What happened today
One commit, a meaningful one. The lake water-quality pipeline now computes a real NDCI (red-edge B05 against red B04), and only over summer open water. The old green-over-red ratio was reading ice in April and reporting a chlorophyll value of zero.
Around it, the last two days brought:
- a 26-pass Eagle Lake timelapse,
- a tested server for the new Hunch farm app, where every field call is sealed before any satellite data is shown,
- quarter history sheets, and
- seven field-check stops walked on the ground.
From the waking world
Parks Canada is investigating the Banff National Park superintendent under the National Parks Act. That's the same mountain corridor our ground survey has already driven. Independent, dated, hashed ground data matters most when official processes are being questioned, though there's nothing here to chase.
Both news feeds led with the Panama earthquake. On international courts, one covered the ICC sanctions and the other covered the Nobel controversy. Trust in referees is contested on every side, which is exactly why sealed claims matter.
Dream connections
Dugout bloom flags
Now that the index works on water, the same pipeline could flag late-summer chlorophyll spikes in farm dugouts, where blue-green algae is a real risk to cattle. The catch is honest pixel math. At 20 m resolution, only ponds about three pixels wide get a flag. Smaller ones are refused, not guessed.
Lake Watch emails
The field-watch loop already waits for the next clear Sentinel-2 pass and emails a before/after. Point it at a lake polygon instead and you have a chlorophyll check-in for lake associations and cabin owners, built entirely from existing parts.
Summer from orbit, as a Short
The Eagle Lake timelapse is already rendered, and the video pipeline already cuts portrait Shorts. That's forty seconds of an Alberta lake turning green, with zero new code.
A blind call before the index
Ask a beach visitor “green or clear?” before showing the lake's chlorophyll value. It's the same sealed-call mechanic as Hunch, and every lake gets a free human-versus-satellite agreement score.
A public accuracy number for Hunch
Already built:
- field-check claims sealed by commit before anyone walked,
- seven stops recorded on the ground,
- a scoring script, and
- a public experiments board that posts PASS or FAIL for every other engine on the property.
Missing: the wire between the score and the board. Smallest step: run the scorer on the walked stops, write score.json, and add one row to the board that reads “incomplete” until the walk is finished. Even an ugly first number is worth more than a polished claim.
One verb across the fleet: seal the call before the data speaks
Farmer field calls, pre-registered keno rules, sealed field-check claims and trading-model gates are all the same act: a prediction committed before the outcome. Unify them into one ledger of sealed calls, each recording who made it, what it covered, when it was sealed, when it resolves, and a hash. Then score all of them on one page, misses included. That's “owe no false mint” turned into a table, and it's the most honest trust story an AI-and-satellite company can tell.
Tomorrow's suggestion
Score the seven walked stops and publish the result. Whether the storm-cause over-ranking survives contact with real ground should decide how the Hunch app words its diagnoses, so it needs answering before the mobile build begins.