Openness is the product
How we validate
Any map can look convincing. We test ours against evidence it never saw, and we publish the misses as well as the hits.
The test
We rank every 30 m cell in the pilot watershed (650 km² around Alma and Fairplay, Colorado). Then we check where independent evidence falls in that ranking: claims people are paying to hold, operations registered with the Mine Safety and Health Administration, mines named in old newspapers and facts from historic USGS reports.
AUC is the chance that a random piece of evidence ranks above a random piece of background ground. 0.5 is a coin flip; 1.0 is perfect.
Results, October 2026
| Test | Result | Verdict |
|---|---|---|
| Active lode claims (not used in the model) | AUC 0.83; 3.4× expected in top 10% | Strong |
| Federally registered mining operations (MSHA) | 6 of 8 in the top 7% of ground | Strong |
| Historic mines within the mineralized range | AUC 0.76 | Good (partly overlaps model inputs) |
| Historic newspaper mine and placer mentions | AUC 0.54–0.57 | Weak |
| Placer claims within stream gravels | AUC 0.33–0.42 | Weak: placer model needs field data |
| Historic report facts, gold vs. barren | No separation yet | Inconclusive: barren locations too coarse |
| Reading historic report tables | 46 of 46 cells correct on a test page | Strong |
What we're doing about the weak spots
Satellites and lidar find the right kind of ground, but they can't tell rich gravel from poor. That takes field results. In the 2027 season we will pan and test top-ranked and low-ranked ground side by side, record every result including the empty ones, and publish the hit rate here.
Sources
US Geological Survey (MRDS, USMIN, geologic maps, NURE geochemistry, 3DEP lidar, Bulletins 911 and 955-D), NASA (EMIT, ASTER), ESA Copernicus (Sentinel-2), Bureau of Land Management (mining claims, surface management), MSHA, and the Library of Congress Chronicling America newspaper archive.