements established criticality frameworks (EU CRM, USGS, and structural taxonomies such as the Payne Institute’s) rather than replacing them — they score how critical a material is; this asks why its bottleneck concentrates (a side-by-side comparison is in Report 04 §5). And after the review that added the eighth mechanism, the set is now treated as frozen: awkward cases are shown as blends or flagged for research, not absorbed by inventing a ninth category.