The pivot — aiming at the niche that's real
We didn’t kill the project. We changed what it claims to be.
The reframe
The stress test’s strongest conclusion was that we had aimed at the hardest version of the problem — an autonomous statewide wildfire alarm network — while the evidence pointed at a different, real, and winnable niche. We’re re-aiming.
UintaWatch is an open, off-grid sensing lab for the detection gaps that cameras and satellites provably cannot cover: night smoldering, cloud-obscured events, low-intensity smoldering on contained or “extinguished” fires, and far-field smoke discrimination.
What published research says we can aim at
A January 2026 study in Fire (Natural Resources Canada, 9(4):141) field-tested continuous ground smoke sensors across 20 prescribed fires and wildfires — and the sensors were purchased by DHS S&T, the same program the IG audit condemned. The results give us external, published bars to clear instead of self-referential success:
| Published result | Our target on open hardware |
|---|---|
| 100% detection of fires >100 ha at median ~1.5 km | Match it with ~$50–100 open nodes and LoRa instead of internet backhaul |
| 30% detection of fires ≤1 ha | Improve via multi-node corroboration and wind-aware placement |
| Nighttime smoldering detected where satellites see nothing | Primary target — this is where we aim |
| Far-field regional smoke = major unsolved false-positive source | Solve it with open data and published algorithms |
| Closed vendor hardware, internet backhaul | Open hardware, off-grid — the unclaimed difference |
Sensors purchased by DHS S&T, deployed across 20 fires in Canada, 2023–2024 (Fire 9(4):141).
Clearing those bars is a positive, publishable result. If some turn out unreachable, that’s a published boundary — useful to everyone, still a win for the lab.
What changes in public
- Our website copy now labels hypotheses as hypotheses. The BME688 is one input under test, never “the fire detector.”
- We added an Open Questions page: what published research already shows, and what we’re measuring next.
- We’re starting from data, not hardware: the Canadian study’s full dataset is open (MIT license), so the lab’s first experiments can happen before a single node is built.
- Meshtastic stays as one tested transport — not a sacred architecture.
What doesn’t change
Open source, build-in-public, and the conviction that Utah’s detection gaps deserve better instrumentation. The next entry lays out the working roadmap.