Report 04 · August 2026 · Synthesis

Why stages concentrate: a mechanism taxonomy from 58 value chains

Critical Materials Atlas · independent public-data research · derived from the 58 end-use value chains, each on public production, ownership and customs data

Abstract. The atlas’s value-chains layer traces 58 end-use systems — from chips and solar to fertilizer and jet engines — from raw material to finished product, and marks the single stage in each that actually binds supply. One pattern dominates: the binding stage is rarely the mine. Of 58 chains, only 4 have a chokepoint that is genuinely geological — a deposit that cannot be substituted or relocated. The other 54 were made. But “built versus geological” only labels a stage after you already know where it binds. This report proposes a mechanism taxonomy: seven reasons a stage concentrates, each stated as a rule about the process itself, so the concentration is predicted rather than described — and each mechanism carries a different fix. Two chains falsify the simplest version and set the multi-regime structure: indium refining concentrates because it is a by-product of zinc, not because it is hot; single-crystal turbine blades concentrate because of qualification, though they are hot. The result is a decision layer: the mechanism tells you whether a dependency can be out-built, and how.

Addendum, 29 August 2026 — after an expert review of the taxonomy. A methods review of the classifications produced three changes, reflected on the live Chokepoint Map (this report preserves its original August figures): (1) an eighth mechanism, “fragility,” was added — a supply riding on a few ageing or irreplaceable facilities with no buffer (medical isotopes), which the seven-way scheme had folded into “capability”; (2) strontium was corrected from “diffuse” to geological (Iran ≈ 58% of celestite), moving the geological count from 4 to 5 of 58; and (3) five chains are now shown as a genuine blend of two mechanisms (helium, nuclear, vanadium, beryllium and the isotopes case), four of them flagged for deeper research on the exact split. The taxonomy got more honest for being reviewed — which is the posture the report argues for.

Key findings

Only four chokepoints are geological. Phosphate (~70% Morocco), the platinum-group Bushveld, niobium’s single Brazilian mine and boron’s Turkish borate reserves are the only stages fixed to an unrelocatable deposit. For these you cannot add supply — you substitute, recycle or stockpile. Every other chain’s chokepoint was made, and made things can be remade.
Concentration tracks the process, not importance or tonnage. Cement and aluminium are both bulk materials, yet cement has no chokepoint and aluminium is ~59% one country. The difference is physical: aluminium smelting is continuous and hot, so it must be one large unit at cheap power; a cement kiln is switchable and local, so it spreads. The binding question is whether the stage can be small.
Seven mechanisms cover the layer. Thermodynamic (6 chains), by-product (15), built capability (20), geological (4), governance (2), policy (1), and diffuse — the switchable stages that never concentrated (10). Each is a testable claim about the binding stage, stated in the table below.
The mechanism is the fix. A thermodynamic chokepoint is a build-a-smelter-at-clean-power problem; a by-product cannot be willed into existence and must be recovered from its host; a built capability takes years to qualify; a geological one cannot be out-built at all. Naming the mechanism turns a list of dependencies into a list of decisions.

1 · Why sort by mechanism

The layer’s first finding — the chokepoint is rarely the mine — is descriptive: it tells you the bottleneck has usually moved downstream, to a refinery, a fab, an enrichment plant or a transformer. The sharper question is why a particular stage gathers into few hands, because a rule about the process predicts the concentration and, more usefully, names what it would take to undo it. Splitting chains into “built” and “geological” labels them after the fact; a mechanism classifies them from the physics of the binding step.

A stage concentrates when something about the process forces it to be large, single-sourced or hard to replicate — continuous heat, coupling to a host material, a qualification barrier, or the geology itself. Where nothing forces it, the stage spreads. Concentration is the exception that needs a cause; diffusion is the default.

2 · The seven mechanisms

The 58 value chains classified by the mechanism that concentrates their binding stage, derived from each chain record’s `chokepoint` field (single source of truth; the counts are re-checked against the records on every build). “Chains” = number in that class. Each rule is a falsifiable statement about the process.
Mechanism	Chains	What makes it concentrate	What it would take to move it
Built capability	20	A refinery, fab or separation plant whose barrier is capital, process know-how and years of qualification. Switchable in principle, concentrated by accumulated capability.	Fund and qualify capacity elsewhere over years — slow, but not fixed by nature.
By-product	15	The material is not produced for its own sake; output is bounded by a host (zinc, copper, gas, steel), so it cannot scale on demand.	Incentivise recovery at existing host refineries, and recycle — it cannot be willed into existence.
Diffuse (spreads)	10	A switchable, cool process with no concentrating force — so it spreads. The constraint is scale, logistics, labour or time, not one country.	Nothing to break: build capacity broadly. The rule working in reverse.
Thermodynamic	6	A continuous, high-temperature process that cannot be stopped without wrecking the plant — so it must be one very large unit, sited where power is cheap and never fails.	Build capacity at cheap, reliable (clean) power — capital plus energy policy.
Geological	4	A deposit that cannot be substituted or relocated. The only mechanism where you cannot add supply.	Substitution, recycling, stockpiles — not new mines.
Governance	2	Supply concentrated in conflict-linked or artisanal production from specific regions (the 3TG minerals). The scarce thing is documented, responsible supply.	Traceability, due diligence, alternative sources.
Policy	1	Concentration created or enforced by a rule — an export ban, resource nationalism.	It can reverse in either direction, quickly — trade and diplomacy.

Two features of the table matter. First, diffuse is not a failure to classify — it is a finding: ten chains (cement, wind, grids, EV assembly, heat pumps, copper’s lead-time, lead’s recycling loop, baryte, strontium, data-centre power) have no single-country chokepoint, and the reason is always the same — the binding stage is switchable and local, so nothing pins it. Second, geological is the smallest class by far. The intuition that critical-material dependence is mostly a mining problem is almost exactly wrong: it is overwhelmingly a processing, coupling and capability problem, which is far more tractable than geology.

3 · The two cases that set the taxonomy

The simplest version of the idea — a single rule that a continuous-hot process concentrates while a cold one spreads — is real but incomplete. Two chains break it in different directions, and each forces a separate mechanism.

Two falsification tests. Each would look identical to the others on a “built vs geological” chart, yet each concentrates for a different reason and admits a different fix — which is the reason to sort by mechanism rather than by outcome.
Chain / stage	Hot?	Continuous?	Why it actually concentrates
Aluminium smelting	Yes	Yes	Thermodynamic — a potline freezes if power drops, so one big unit at cheap power.
Indium refining (~70% one country)	No	—	By-product — refined electrolytically at modest temperature; it concentrates because it rides on zinc, not because of heat.
Single-crystal turbine blades	Yes	No (batch)	Built capability — cast one at a time; concentrated by decades of metallurgical qualification, not power-siting.

Indium proves that heat is not necessary for concentration (a cold, switchable process can still be single-sourced if it is coupled to a host); single-crystal blades prove it is not sufficient (a hot process can still spread across qualified suppliers, or be pinned by qualification rather than power). The taxonomy therefore needs at least these distinct regimes — thermodynamics, by-product coupling, and qualification — plus geology, governance and policy. The thermal-continuity rule earns its place as one mechanism of seven, not as the whole story.

4 · What the taxonomy is for

The point of naming the mechanism is that it names the response. A policymaker or a firm looking at a dependency does not primarily need to know how concentrated it is; they need to know what kind of concentration it is, because that determines whether it can be undone and at what cost. A thermodynamic chokepoint yields to cheap, reliable power and capital. A by-product chokepoint does not respond to price at all — you cannot mine gallium; you recover more of it from the aluminium and zinc you already refine, or you recycle. A built-capability chokepoint (a chip fab, rare-earth separation, a turbine-blade foundry) is a multi-year funding-and-qualification programme, hard but finite. Only the four geological chokepoints cannot be out-built, and there the entire toolkit is substitution, recycling and stockpiling. The taxonomy converts a wall of concentration statistics into a short list of decisions, each with a known shape.

What this report is, and is not. This is a classification synthesis, not a statistical test. Each chain’s mechanism is a reasoned judgement about its binding stage, made falsifiable by stating the process rule explicitly (a reader can check whether indium refining is really electrolytic, or whether a potline really cannot be stopped). Every underlying figure — the concentration shares, the export-control dates, the histories — carries a per-figure confidence tag on the chain’s own page (measured, estimate, snapshot or proxy), and the boundaries of the customs data are flagged throughout. The taxonomy is a lens for reading the layer, offered on its logic and its sources, not a claim of statistical proof. Recoding-consistency check (added after review): all 58 stages were re-coded by a second, automated classifier given the stage facts and the eight definitions only (atlas label and physics hidden); agreement 86%, Cohen’s κ = 0.83. Because the re-coder is a model and not an independent human expert, this measures consistency (and possibly shared priors), not external validation — and a forced single label understates agreement on the chains shown as blends, where the eight disagreements concentrate (reproducible: build_taxonomy_agreement.py).

5 · How it relates to the established taxonomies

Criticality already has taxonomies, and this one does not replace them — it answers a different question. The established schemes classify materials by how critical they are; this one classifies chokepoints by why the concentration exists, and therefore what would move it. They sit one layer apart: a criticality screen tells you which dependencies to worry about; the mechanism taxonomy tells you, for a given dependency, what kind of problem it is. The comparison below is our reading of each framework’s stated design, not a claim about their internals.

Complementary, not competing. A criticality screen (EU / USGS) and this mechanism taxonomy differ in *unit of analysis* (material vs stage) and *question* (how critical vs why concentrated). The Payne structural view is the closest neighbour — a taxonomy of *why* — where this work adds the specific *binding stage* and a physical rule per mechanism. Sources: European Commission CRM/CRMA 2023; Nassar & USGS methodology; Payne Institute (Colorado School of Mines).
Framework	Unit of analysis	Question it answers	Output
EU CRM / CRMA (European Commission, 2023)	Material	Which materials are critical / strategic?	A two-axis screen — supply risk × economic importance — yielding a list (34 critical, 17 strategic).
USGS criticality (Nassar et al.; MCS)	Material	How critical, and how exposed?	A risk score / matrix — supply risk × impact of a disruption — plus a red/yellow/green early-warning screen.
Payne Institute structural view	Material / supply chain	What structural type of criticality?	A structural classification of why a material is critical (e.g. by-product, geographic, processing), close in spirit to this work.
This atlas — mechanism taxonomy	Stage (chokepoint)	Why does the binding stage concentrate — and what fixes it?	Eight mechanisms, each with a falsifiable physical rule and a matched response, assigned to the one binding stage of each of 58 chains.

So “were we adding an unnecessary taxonomy?” is a fair challenge, and the honest answer is: not orthogonal, but differently organised. We should not oversell it — several mechanisms plainly overlap the established supply-risk dimensions (our “governance” is their conflict-mineral/ESG axis; “policy” is their export-control/trade-measure axis; “by-product” and “geological” are standard structural explanations). What the mechanism taxonomy adds is not a new axis of criticality but a mechanism-oriented crosswalk: it fixes the classification to the single binding stage of each chain and attaches a falsifiable physical rule and a matched response to it. So the two compose in practice — run an EU or USGS screen to decide which dependencies matter; read the mechanism to decide what to do about each one (does it yield to power and capital, to recovery and recycling, to a multi-year build-out, or only to substitution and stockpiles) — while accepting that the category boundaries are a one-coder scheme awaiting a second rater, not a validated instrument.

Method

Each of the 58 chains carries a structured chokepoint record — its binding stage, the concentrating mechanism, a one-line statement of the physics, the holder, the share and any export control — authored from public production and trade data (USGS Mineral Commodity Summaries, BGS World Mineral Statistics, IEA, OECD/NEA, the European Commission’s CRM studies, industry associations, and CEPII BACI customs data). The Chokepoint Map and the counts in this report are derived from those records by a single build step (build_chokepoint_map.py), and a pre-flight check re-derives and compares them on every build, so the classification lives in one place and the tallies cannot drift from the per-chain pages. Mechanism definitions and the fix per mechanism are held once, centrally. The layer is production-primary: trade is demoted to a flagged overlay, never used to split a broad customs code. Counts are as of the 58-chain layer at publication; adding chains updates the map automatically. Full per-chain sources and confidence tags are on each chain page, reachable from the value-chains hub.

How to cite. Critical Materials Atlas (2026). Why stages concentrate: a mechanism taxonomy from 58 value chains. Report 04. Zenodo. https://doi.org/10.5281/zenodo.21948855

@techreport{cma_report04_2026,
  title       = {Why stages concentrate: a mechanism taxonomy from 58 value chains},
  author      = {{Critical Materials Atlas}},
  institution = {Critical Materials Atlas},
  number      = {Report 04},
  year        = {2026},
  publisher   = {Zenodo},
  doi         = {10.5281/zenodo.21948855},
  url         = {https://criticalmaterialsatlas.org/report-mechanism-taxonomy.html}
}

Archived on Zenodo — concept DOI 10.5281/zenodo.21948855 (always resolves to the latest version).