We are rewriting this page so that every capability described on it is one the platform demonstrably has, and every figure on it can be traced to a source you can check.
The previous version of this page described detection hardware and survey aircraft that GeoMetals does not operate. It has been withdrawn rather than reworded. A summary of what the platform actually does today is below.
GeoMetals is a data and analysis layer over published mineral-resource information. It compiles, cross-references and visualises material that is already in the public record. Everything listed below is running code you can exercise against the public API.
A register of critical-mineral sites — rare earths, copper, lithium, cobalt, nickel, graphite and others — compiled from USGS MRDS and ICMM. Each record carries location, country, commodity, status and the source it came from. It does not carry grade, tonnage, priority or confidence: those sources do not publish them for these records, so the fields are null rather than estimated, and the API refuses to render them as measurements.
We run no prospectivity model, so the register contains no generated records at all — every row traces to a published source. If that ever changes, model output will be stored and displayed as a prediction, never as a discovery, and counted separately.
Ingestion from named, publicly accessible sources, so that any figure we show can be traced back to the body that published it.
Row counts read from /api/metrics on 20 August 2026, except the 5,013 / 177 split of the 5,190 compiled records, which is counted in the compiled source file itself — /api/metrics reports the compiled set only as a single seed_deposits total. The 18 tracked tickers come from the separate stock endpoint, not from /api/metrics. A SEDAR+ scraper also runs but has stored no filings to date, so no SEDAR+ figure appears anywhere on the platform. We do not ingest BGS World Mineral Statistics, Geoscience Australia, ASX JORC reports or Sentinel Hub imagery — an earlier version of this page listed all four, and none of them were ever connected.
A caching proxy over public Terrarium elevation tiles, used to build elevation profiles and mountain cross-sections at deposit locations. Includes disk caching, concurrent tile fetching, upstream rate limiting and retry with backoff. The elevation data is third-party open data; we render and cache it, we do not measure it ourselves.
A ternary verifier grades a statement TRUE, FALSE or UNKNOWN against a fixed table of twelve hand-written rules, and it returns UNKNOWN rather than guessing. It checks the internal consistency of what the platform displays. It does not confirm anything about the ground, and it is not a compliance review.
/api/halluc publishes the composition of that rule table — 8 true, 2 false, 2 unknown, i.e. 66.67% / 16.67% / 16.67%. That is the fixed makeup of twelve rules, not a measured error rate, not an accuracy score, and it does not move. We publish no measured accuracy figure because we have not measured one.
Read from the platform's public /api/stats endpoint on 20 August 2026 — the field names above are the response keys (total_deposits = seed_records + live_records), so you can check each figure yourself. These are record counts, not confirmed resources, and the live_records figure changes whenever the MRDS ingest worker next runs; it stood at 2,000 on that date. /api/stats deliberately reports no priority or confidence aggregate at all, because no such score exists on these records. The platform serves no model predictions: there is no prospectivity model, so nothing is published as a predicted deposit.
An earlier version of this page asserted the following. None of it is supported by anything the company has built or measured, so it has been removed rather than softened.
We will walk you through the data model, the sources behind every record, and the limits of what the platform can currently tell you.