Critical Materials Atlas
Method · the reconciliation, improved

The production-constrained anchor

A trade reconciliation should respect an attribution bound: a country can’t be credited as the mine origin of more than it produces minus what it absorbs at home. Feed it a consumption estimate built without trade, and that becomes a testable constraint — where it binds — that repairs the origin gap, where customs credit the refiner and re-exporter, not the mine. Honest caveat up front: the consumption term does real work in only about half the corrected rows — each row below shows exactly how far it moves the answer.

Research / method layer. This demonstrates a technique on top of the atlas’s reconciled trade. The consumption input is an estimate (activity × calibrated intensity), and the blend weight is a documented default (w = 0.5). The direction and magnitude of the corrections are robust; treat the exact corrected percentages as indicative.
How it works. Three independent inputs: production (World Mining Data, per country) · consumption (trade-independent bottom-up = national activity × a calibrated material intensity, at uniform per-country intensity — so a country’s modelled consumption share equals its share of the driver activity, and the model cannot detect a country using more or less of a material than its industrial size implies) · observed exports (the atlas’s reconciled mirror trade). For each producer, exportable supply = max(0, production − (1−R)·consumption), where R is the recycled share. Two levers: a hard guardrail (no origin can export more than it can supply) and a soft prior (pull a deleted origin back toward what production says it must be). We assert only the safe direction — a mine leader credited far less than it can supply (classic refiner-fronting) — and flag the rest. Reproducible: build_anchor.py → anchor.json.

Why not imports + recycling? The full material balance is exports = production + imports + recycling − consumption − losses. But the anchor answers a narrower question — whose mine did this come from? — not “how much can this country ship?” Imports are deliberately excluded: if China refines imported Congolese cobalt, that cobalt is Congo’s origin, not China’s — adding imports would re-create the very origin gap we correct. Re-exported imports still show up, as the review flag (a country exporting more than it makes minus uses). Recycling IS included (the R term): a country’s own recycling meets part of its use, so only (1−R) of consumption draws on primary supply — R is an approximate world recycled share (USGS/EU CRM), 0 where negligible. The honest remaining gaps are losses and stock changes, which are poorly measured everywhere.

Why this is the moat

Mirror-trade reconciliation (matching what exporters and importers each report) is well-trodden. Constraining it with physical production and an independent consumption estimate is not — and it is exactly what fixes the atlas’s hardest problem: the origin gap, where a material’s true source disappears from trade because it is refined and re-exported by someone else (DR Congo’s cobalt fronted by Chinese refiners; Indonesia’s nickel; South Africa’s platinum). The anchor puts the origin back, not by assertion but by a bound the data has to respect. Two honest boundaries: (1) stage matching — the constraint holds only where the mined form is the traded form (lithium spodumene→carbonate, graphite flake→processed need the two-stage version, the next build); and (2) the consumption term only bites where the mining country is also large in the host industry. And here is the honest limit: modelled consumption is a country’s share of the driver activity at uniform intensity, so a country’s consumption share simply is its activity share. That means the bound bites for China (fluorspar, tungsten, phosphate) because China is large in every heavy industry — a fact about industrial scale, not a discovered property of these supply chains — and it cannot detect a country using more or less of a material than its industrial size implies. Everywhere else the mine leader is a small industrial consumer, so the repair rests on production alone. The Bind Δ column above says which is which, row by row — that transparency is the point, not a general claim to “obey physics.” See the consumption input →