Method

How we make an unverifiable claim testable.

A junior cannot verify that a target is good without spending a hundred and fifty thousand dollars drilling it. Faced with unverifiable claims from an unknown supplier, the rational buyer does nothing. That is an adverse-selection problem, not a marketing problem, and no amount of posting fixes it. So we manufacture verifiability before we sell anything.

The protocol

Blind retrospective validation

Mask. Remove every input that could leak the answer — mine locations, occurrence databases, producer-held tenements, any report that names a discovery. The exclusions are documented precisely, because this is the step where most validations quietly cheat.
Run cold. Execute the pipeline with no knowledge of where the deposits are.
Freeze and timestamp. Publish a cryptographic timestamp of the output before unmasking. No retro-fitting is possible after that point.
Unmask and score. How much of the known endowment falls inside the top decile of ranked area, and what is the enrichment over random?
Publish the misses. Including the deposits we failed to find and the false positives that turned out to be salt pans.

The sentence this produces

“The top five per cent of ranked area captured six of nine known deposits — a twelvefold enrichment over random.”

That is a sentence an exploration manager can take to a board. It costs nothing but three weeks and intellectual honesty. Three validations — one gold belt, one lithium pegmatite field, one copper system — are completed before a single sale is attempted.

The pipeline

Seven stages, all of them arguable.

Acquire

Every input is public, free and licensed for commercial use. Sentinel-1 and 2, Landsat, ASTER, free spaceborne hyperspectral, government magnetics, radiometrics and gravity, national geology and geochemistry, the open-file exploration archive, and the mining cadastre. The only paid data is client-funded and bought as a pass-through, after free data is genuinely exhausted.

Each dataset is logged in a licence register with its commercial-use and redistribution status before it is used in any paid deliverable.

Lead with geophysics

Airborne magnetics maps structure, intrusions and magnetite-destructive alteration straight through the regolith that defeats optical sensors. Radiometrics measures surface geochemistry directly rather than inferring it from reflectance. In Australia both are free, statewide and excellent. They are also the layers experienced geologists trust, which matters commercially as much as technically.

Most low-cost providers stop at Sentinel-2. That is the single clearest difference between this method and theirs.

Confirm with satellite

Sentinel-2 SWIR ratios, ASTER thermal bands and free hyperspectral map clay, iron-oxide, carbonate and silica alteration. Seasonal stacking suppresses the false positives — dry lakes, borrow pits, tailings, agricultural land — that make satellite-only targeting a red flag to anyone with a budget.

Satellite is one evidence layer of seven here, weighted at fourteen per cent. Leading with it marks you as an amateur.

Fuse and rank

Evidence layers are combined under weights calibrated against known deposits in that specific belt, never generic global weights. The output is a continuous prospectivity surface and then a discrete ranked target set, with a documented score decomposition you can argue with.

One config file, one run hash. A client can re-run the exact configuration that produced their register.

Screen for tenure

A target on ground you cannot get is worth nothing. Every target is intersected with granted tenements, pending applications, protected areas from openly-licensed national datasets and other statutory exclusions, with a buffer for positional error.

We do not use the World Database on Protected Areas, because its licence prohibits commercial use. Open national datasets replace it and are higher resolution.

Screen for mineability

The layer no other low-cost targeting provider sells. Every surviving target is scored on whether it could ever become an operation. Two targets with identical geological scores can differ tenfold in expected value once this runs.

This is where a mining engineer is worth more than a geologist. Geologists find mineralisation; mining engineers find mines.

Try to destroy it

Before delivery we spend an hour trying to break our own top three targets. Whatever survives goes in. Whatever does not is demoted, with the reasoning written into the dossier under a mandatory field: the strongest argument against this target.

Clients find that field more persuasive than anything else in the document. So do we.

Evidence weights

Published, not hidden.

Default weighting for an orogenic-gold model. Calibrated per belt against known deposits before any target is scored, and never adjusted afterwards to flatter a result.

Layer	Weight	Source
Structural architecture	25%	Airborne magnetics + SAR + DEM fusion
Radiometric alteration	18%	K/Th ratio from government radiometrics
Lithological favourability	15%	Government geology + magnetic domains
Spectral alteration	14%	Sentinel-2, ASTER, EMIT / EnMAP / PRISMA
Geochemical anomaly	12%	National stream sediment, soil and laterite
Metallogenic context	16%	Known-occurrence databases, permissive tracts

Mineability screen

Could it ever be mined?

Eight factors, each scored minus one, zero or plus one, applied as a bounded multiplier between 0.4 and 1.2. Deliberately bounded: the screen re-ranks, it never invents value.

Depth to target vs likely mining method
Terrain and haulage
Process water availability
Grid and power proximity
Strip-ratio proxy
Land use and permitting friction
Distance to an existing plant
Obtainability of the ground