For each metric on responsible gold, the full spread of published readings: each one carries its source, its reference year and how it was produced.
Most headline numbers about gold descend from a handful of origins. This page shows the spread instead of crowning a winner.
Contested counts publish as attributed ranges, with the floor source and the ceiling source named. Self-reported figures carry "per X, unaudited" and never enter headline stats.
The year shown is the year a figure refers to, not the year it was reprinted; re-citations are labelled as re-citations.
Each metric leads with the weakest method among its readings: honesty leads worst-case.
Shares, undeclared production and trade flows, mostly modelled from trade-gap analysis and single origins.
Almost no spread, and that is the finding: one 2017 IGF estimate has been recirculated for nearly a decade by the LBMA, the WGC and the World Bank without re-derivation. The credible range is 15-20%, vintage 2017-2019, and because ASGM production is unregistered by definition, every reading here is modelled from the same single origin.
Trade-mirror estimation carries bounds by construction: you infer what was never declared by comparing what importing countries record against what exporting countries admit. The 321-474 t band is the method’s honest width, from a single producer (Swissaid) whose method has been widely checked, and there is no registry to verify it against.
Livelihood counts in artisanal and small-scale gold mining, with the all-minerals figure shown as a labelled out-of-scope reading.
The gold-only estimates (around 10M versus 15-20M) differ on vintage and modelling, and none rests on a census. But the bigger hazard is scope: the widely quoted 45M is all-minerals ASM (World Bank 2024, Delve 2020 lineage), shown here as a labelled out-of-scope reading so it stops migrating into gold copy.
The registry- and market-observed layer: vault holdings, price records and demand estimates, with the estimator named.
These readings do not disagree; they count different things: one year of all imports versus eleven years of African imports. Both trace to a single producer, Swissaid, working from mirror trade data, which is why the hub carries them attributed rather than as settled fact.
Both readings come from the party being measured: GoldBod is buyer, exporter and scorekeeper at once. The independent record pushes back: a MyJoyOnline analysis argues the financing structure left a GH¢9bn hole at the Bank of Ghana, and critics warn monopoly pricing invites smuggling. Revenue booked is not value created; the spread between the claim and the critique is the story.
No disagreement to explain: these are the best-measured numbers on the page, one from the LBMA’s monthly vault registry, one from the traded market. They sit here as the calibration point, what "assayed" looks like, against which every modelled and asserted figure above should be read.
Register counts and certified volumes, including the counts that disagree on the schemes’ own pages.
Certification lapses plus stale counters on the scheme’s own web properties. Per this hub’s policy, a contested count publishes as an attributed range, floor and ceiling both named and dated, never as one number.
These are different instruments measured in the same unit: a certification label (Fairmined), fair-trade terms (Fairtrade), an escalator sourcing model (Swiss Better Gold) and a state monopsony (GoldBod). The three-orders-of-magnitude gap is structural, not a dispute. And every reading is party-reported and unaudited, so each carries its attribution on its face.
Two registries counting two different things: the RMI list absorbs LBMA and RJC audits through cross-recognition, so it is larger by construction, while the gold GDL counts only LBMA-accredited gold refiners. Add the "100+" gold-plus-silver framing and the same industry yields three defensible counts.
Every reading traces to a named source in the July 2026 crawl. Readings are never averaged and never ranked; where a spread exists, the spread is the finding.
Reviewed 2 July 2026.
The ASGM share of world gold traces to one 2017 estimate; the certified-organisation counts disagree on the scheme’s own pages. Treat every unsourced gold statistic as a citation waiting to be checked.