Loading…
Loading…
The sources behind our critical-minerals supply-chain research — institutional reports, academic methods papers, Chinese-language policy and academic writing, national-agency intelligence, and the primary disclosures of companies that sit at supply chokepoints. A source is marked read only when it was actually read and its findings extracted, never from a search snippet; sources we tried and could not access are listed as such rather than dressed up as coverage.
39 sources · 36 read · 0 in review · 3 inaccessible · updated 2026-08-22
China's own framing leads with MINE-END weakness (Li 13.6%, Ni 4.1%, Co 1.0% of global mine output) — same data as the West, opposite narrative. Nickel ore 85.22% from the Philippines.
The standout admission: beryllium 69% import-dependent, 90% of that from the US. Calls for a 风险评估监测预警体系 (risk-assessment monitoring early-warning system) — our product, stated as China's unmet need.
Regulator-authored chain-of-custody ontology: 一物一码 item-level coding, monthly filing, procurement + sales counterparty fields — maps ~1:1 onto our named_counterparties schema.
PRIMARY TEXT READ incl. both Annex .wps files. Scope: 13 substances (Sm/Dy/Gd/Tb/Lu/Sc/Y + alloys + Dy/Tb oxides), denominator = the foreign-made listed item, not the end product. Never in force for a single day; suspended to 10 Nov 2026 with automatic revival. No incumbent tooling exists.
Compliance notification does not discharge the due-diligence duty (尽职调查); red-flag knowledge standard (明知); enforcement against third-country transshipment. Law-firm 解读 articles are the highest-yield source class for obligations firms cannot discharge.
China's structural self-diagnosis: 中间大、两头小 (big middle, small at both ends) — contradicts the Western 'end-to-end dominance' framing. Calls for a global new-energy mining big-data platform.
Stockpiling now statutory with a three-way split — product, capacity (产能储备) and in-ground (产地储备) reserves; the latter two have no US SPR analogue.
Complete operative text read: suspends the six October announcements 自即日起至2026年11月10日. Automatic revival on the plain text; extension requires a new announcement.
The CRMA's scientific basis. Six-layer value-chain schema (raw material → processed material → component → assembly → super-assembly → system) with a tradability test for the component boundary — adopted as our schema vocabulary.
Downloaded and parsed: one XLSX of material-demand scenarios. Contains NO component layer — the six-layer chain exists only as prose tables in the report annex.
JS bundle decompiled: a presentation layer over JRC132889, navigation-only data, no ingestible payload.
READ in full (372pp, via Wayback — live blob GUID rotated). Refining concentration verbatim (CN: 50% Cu / 70% Li / 75% Co / 85% magnet-RE separation / >90% battery graphite); USD 6.5T/yr at risk; EU-vs-CN price wedge (Ga/Dy/Tb ~5x, Ge ~3x, W 6x) validates our SMM lane; uranium ownership 6 firms = 88%. Data Explorer dataset free, CC BY 4.0.
READ in full (57pp, 82 companies). The strongest third-party validation of our thesis: mining firms trace only ~20% to final manufacturing; downstream cites 'limited leverage' at 3x the upstream rate; ~75% intend to increase traceability spend. An outside-in OSINT layer needs none of the supplier cooperation they lack.
READ + DATASET PRESERVED at data/external/ (86,988 measure-level rows, 2009-2024, HS6, 13 measure types, verified). Restrictions up fivefold since 2009; 70% of cobalt/manganese exports under at least one restriction; India among top-5 implementers. Directly ingestible into IPTM.
READ (via Wayback). CORRECTION to our phase-1 notes: contains NO company-level ownership figures — ownership is qualitative only; its numbers are geographic (top-3 countries: Li 92%, REE 90%, Co 78%). The quantified ownership analysis in the canon is in the IEA CMO, not here.
READ in full. Mostly derivative of IEA 2023; corporate concentration flagged but unquantified. Lowest value of the canon set.
The 'Substitutes:' field verified: rich and application-specific for gallium/cobalt, one useless sentence for rare earths. PDF-only — the field exists in no structured release.
Downloaded and INGESTED: country-level production/capacity/reserves, 77 commodities, CC0. Now feeds data/intelligence/usgs-production-shares.json (17 materials with shares + HHI).
READ in full text: equilibrium displacement + constrained IO optimization (Clarabel/CVXPY) + ML restriction probabilities (17 China scenarios set to 100%), SPOF override, short-run elasticities extracted. Stops at 402-industry level — the company/node layer is open.
READ (PDF is watermarked DRAFT). Substitutability = 44 ordinal scores (22 materials x 2 horizons), per-material not per-application, prose-locked. A per-application layer with qualification times would exceed the federal state of the art.
Closest free platform: country/HS level, Comtrade 2017-2022 only, no company/facility layer. Self-declares the gaps our corpus fills ('HS classification... provides a poor match', country data 'fragmented and incomplete at best').
USGS = economy-wide/historical, DOE = energy-specific. Read via reader proxy (direct PDFs 403). Confirms 2025 list adds incl. lead.
READ in full (106pp). Build-prune LLM ensemble, 10x per good, embedding-matched to HS6 (LLMs hallucinate codes). 5,633 nodes / 980,018 edges, avg degree 348 — recall-heavy. NO per-edge error rate reported; validation is industry-level correlational (0.208 SD). Verdict: binary adjacency prior ONLY, never intensity/BOM — and the conservative prune biases against exactly the exotic critical-material edges we care about.
Hard paywall (403 direct and via reader proxy; no preprint found). Its 'network index of criticality' construction remains unverified — do not cite from summaries.
READ in full. Recipe verified and reusable (HS-sequence tracing, role classification, terminal-node transparency typology, reachability Monte Carlo — one trader's removal collapsed resilience 0.80->0.15). THE CATCH: it rests on paid shipment data with shipper/consignee names — incompatible with zero budget; our OSINT substitution replaces that layer. Their 34% non-transparent share is a usable benchmark.
Abstract/highlights read; FULL TEXT BLOCKED (ScienceDirect captcha on every route despite CC-BY). LP + I-O model over 56 commodities' end-use data; mica/lithium/fluorspar worst. The LP formulation itself remains unread.
Abstract/highlights read; BODY BLOCKED. The whitespace confirmation verbatim: 'resilience strategies mainly target upstream disruptions... while midstream and downstream vulnerabilities remain underexplored' — our downstream/company-level thesis confirmed across a 327-study PRISMA sample.
READ in full. Mixed-frequency SVAR, 6 base metals, all-free data. Transmission demonstrated = shock -> commodity price/volatility, NOT equity — consistent with our null equity backtest: the strong channel is the commodity layer; the equity layer is where the signal dies.
READ in full. Formula extracted: Spr = avg[extraction share, processing share, US import share] x (100-FreedomHouse)/100. All-public data; validated out-of-sample (Bi/Mg/Sb Tier-1 BEFORE China's 2024 controls hit exactly those three). The cheapest benchmark to compute against our scores.
READ (German, Rohstoffliste 2023). HHI x weighted-country-risk grid with citable thresholds (1,500/2,500; GLR ±0.5) worth adopting with attribution; 221 traded intermediate products by HS code; 140 products (46%) high-risk; China #1 producer for 90% of refining products. Monthly Preismonitor incl. recycling-material prices.
READ. World Mineral Production: the only continuous series back to 1913, free Excel/JSON from 1970 — but commercial use requires NERC/BGS permission (verbatim terms captured). UK methodology adds companion-metal fraction as a supply-risk indicator. Flag licence before any ingest.
READ (Japanese). Stockpile: 34 mineral groups/55 elements, 60 days (42 national + 18 private) — and the reform made actual targets/volumes explicitly NON-PUBLIC on economic-security grounds, so Japan's buffer is now inference not lookup. Material Flow series = the model for demand-side national flow maps (compound-level tonnage).
THE actionable find of the national-agency pass: facility-level (smelters, refineries, mines, advanced projects, 34 CMs), updated annually, Open Government Licence Canada — commercial use OK. The cleanest openly-licensed facility feed reviewed; directly ingestible into node-registry.
READ: yttrium 333t->17t to the US; EU imports rebounded, US never did; Lynas first ex-China Dy oxide May 2025; US $110/kg price floor; DoD bars CN/RU/IR/KP rare earths by 1 Jan 2027 — a second dated compliance obligation.
The single best business-model source found: qualification-vs-contracted distinction, feedstock (not capacity) as binding constraint, equipment supplier-development pain — with the caveat that the equipment quote is hedged ('sometimes').
A four-smelter major states its supplier due diligence bottoms out at 'publicly available information' — 4 on-site supplier assessments per year. The labour-replacement pitch, verbatim.
1.1 Mt/yr feed, $6.6bn capex, ops 2029; 12 of 60 US critical minerals produced today, 14 by 2028.
Falsifies the 'no Western kiln supply' claim: 110 m cracking kiln, ~€13m, described as the plant's longest-lead-time equipment (~24 months order to commissioning).
The one genuinely concentrated equipment class is Japanese, not Chinese: ULVAC claims >70% of continuous magnet-furnace lines, >24-month delivery, new plant builds ~12 units/yr.