Evidence has levels. We show which one each result has reached.
A spectral index that matches its published formula is not the same as a mineral confirmed by X-ray diffraction. Terra keeps those apart, on screen and in every report.
Seven levels, four reached.
Each level is a separate question, asked in this order. Today Terra's strongest capabilities have reached level four. Levels five to seven need laboratories, field work and reviewers outside the project.
- 1
Measurement available
The data exist and are read as published.
- 2
Numerically verified
The calculation agrees with an independent implementation or a hand calculation.
- 3
Method reproduced
A published method is implemented as published and checked against its own example.
- 4
Reference compared
The output was compared with an external reference, another product or a published map.
- 5
Independent physical evidence
XRD, microscopy or a drill intersection at the place.
- 6
External review
A qualified person outside the project reviewed it.
- 7
Predictive contribution
Better at finding deposits on independent, spatially held-out data.
The physics, interactive.
These are the absorption features the satellite lab measures. The curve is a drawing to show the idea; the band positions are the real ones.
1.76µm
Alunite
OH overtone
Advanced argillic alteration, a classic halo of epithermal systems. At Cuprite EMIT and ASTER agree on it with an AUC of 0.986.
Seen by: EMIT resolves it; ASTER sees alunite through its 2.17 µm side
What may be said, and what may not.
The claims guard is code, not a style guide. It checks every report before it prints.
What Terra can claim today
- Terra reads public satellite, geophysical and geochemical data as delivered and processes them reproducibly.
- The calculations are correct within the tested scope: 489 automated tests, hand calculations and an independent reference.
- EMIT and ASTER see the same surface pattern at Cuprite and Goldfield for alunite, Al-OH and ferric iron.
- Element results keep their limits, methods and media. Drill samples are never compared with a surface pixel.
What it will not tell you
- That a deposit has been discovered or confirmed, or that mineralisation is economic.
- That a mineral has been identified at a place. That needs XRD, microscopy or field spectroscopy.
- That rankings predict discoveries. Independent predictive improvement has not been demonstrated.
- That any capability has been externally reviewed. The Cuprite review package is ready and waiting for a reviewer.
Results that did not flatter us.
A benchmark is only worth something if it is allowed to say no.
+0.002
Magnetics at Goldfield
Adding the 2002 aeromagnetic survey to the Goldfield ranking moved the AUC by 0.002. That is no improvement, and the benchmark says so.
0.657
Transfer to Tonopah
The frozen Goldfield configuration scored an AUC of 0.657 in Tonopah, a district it never saw. Distance to known mines alone scored 0.881. Transfer was not shown, and that is the result we publish.
0.028
The control pair
At Cuprite, a sensor pair that physically should not agree was tested alongside the real ones. It does not agree, which is what makes the other numbers worth reading.
6 of 35
Samples fit for a pixel
Of 35 USGS samples at Cuprite, six passed the rules for comparing a sample with the pixel above it. The other 29 were rejected with a reason.
Six rules for ground evidence.
Written down once, enforced by the software every time.
- 1Results below a detection limit stay as intervals, never as zero or half the limit.
- 2Analytical methods, sample media and digestions are never pooled into one number.
- 3Drill samples are never compared with a surface pixel.
- 4Each sample counts once, however many reports repeat it.
- 5A pixel whose geolocation cannot be measured is not compared with anything.
- 6Words such as discovered, confirmed or economic are blocked before a report prints.
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.