Four areas, every number, every limit.

These are the areas the workbench has been tested on. The figures come straight from its records, and each case ends with what it does not show.

Cuprite, Nevada, USA

Two satellites, twenty years apart, see the same rock.

0.986

AUC, EMIT alunite against ASTER alunite index

EMIT (2023) and ASTER (2003) were compared pixel by pixel over 32,494 pixels. Alunite, Al-OH and ferric iron agree. A control pair that should not agree does not.

Al-OH agreement (AUC): 0.787
Ferric iron agreement (AUC): 0.806
Control pair, no agreement as expected: 0.028
USGS samples, 6 fit for a pixel comparison: 35

Where it stops. Agreement between two sensors is not a mineral identification. No laboratory mineralogy is stored for Cuprite yet.

Goldfield, Nevada, USA

An honest benchmark beats a flattering one.

+0.002

AUC gained by adding the 2002 aeromagnetic survey

223 samples and 15,388 determinations, none counted twice. Ferric oxide from ASTER is weakly associated with measured iron. Magnetics did not improve the ranking, and Terra says so.

Samples: 223
Determinations: 15,388
Censored Kendall tau, ferric oxide against iron: 0.224
Independent pixels in that test: 41

Where it stops. Association is not cause. Rankings at Goldfield stay experimental.

Kalgoorlie, Western Australia

A million assays, handled like a million assays.

1,133,885

Determinations from GSWA and company surface samples

80,178 samples, imported without downloading the statewide database. Gold results without a method stay in their own group, and no group mixes surface and drill samples.

Samples: 80,178
Largest difference to an independent reference: 4.83e-10
Gold groups mixing surface and drill core: 0
Censored gold statistics over the whole area: 4.0 s

Where it stops. No satellite analysis has been run at Kalgoorlie yet, so no pixel comparison is made there.

Wadi Ham, Fujairah, UAE

Copper gossans in ophiolite, marked experimental.

Landsat scenes in the copper reconnaissance run

A Landsat workflow for gossans over the Semail ophiolite. Public exploration records for the UAE are thin, and Terra reports that coverage as insufficient instead of guessing.

Area at 30 m: 361 km²
Processing time: 3 min 38 s

Where it stops. Experimental model, no ground truth in the workbench.

Bring us the ground you are thinking about.

Name an area. We set up a pilot workspace with a first investigation and walk through it with you.

Request access