Lab Journal
The process,
in public.
Where we started, what we're changing along the way and why, and every experiment as it happens — including the results we'd rather not publish.
The map learned to talk back
A streaming field agent inside the app, callouts that finally look like ours instead of Apple's, and two findings from a v4 QA pass — both fixed the same night. The interesting bug was an inverted boolean we'd been misreading for a week.
Read entryWe tried to break our own app
Seven bugs, three of them functional, all of them ours. A headless simulator QA pass found them, fixed them, and left two sessions dead on the floor — both killed by the agent running the QA, not the app. Here's the log.
Read entryThe pivot — aiming at the niche that's real
UintaWatch is re-aiming at a validated, winnable niche with published success criteria: the detection gaps cameras and satellites provably can't cover.
Read entryWe audited our own words for AI slop
This month we pledged that every claim on this site would carry a source. Then we ran a slop detector over the whole repo to see if the writing itself held up. Here's what it found, and what we changed.
Read entryWhat comes next
The working roadmap — from reproducing published federal results on open data, to five deliberately different node configs, to first field numbers around partnered prescribed burns.
Read entryWe tried to disprove ourselves
We asked another AI to break the UintaWatch thesis, then verified every key claim ourselves. The sensor physics didn't survive. Here's what we found, sources included.
Read entryWhere we started
Two sprint weekends, one thesis — cover Utah's wildlands with community-built solar mesh sensors that detect fires before smoke is visible. Here's what we built and what we believed.
Read entryFollow the lab.
Entries are also available via RSS.