Updates that can't brick a node
Two firmware slots. A new version only counts once it has connected and delivered a reading. If it can't, the bootloader goes back to the old one.
Technology
Six sensing points in and under the box, read by a node that sleeps almost all the time on a solar-charged supply.
From hive to screen
Holds 64 readings in memory that survives deep sleep
A low-power node wakes, measures, sends and goes back to sleep. No network? It keeps the reading and tries again next cycle.
Keeps every message in a local database
A gateway collects the whole apiary over Wi-Fi and forwards it over 4G. When the link drops, it stores, then replays in order.
Replays are harmless: duplicates are ignored
Each reading lands once, however many times it arrives. If a stretch of data is missing, the platform notices and asks the gateway for it.
On-board analysis
The node captures a short sample and computes its frequency bands on the chip. Only the bands travel, so a thin 4G link carries a colony's signature instead of an audio file. The same pipeline reads vibration from the accelerometer.
Frequency bands, configurable per node (illustration)
Two firmware slots. A new version only counts once it has connected and delivered a reading. If it can't, the bootloader goes back to the old one.
A staggered queue updates a few nodes at a time, so one uplink on one solar station is never asked to serve a whole apiary at once.
Below a safe voltage a node declines the download and asks to try again later. Power is the scarcest thing on the hill.
Each station reports its rails, so state of charge and autonomy are on the dashboard rather than being a surprise.
Each gateway has its own identity and may only publish under its own namespace. One apiary can't speak for another, enforced at the broker.
A failed sensor reports as missing, never as a zero that looks real. And gaps in the record are shown, then repaired where the gateway still has the data.
