Our standards
Detection tools tend to fail quietly: they memorize their home terrain, over-alert, or stop working when the landscape changes. We don't publish our methods. We do publish the standards every result has to meet, and the results themselves.
Evaluation discipline
Nearby terrain looks alike, so testing a model next door to where it trained is quiet cheating. We hold back whole blocks of geography — whole counties, and in one case a whole state — and only report what was earned there.
For validation studies, we write down what will count as success before we run the test, and then we report what happened either way. The misses stay in the report next to the wins.
When a signal isn't there, we prove it isn't and move on. We walked away from an entire product idea — detecting orphaned oil wells from imagery — after our own controlled tests showed it couldn't beat chance. That discipline is also why you can trust the results we do stand behind.
Sample sizes, confidence ranges, where performance drops and why, and what a claim does not cover. All of it goes in the deliverable itself, not a footnote. Estimates are labeled as estimates.
Data provenance
Every input we use is free, public, and taxpayer-funded. That keeps our work reproducible and auditable — and it means no data vendor sits between you and your results.
| Source | Provider | What it gives us |
|---|---|---|
| 3DEP LiDAR (~1 m) | US Geological Survey | Bare-earth terrain — the shape of the ground with the forest digitally removed |
| State AML inventories | PA DEP, Ohio DNR, others | The official hazard records we grade ourselves against |
| Hydrographic archives | NOAA NCEI | Side-scan sonar and bathymetry for the maritime work |
Public materials show aggregate statistics and illustrative figures only. Precise coordinates of candidate hazards — and of culturally or legally protected sites — go to the responsible agency, not onto the internet.
The fastest way to evaluate any of this is to point us at ground where you already know the answers, and see what comes back.
Start a conversation