Calibration Mode: The First Ten

Every post so far has described a capability. This one describes a status.

Because the honest thing to say about KAIRair right now isn't "it's done." It's: there are ten stations on a bench, of four different types, and they're not going anywhere until they agree with each other about the air. That's the phase the whole project is in. Not launched. Not shipping. Calibrating.

I think the calibration phase is the most revealing thing to write about, precisely because it's the part most projects hide. So here's the state of the network, red cells and all.

---

what calibration mode actually is

Calibration is the boring-sounding work that everything else depends on: making sure the number a station reports is the number the air actually is.

It sounds trivial until you have more than one station. Put two stations side by side, in the same air, and they should read the same. If one says 21.4°C and the other says 20.1°C, at least one of them is lying, and you don't yet know which — or whether both are off in some direction. Multiply that across ten stations and four hardware types, each with its own sensors and its own quirks, and "do they agree" becomes a real question with real work behind it.

So the ten stations sit together. They run. They report. And their readings get checked against each other and against known conditions, so that the differences between them can be understood, and where necessary, corrected. The USB-powered baseline stations help here — always awake, never power-starved — as the steady ruler the battery and solar stations get measured against. When a battery station drifts, you want to know if it's the sensor or the power, and having a station you know is well-fed makes that answerable.

Calibration mode is that whole disposition: hold the fleet close, trust nothing yet, and earn each station's readings before letting them out.

---

why it's the gate, not a formality

Here's the thing I keep coming back to, because it's the entire reason to take this slowly.

A network's only real asset is trust. Not the map, not the API, not the pretty sky — those are all worthless if people don't believe the numbers. And trust is annihilatingly fragile: it takes one person catching one station confidently reporting nonsense to make them doubt all of them. A wrong reading isn't a small bug. It's an attack on the credibility of everything else the network says.

Which leads to the sentence I've repeated through this series on purpose: a network of confident, wrong readings is worse than no network at all. No network, people know they don't know. A wrong network, people think they know and they're mistaken — that's misinformation with a nice interface. The bar isn't "the station turns on and shows a number." The bar is "I would trust this number in your backyard as much as I trust it on my bench."

Ten stations is small enough to actually clear that bar by hand, one station at a time. That's not a limitation of the MVP. It's the point of doing an MVP — a scale where you can be genuinely sure, before you're at a scale where you can't.

---

the state of the network

Here's where things honestly stand. Green is solid and running. Yellow is real but rough or incomplete. Red is not done.

Piece	State	Honest note
The network & live map	🟢	Live at kairair.com; stations report, the map updates in real time
Source-of-truth definitions	🟢	Boards, sensors, kits → generated firmware, diagrams, BOM
Shared-HAL firmware	🟢	One firmware, config generated per station type
Signed OTA	🟢	Verified updates; key rotated and handled deliberately
Dynamic sensor capture	🟢	New measurements absorbed on first report
Research API + MCP	🟢	Keyed, versioned, and conversational
Weather experience	🟢	Live per-station scene; recently fixed for phones
The first ten stations	🟡	Built; in calibration, not yet released to hosts
Off-box backup	🟡	Plumbed and scheduled; not yet switched on — being finished
Automated test suite	🔴	Thin; correctness leans on validation + calibration, not a test harness
Scale beyond a handful	🔴	The architecture is built for it; the fleet isn't there yet

I'm leaving the red in on purpose. A build-in-public series that only shows green is marketing wearing a lab coat. The test suite is genuinely thin — a lot of the correctness story currently rests on the definition validator and on hands-on calibration rather than an automated harness, and that's a gap I'd name in a code review of my own work. The backup is scheduled but not yet live, which for a network whose signing key you can't afford to lose is exactly the kind of thing you finish before you scale, not after.

None of that is hidden. It's the actual frontier.

---

what "ready for hosts" will mean

So what's the finish line for this phase? When do the first ten leave the bench?

When I'd put one in your yard without flinching. Concretely: when the ten agree with each other and with reference conditions within a margin I can state out loud; when a battery station's lifespan is a known quantity and not a hope; when the backup is on so a lost key isn't a catastrophe; when the onboarding is smooth enough that a host with no idea what an ESP8266 is can go from box to live-on-the-map without me on the phone.

That's the gate. Ten calibrated, trustworthy stations, of four types, ready to be handed to people who will judge the entire network by the one sitting in their yard.

Ten sounds small. It's supposed to. Ten is how many you can be sure about — and being sure, at ten, is what earns the right to ever be at a hundred.

---

Part eleven of the KAIRair build series. Previous: Weather You Can Feel. Next, and last: more than weather — why this was never really about the weather, and what a network built to make air visible is actually for.