The dock said “Normal” for twelve hours. Its own data said otherwise.

Just before midnight on a warm July evening, a UK-operated DJI dock quietly stopped cooling itself. Its climate system's power draw fell to less than half its normal idle level and stayed there. The camera stopped producing video at the same moment. Telemetry kept flowing, and the manufacturer's console showed what it would show all night: Idle. Normal status.

By morning the cabinet was five degrees above ambient and climbing — with a drone and its battery inside. By late morning it had passed 35°C, and the battery was tracking it upward, on a day forecast to get warmer.

At 08:30, Dock Pulse opened a thermal-runaway alert: internal temperature more than five degrees above ambient, sustained. At 09:55 it opened a second: climate-system draw at 44% of this dock's learned idle median, with equipment inside. The operator was notified before 10:30.

What followed took under two hours, all of it remote. The dock was restarted — video returned; cooling didn't. A manual cooling cycle was commanded — the compressor engaged instantly, pulling the cabinet from 35°C to 27°C at half a degree per minute, the battery following it down into safe storage temperature. The conclusion wrote itself: the hardware was flawless. The dock had simply never issued the command. The automatic climate loop had hung — and when it resumed, both alerts closed themselves, timestamped, no human tidying required.

The part that sells the lesson

Sixteen days earlier, the same dock did the same thing — same late-evening onset, same dual failure, cabinet to 36°C, a full day of failed camera captures. Nobody knew, because there was nothing watching: this was before Smart Alerts existed. That incident only came to light later, in historical telemetry — during the development of the very feature that would catch its sequel. The first failure became the training data. The second met the finished system, and was diagnosed to component level before lunch.

A dock cannot be trusted to report a failure of the system that does the reporting. Condition monitoring has to live outside the machine — watching derived signals like power-draw baselines and thermal deltas, not the machine's own opinion of itself. That's what Smart Alerts does, on every dock, every ten minutes.

The battery inside spent a morning warmer than anyone would choose. It spent the afternoon at 21°C. That difference is the product.

Postscript: it happened again — and that is rather the point

Sixteen hours after recovery, the same dock re-entered the same failure state — late evening again. This time nothing was “discovered”: the alert opened within 23 minutes of the signature forming, and the operator was notified minutes after that, before a single degree had moved. The documented remote procedure recovered the dock the next morning, and a reproducible reproduction recipe — built from two mornings of timestamped intervention boundaries — is now with the manufacturer as a formal defect report. Until the root cause is fixed, the monitoring stands guard.

That is what resolution looks like in real operations: not a fault that never returns, but a fault that can never again go unseen. The first incident cost twelve silent hours. Its latest recurrence cost twenty-three minutes of detection time. That trajectory is the product too.