Critical Materials Atlas
What you can trust

Built to be broken — here is what survives.

Most dashboards ask you to trust them. This one shows its work and tries to break itself. Every number traces to an open script; ten automated guards run on every change; and the atlas has been through three rounds of adversarial red-team review that caught — and published — three real errors. This page is the short version: what the atlas can say with confidence, and how it knows.

The discipline

Rigor is not a claim; it is a set of habits you can check. These are the atlas’s.

100%
of claims trace to an open, re-runnable script — no black boxes
10
automated guards run on every commit (dead links, stale data, unproven claims)
3
rounds of hostile external review — each attacked the previous round’s fixes
3
real errors caught and published, not buried

The rule throughout: never hand-tune a number to look right. Where the data cannot settle a question, the atlas reports the range and says so — concentration, for instance, is shown as a spread across four reconstructions, not a single figure, because no reconstruction of customs data is ground truth.

The reliability map

Not every number is equally solid, so the atlas grades each one. Of 31 tracked commodities, most carry a robust concentration and trend signal; a handful are flagged — up front, by an ex-ante screen, not after the fact. Draw headline conclusions from the robust rows.

21
robust — solid to headline
4
spread-sensitive — read with the band
2
shared HS code — not separable in trade
3
thin-fragile — too little trade to measure
1
engine-understated — trust BACI here (cobalt)

See the full per-commodity table on the Method page.

What the atlas can say with confidence

The findings below rest on robust rows only — and each ends where a finding should: with what to do about it.

The chokepoint is rarely the mine

Across 58 end-use supply chains, only 5 are constrained by geology — a deposit that cannot be moved. The other 53 concentrate at a stage that was built: a refinery, a fab, a separation plant, an export rule.

So what: most dependencies are not permanent. An export ban on ore cannot touch a refining chokepoint — but capital, energy policy and years of qualification can rebuild one. Target the stage that concentrates, not the mine. The Chokepoint Map →

The seller is usually not the source

For 17 of 32 materials the top exporter never mined the material — it refined someone else’s ore and re-sold it. The gap holds when measured in tonnes, not just dollars, and survives a permutation test (it is far more than chance).

So what: a customs ledger of “who exports” systematically mislocates the real dependency. Trace to the refiner, and to the mine behind it, before you name a supplier secure. The origin gap →

Concentration is moving — and we can date it

On the robust series, export concentration is rising for coking coal, manganese, bauxite, rare-earth magnets, magnesium, tungsten and strontium — several already single-country dominated (magnets, magnesium and tungsten toward China; bauxite toward Guinea). It is falling for copper, nickel, palladium, helium, vanadium and tantalum.

So what: the materials tightening are not always the ones in the headlines. A rising-and-already-dominated material (magnesium, tungsten) is where a shock would bite hardest and where diversification has the most to undo. The tested trends →

When the chokepoint is the mine, geology says why

The five geological chokepoints — phosphate rock, the platinum-group metals, boron, strontium, and niobium (the steel-alloy chain) — cannot be out-built. There is no second Bushveld for the platinum-group metals, no second Araxá for niobium.

So what: for these, the toolkit is not “build capacity elsewhere.” It is substitution, recycling and stockpiling — a different, slower policy, and worth starting early. Why the mechanism matters →

What we got wrong (on purpose visible)

A research site that never corrects itself is not being checked. The atlas keeps a public ledger of its own errors — because the corrections are the evidence the method works.

  • A price-coupling claim, withdrawn. We had argued by-product metals track their host’s price; a proper test on real price series refuted the general law (only bismuth←lead survived). The supply claim stands; the price claim was retracted.
  • Cobalt, diagnosed backwards, then corrected. A first pass called a cobalt trend conflict a “thin-code artifact” on a buggy extraction. Screening thinness across every commodity up front exposed the error: cobalt is a real $0.8–4.9B market, and the finding is that the engine understated DR Congo’s real dominance — a genuine limitation, kept in view.
  • A screen that flagged the wrong thing. An early fragility rule mislabelled nine genuinely-concentrated commodities (bauxite, magnets, tungsten) as unreliable. Fixed: single-country dominance is a finding, not a data flaw. The real fragile list is three.

The full history is in the changelog; the standing invitation to break the atlas is here.

Explore the 32 materials → Read the full method Audit the atlas