⚪⚫ They Turned On a Sensor Network at Lake Tahoe. Nobody Wrote the Docs.
Two days after UNR switched on the Tahoe Environmental Observatory Network, I started reading it. No API documentation, then or now. Ten endpoint slugs guessed from a truncation pattern, forty-three sensors that turned out to be thirteen devices, a soil pH of 1.8, and a five-digit parameter code that turned out to be the pollutant the lake’s TMDL actually regulates. Plus nine things I got wrong — including one that silently returned zero rows from every gauge for hours while reporting success.
Read post →🎲 I Wanted to Play Star Frontiers Again. I Accidentally Built a Game Engine.
It started with a photo of a 1983 Referee’s Screen on Instagram. Four games and about two thousand lines of JavaScript later: a shared engine, dice with real silhouettes and synthesized audio, and a survival game built on genuine USGS topo tiles of Railroad Valley, Nye County, where every distance and walking time is computed rather than invented. Plus three things I got wrong in public — impossible dice checks, atmosphere where information was needed, and death clocks nobody could see.
Read post →🔥 Watching a Mountain Burn
The Hawk Fire burned across Peavine Mountain north of Reno on 22 August. Ninety-two minutes after ignition, a thermal satellite saw it 315 metres from an aspen grove where Basque sheepherders carved their names into the trunks a century ago — a place I hid a geocache in 2008 and used to ride to in the 1990s. Sentinel-2, VIIRS and a watershed analysis, all free and public, from eight hundred miles away. Includes the terrain conclusion I got wrong first, and why a healthy recovery curve would be bad news for the carvings.
Read post →🌲 Finding the Trees
A stretch project into airborne LiDAR and point clouds — PDAL, canopy height models, individual tree detection — taken from a synthetic Finnish forest through to real national data. The detector scored 78.4% recall against 450 trees I placed myself. Then recall went up as the data got worse, perfect ground calibration hid a canopy bias, and a 100% precision turned out to be structurally guaranteed. Validated against 1,295 stands from the Finnish Forest Centre inventory.
Read post →🏞 The Creek That Isn't There
Big Creek runs dry more than a third of the time, then floods 200 times over. Asking Snowflake's warehouse-native AI agent about the creek behind my hometown, and why it only worked because a tested dbt model was underneath it — plus 170 years of Groveland history explaining why the creek was never the town's water supply.
Read post →🌲 Half a Million Trees
Learning Snowflake and dbt by pointing them at the federal Forest Inventory and Analysis dataset for Washington — 531,490 measured trees, where the wood is, where it dies, what fire does to both, and five confident predictions that turned out wrong, including two the data caught quietly lying.
Read post →π Bob's Gang Wanted a Trade Journal. GitHub Pages Wanted Me Dead.
Building an AI-assisted trade journal for a Rotary investment club: a domain that jumped repos and briefly deleted my own website, a cache that wouldn't let go, and the no-cors redirect bug that ate every form submission with a cheerful success message.
Read post βπ Boo Before You're Ready
A twelve-foot animatronic skeleton in the garden-center aisle in July isn't really about Halloween. It's the newsvendor problem, a tariff that changed its legal basis twice this year, a real-estate arbitrage disguised as a costume shop, and a seasonal-adjustment model quietly getting fooled.
Read post β