STRATA · Methodology
Strata observes the world's mineral supply chains from primary sources, estimates where demand comes from - and always tells you which of the two you are looking at. This page walks the method for copper, nickel, and lithium, and how it couples to GCAM. Every figure below is read live from the same data the product serves.
Scope & limits
01 · The lens
Strata maps every commodity onto one eight-stage scaffold, ore to end use, so any two are directly comparable. But a real commodity forks - some routes skip stages - so we read this scaffold three ways: what the stages are, how this commodity actually travels them, and how much of each we cover.
One scaffold, every commodity - so any two line up
Strata maps every commodity onto the same eight stages, ore to end use. That is what makes copper, nickel and lithium directly comparable - but a real commodity rarely walks all eight in a straight line. It forks, and some routes skip stages. The primer decodes the scaffold; the journey and routes below show how this one actually travels it.
How nickel is made - inputs, outputs & routes
The metal flows left to right. At each stage something is added (from above) and something leaves (below). The three tracks show which stages each route uses - and which it skips.
Hover a stage to read what happens there - click to pin it open. Teal = added at that step, gold = leaves at that step; the coloured tracks are the three routes.
Recycling loops old scrap back into the chain - it re-enters downstream, it is not a new-metal stage. Grades are indicative; sources are linked below.
Nickel does not travel one line - it splits early into two ore families that take different routes and rejoin at the end. Sulfide ore is concentrated, smelted, and refined to pure metal; laterite ore (~70% of world supply today) is either smelted straight into stainless-grade metal or acid-leached for battery chemicals. Below is the whole scaffold; the forks show which stages each route actually uses.
Nickel is dug out of the ground as one of two very different rocks: hard sulfide ore, or soft, near-surface laterite (weathered tropical soil).
Sulfide ore is crushed and floated in water so the nickel-rich grains bubble to the top, upgrading it ~10x. Laterite has no separable nickel grains, so it cannot be concentrated - it goes forward whole.
The feed is melted at ~1,400-1,600 C. Sulfide concentrate becomes a nickel 'matte'; laterite becomes either ferronickel/nickel pig iron (an iron-nickel alloy that goes straight to stainless) or a matte.
Matte is electro-refined or gas-refined into pure Class-1 nickel metal; separately, iron-rich limonite laterite is dissolved in hot acid (HPAL) to pull out a nickel-cobalt hydroxide for the battery chain.
For batteries, nickel (as MHP or dissolved metal) is turned into nickel sulfate crystals - the exact salt a battery cathode plant needs.
Nickel is built into things: melted with iron and chromium into stainless steel, co-precipitated into battery cathode powder, or alloyed into jet-engine superalloys.
Old stainless steel and spent batteries are collected and re-melted or leached, returning their nickel to the chain instead of mining new metal. This box is dashed because it feeds back in, it does not add new metal.
Finished nickel ends up in things people buy - kitchen sinks and buildings (stainless), electric-car batteries, and aircraft engines.
grades & routes sourced from: USGS Mineral Commodity Summaries 2025 - Nickel · Nickel Institute - Nickel industry (processing series, Parts 1-6) · INSG - The World Nickel Factbook 2024 · worldstainless - The Global Life Cycle of Stainless Steels · LME - Primary Nickel special contract rules (99.80% Class 1)
The same chain, sized by how much of each stage we map
Band height is facilities strata maps; a stage figure links to its source. Full value chain (companies + control) →
Coverage across commodities - covered, partial, or gap
| commodity | Extraction | Refining | Fabrication (semis) | Use | Recycling |
|---|---|---|---|---|---|
| copper | covered | covered | covered | covered | covered |
| nickel | covered | covered | partial | covered | covered |
| lithium | covered | covered | gap | covered | partial |
hover a chip for the stage headline · a "gap" chip is the product stating where its own coverage ends - not a blank we hide
02 · The membrane
Two kinds of numbers exist in strata, and they are visually and structurally distinct everywhere they appear.
Reported by an authoritative source - USGS, BGS, ICSG, a company filing, a national agency. The row carries its citation; click it and you land on the publication.
Computed as observed activity × a verified material intensity. Always marked imputed, always shows BOTH ingredient sources, and expands to the full calculation on click.
The rule that holds the product together: a reader can always tell which of the two they are looking at - and when the data cannot support a number, we show nothing rather than a wrong number. Every observed row links to its publication; the full catalogue, ranked by authority, lives at /sources.
03 · The landscape
Every figure below is counted from the shipped data, not typed into this page - so it cannot drift from what the rest of the site serves. The evidence split sits beside the totals on purpose: a headline count without it is the overclaim the membrane above warns about.
The life cycle from the plant floor - facility rows by value-chain stage
These are facility x commodity ROWS at a stage, not distinct plants - one plant serving two commodities is two rows, and a mine-plus-smelter complex is counted at both stages, so nothing here may be added. 52 of 65 commodities carry a facility layer of their own; 13 rare-earth elements inherit the shared rare_earths set and are excluded from these counts. Assembled from 11external registers - several are mineral-title cadastres or deposit catalogues rather than plant registers - plus strata's own derivation layer; most of the 470,051 rows carry no stage signal.
How strong is the evidence behind a row
These partition all 470,051 records exactly, strongest first. Only Observed and Corroborated are load-bearing: 15,437 (3.3%) seen in a primary register or agreed by two independent ones. The rest is inventory - largely a 2016 USGS MRDS snapshot, which records that a deposit exists, not that a plant runs. These are facility x COMMODITY rows, so a polymetallic mine counts once per commodity and this is not a count of distinct plants.
Where it comes from - one mark per source
Sorted strongest evidence first, so each row shows not just how many sources feed an axis but what kind. Supply rests almost entirely on official surveys; environmental does not, and saying so is the point.
194 carded sources · 119 live fetchers, 74 extracted seeds · 69 record types · 32 distinct USGS products alone · 16 declined on licence
04 · Supply, observed
Sources publish production per chain stage - mine, smelter, refined. The same metal appears at every stage, so summing stages manufactures phantom supply. Strata keeps each stage as its own panel and never sums across them.
How do we know a row's stage? We don't infer it - the source reports it.
USGS and BGS publish production as separate statistical series per stage. Strata preserves each row's verbatim wording and only GROUPS that vocabulary into stage panels:
A row whose label fits no family lands in "Other forms" rather than being force-classified, and every row keeps its provenance - so any grouping is auditable back to the source line. That is the difference between a data-integrity rule and a model assumption. A free bonus: nickel's Class-1 vs Class-2 split (refined metal vs NPI / ferronickel) simply falls out of the panels, because the sources were reporting it all along.
05 · Reading the concentration bar
Every screen carries an HHI concentration bar - the product's headline supply-risk number. It is not a black box: it is the sum of squared supply shares, computed separately at each stage, and we show the shares that build it.
The concentration bar on every screen is a single number - the Herfindahl-Hirschman Index (HHI): take each supplier's share of the total, square it, and add them up. It runs 0 (many equal suppliers) to 1 (one supplier owns everything). Squaring is the whole point - it makes a dominant producer count far more than several small ones. We show the shares, so you can rebuild the number yourself.
the top five shares alone already sum to 0.52; the dominant producer's square carries most of the index - which is exactly what a concentration measure should do
The bands
The ladder runs green to amber and stops there - an index is a description, not an alarm, so there is no red. A missing bar means not enough suppliers disclosed to compute it, never "low".
HHI of which stage?
Because a metal's mine geography is not its refinery geography, we compute HHI separately at each stage and never blend them. The mine map and the refinery map are different maps - and the gap between them is the single most useful supply-risk signal on the page.
copper is the lesson: mining is spread across the Americas and Africa (diversified), but refining routes through China - the risk lives downstream, and only a per-stage HHI shows it. "% attributed" on the live rows sizes how much of the total the named suppliers cover; a "Rest of world" residual is kept in production sums but excluded from the index, so it can neither dilute nor inflate the concentration number.
06 · The resource base
Production is a yearly rate; beneath it sit the reserves and mapped deposits that decide whether supply can scale to the transition's build-out.
The eight-stage scaffold starts at the mine; beneath it sits the resource base that decides long-run adequacy - economic reserves in the ground, and the wider inventory of mapped deposits. It is where "can supply scale to meet the build-out?" is actually answered.
million tonnes contained
A deposit count is not a supply map
Of 47,988 mapped deposits, 79% sit in United States - not because it holds the metal, but because the USGS MRDS inventory is exhaustive at home and sparser abroad. So we label the count as "how well mapped" and read the true supply leader (Chile) from reserves and production instead. It is the same rule as "a census share is not a world share", one layer deeper.
07 · Secondary supply
Recycling is a second source of metal - large for a mature base metal, negligible for a young battery metal - and it must never be summed into primary mine production.
The life-cycle infographic shows recycling as a stage; here is how much it actually provides, and why it stays on separate books. Recycled metal re-enters the chain downstream, so it must never be added to primary mine production - that would be the phantom-supply error again. How much a metal recycles is a first-order lever on the demand-vs-supply gap, and it varies enormously by metal.
about a third of copper supply is recycled input; most US scrap (68%) is exported
alloy and stainless scrap recirculates at base-metal rates, shown as a rate not a share
almost nothing is recovered yet - the single biggest lever on future lithium supply
End-of-life recycling rates come from Yale STAFDB; scrap flows and recycled-input share from USGS. The contrast is the point: a mature base metal recovers most of its scrap, while a young battery metal recovers almost none yet - which is exactly the lever a transition scenario tests.
08 · Facilities, not just countries
Country totals answer how much; the censuses answer which plants, owned by whom. Counts and China shares below are live from the plant censuses (operating facilities; shares are of the censused fleet, never presented as world shares).
plus an 84-mill fabrication census (wire rod, tube, sheet) with ownership resolved
plus the 98-plant Indonesia smelter census (CGS): 86% of operating Ni-equivalent capacity is China-linked
the converter census is the chain's chokepoint view - HHI withheld until enough plants disclose capacity
ITA/USGS 3TG public-domain roster (the license-clean alternative to LBMA): 266 refiners, incl. the 4 Swiss majors the roster omits; per-refiner capacity withheld (no free source)
09 · From operator to ultimate owner
A census names the operator; resolution names the ultimate parent behind it - and weighs control by output, not just plant count.
The census names the plant; resolution names who ultimately controls it. Each operator is walked up its ownership chain - SEC EX-21 subsidiary lists and the CorpWatch control graph - to an ultimate parent, its home jurisdiction, and a control bloc. Blocs are relational: the same company is "Domestic" at home and "foreign" abroad, so control is measured against the asset's own country, not in the absolute.
Count the plants, or weigh the output?
A share by plant count and a share by capacity are different numbers, and the gap is the story. Across nickel's 19 resolved operations:
a minority of the plants, a near-majority of the metal - a few large Chinese-owned operations carry far more output than their headcount suggests. Only capacity-weighting sees it.
Output by control bloc
10 · Trade
Metal has to physically reach the user, through specific partners, exposed to specific measures - a supply-risk layer a production map never shows.
Metal is mined in one place and consumed in another, and the movement between is its own risk layer - the part a "who mines it" map never shows. Strata reads it from three public sources: USGS for net-import reliance, CEPII BACI bilateral flows for who ships to whom, and the OECD inventory for the measures that can choke a flow.
US demand met by imports (2025, USGS)
Copper content of blister and anodes - single-partner dependence is a trade-partner HHI of its own
largest global shipper by value (BACI); 15 export measures in force, e.g. Argentina export tax
11 · The screening risk score
Concentration, dependency, control, restrictions and governance roll into a single 0 to 100 screening score - transparent in its parts, and honest when an input is missing.
Everything above - concentration, import dependency, who controls supply, export restrictions, governance - rolls into one 0 to 100 screening score. It is transparent (every component and its weight is shown) and honest when it cannot compute: a missing input reweights the rest rather than counting as zero, and a missing score means "not enough inputs", never "low risk".
The weights are fixed and published, not fit to an outcome. As a sanity check, the score is cross-referenced against the USGS supply-risk benchmark (OFR 2025-1047), where this commodity ranks 44 of 84 (Moderate). The bands run green to amber; there is no red.
12 · Demand, estimated honestly
Take something you can count - tonnes of stainless melted, GWh of cells built, km of grid line added - and multiply by how much metal each unit needs. Every coefficient is independently verified against its publication before use.
The coverage meter - how much of world demand our lanes explain
the remainder is stated, not hidden - copper's meter is low because grid and building stocks are tracked as stocks, which are never summed with annual flows
13 · Supply meets demand
Sections 03 and 05 built the two halves - what the world makes, and what the transition needs. Here they meet: each implied-demand lane as a share of the world supply it draws on.
Supply was observed; demand was estimated. Here the two halves finally meet: divide an implied-demand lane by the world supply it draws on, and you get the one number a transition analyst actually wants - is this end use a rounding error, or a claim on a large slice of the world's metal?
This is modeled demand over reported world supply - an estimate divided by an observation - so it inherits the estimate's caveat and always reads with a "≈". The denominator matters: an annual-flow lane (stainless melted this year) is measured against annual production, while a stock lane (copper standing in the grid) is measured against reserves. Comparing a stock to a flow is the phantom-supply error in reverse, so we never do it - which is why the demand meter earlier keeps stocks and flows on separate books.
14 · The GCAM relationship
GCAM projects the energy transition but models no critical minerals. Strata turns GCAM's build-out into mineral demand - and exports the whole layer in GCAM's own input format, with both sources attached to every row.
15 · The GCAM-CMM contract
GCAM-CMM is a different consumer from GCAM proper: it wants graded supply curves, not a demand narrative. Strata emits that bundle in the contract's own A-file shape, sized to each file rather than to one roster - the curve covers 11 of the 13 minerals their v8.2 vocabulary spans, while the intensity, price and raw-supply tables run wider (23 to 64 commodities) so nothing they add later is missing.
What one tonne of "available" is, per commodity - 6 of 13 are plain contained metal
Amber means the quantity axis is NOT plain contained metal, and A10.minerals_info states which in that commodity's own row. A blanket basis claim would be worse than silence, because a consumer reconciles against it.
Inside the bundle, per commodity
Each curve is tonnes sorted by extraction cost - the staircase a consumer integrates. Shading separates proven reserves from the resource tail, because that boundary is where the endowment stops being measured.
Measured grade models, and the only undiscovered tier.
contained metal.
tonnage USGS Mineral Commodity Summaries 2026·cost strata a11-cost-benchmarks bundle
A11.curves
65 rows
A11.vintage
117 rows
MCS_reserves
14 rows
MCS_production
14 rows
quota XML
emitted, 7.2 gen
# File: A11.minerals_curves.csv# Units: available: Mt, material basis PER COMMODITY - see A10.minerals_info.resource_basis. Contained metal for most; manganese is contained Mn CONVERTED from USGS's gross-weight ore with cited per-country grades; rare_earths and the neodymium vintage are REO / Nd2O3 equivalent, NOT metal; steel is contained Fe; platinum is the 6-metal PGM basket. extractioncost: 1975 USD/kg of that same basis# Source: Quantity + region: USGS MCS reserves (unit-normalized, GCAM-32 rollup; material basis per commodity in A10.minerals_info.resource_basis). Cost: strata benchmark C1 model (2024 USD, GDP-deflated to 1975 USD; route/class merit order for Li/Ni). Grade-tier split: USGS grade-tonnage model (copper) else a fixed screening split. Uranium: NEA/IAEA Red Book cost-of-recovery tiers (global, authoritative). Public domain / intergovernmental. Per-commodity lineage in A10.minerals_info.region_GCAM32,commodity,subresource,grade,available,extractioncostAfrica_Southern,copper,copper,grade 1,11.55,0.36Africa_Southern,copper,copper,grade 2,7.56,0.972Africa_Southern,copper,copper,grade 3,1.89,1.584Africa_Southern,copper,copper,grade 4,14.1781,2.8512... 61 more copper rows
real rows, 2026-07-28 bundle
Two we do not curve, and why
What we say about it, unprompted
16 · GCAM as a service
PNNL's unmodified GCAM 8.8 answers inside the product. A single GCAM run is a full global solve that takes minutes to hours, so it cannot be run on demand. Instead we solve it ahead of time across its own scenario axes, serve those results instantly, interpolate for the space between solved points, and fall back to the live model when a request lands off the grid. A question outside what the model can stand behind returns a clean, coded refusal - as data, never a guess.
The scenario axes we drive
These are GCAM's own scenario controls - the service steers the model through them directly, so every request maps to a real, documented GCAM configuration rather than a paraphrase of one.
Why six? GCAM exposes thousands of parameters, so an axis has to earn its place: each is dynamically verified to actually move GCAM's outputs, and must map to a policy GCAM 8.8 can model - not one we fabricate. Seven clear that bar (a stop-year axis is held for horizon sweeps) and these six are swept here. Four more candidate policy levers are held back until their full config chains are wired up, because standalone they no-op or rail the carbon market.
How those dials become worlds
A solved world is one value picked on every dial at once. Follow the green thread: it commits to a single setting on each of the six dials, and that set of picks is one full GCAM run.
Multiply the menus and there are 3,150 possible combinations. But three of the dials - carbon price, forcing target and net-zero cap - are alternative steering wheels for the same climate, and a world rarely turns more than one at once. Crossing them makes most combinations contradict, so the space prunes to 165 valid worlds, every one solved ahead of time and served.
the bake · run once, offline
A grid of pre-solved GCAM runs
165 physically-validated GCAM runs. Hover any point to see the scenario it holds - each solid point is a whole GCAM world (energy, industry, land and every gas, not just carbon), and strata reads the build-out from it. Every combination of the six axes, minus the ones that contradict (two knobs cannot both set the carbon price), solved once on a cloud batch; the faint points were pruned before solving.
the serve · every request, in milliseconds
How a question finds its answer on the grid
A typed request is a point in that space. Where it lands decides how it is answered - always exactly, or a clean refusal.
Exact match - the pre-solved run is returned straight from cache.
example SSP2 at a $50/tCO2 carbon price - one of the 165 runs, already solved.
A surrogate trained on the grid (299 models) interpolates from the neighbours; weak fits defer to live.
example the same request at $49/tCO2 - it sits between the solved $38 and $50 points.
No confident neighbours - fall back to the real, unmodified gcam.exe and solve it fresh.
example a combination the bake never pre-solved, too far from any solved point to trust.
Not a question the model can stand behind - a coded refusal with a suggested alternative, as data.
example a $150/tCO2 carbon price - beyond the $100 the service covers.
the scale of it
Each world is one solved GCAM scenario - a single setting of the six dials.
The bulk of each one is emissions and land carbon, not carbon price - about 98 MB per world, roughly 16 GB in all.
Explore every one of the 165 worlds - the full scenario catalog, filterable by driver and SSP.
17 · The rules
194 cataloged sources; each figure is traceable to its publication - see /sources.
Wood Mackenzie, Benchmark, S&P are cite-only and flagged as procurement targets.
Yale-STAF material flows and 2008-vintage footprints render with a historical flag.
Plant-census statistics say so explicitly; world shares come from IEA / USGS series.
