A backup unit that is never inspected is an ornament with green lights.

The first is the obvious one: when the power goes, the batteries hold the system up. The second is less obvious and happens every day: cleaning the power. What arrives at an industrial estate is not a perfect waveform —there are dips when a machine starts, spikes, small cuts— and that ages the power supplies in the equipment. A decent backup unit isolates them from that, and there is a difference between ranges here: some only step in when the power goes out of range and some rebuild the waveform all the time, which is what you want where the mains is bad.

The third is distributing and watching. This house's cabinet power strips can measure the consumption of each line and switch a socket off or on remotely —which means being able to restart a hung piece of equipment without going out there— and there are probes that warn of temperature and humidity inside the cabinet. That watching is what turns a future problem into an alert.

Runtime depends on three things: the capacity of the batteries, how many watts are being drawn and the age of those batteries. The trap is that the relationship is not proportional: twice the load does not mean half the time, it means considerably less. That is why a unit that at thirty per cent load would give plenty of time gives surprisingly little at eighty.

Then there is what has actually been plugged in, which is never what was there on the day of the calculation. And then age: six-year-old batteries give roughly half what they used to give, and they carry on showing green lights until the day of the cut. That is the classic fault: the system that lasted twenty minutes lasts four, and nobody knew.

So the only runtime that means anything is the measured one: everything switched on, at the busiest hour, pulling the incoming power and timing it. It is done at handover and repeated once a year. It takes half an hour and it is the difference between a number off the brochure and a fact.

If the recorder is backed up and the switch that powers the cameras is not, the recorder stays on recording nothing. It sounds absurd and we have seen it many times, because the recorder is what looks important. The rule is that the whole chain has to survive: cameras, the switch that powers them, recorder, storage if it is separate, and the workstation somebody is going to watch from if anybody is watching.

And there are things that are NOT connected. Large monitors that are only any use if somebody is in front of them, printers, a heater, the computer in the office next door. Every watt plugged in comes off the minutes of what matters, and a free socket on a backup unit is a permanent temptation: two years later there is something plugged in that nobody decided on.

The doors deserve a mention of their own, because it is a safety decision and not a technical one: what the access points do during a long cut, whether they stay locked or released, and what the building's evacuation regulations say about it. That is decided with the client and written down.

When the power disappears in the middle of a write, what can be damaged is not the video of that instant: it is the database that says what video there is and where it is. A system can end up with all its drives full of recording and unable to find it, and rebuilding that takes hours or days.

That is why the backup unit has to talk to the server. A program is installed that sees how much battery is left and, when it drops below a threshold, shuts the system down in an orderly way: it closes the video software, flushes what was in memory and switches the machine off. And afterwards, when the power comes back, everything has to start up on its own and in the right order, because nobody is going to go and switch a server on at three in the morning.

The sequence is decided too: if the cut is a long one, what gets switched off first —what is not essential— and what gets switched off last. And the whole thing is properly tested, by pulling the power, on the day of handover. It is the test most installations fail.

A backup unit is not a generator. It is there so a flicker goes unnoticed and so you can shut down in order; if the installation has to keep running through a cut lasting hours, the answer is a generator set with the battery backup covering the minute it takes to start. Selling two hours of batteries to solve a four-hour cut is doing a sum the client will discover on the bad day.

Nor would we fit it inside a closed cabinet without thinking about the heat: batteries hate heat, and a big unit inside a cabinet with no ventilation ages its own batteries very fast.

And we would not fit it without network monitoring in a place where there is nobody. A backup unit that does not report the state of its batteries is exactly as useful as not having one, because the day it fails is the day it was needed.

The IT part is comfortable: these units are watched over the network, they send alerts by email and their status can be fed into the alerting system the client already has. Switching a socket off and on remotely to restart a hung piece of equipment is also within reach of any IT team, and it saves journeys.

What cannot be left to be forgotten is the physical maintenance, because it is the only thing that decides whether the backup exists: batteries run down, and a backup unit that is never inspected is an ornament with green lights. That means a runtime test once a year with everything on, and a battery replacement by calendar, not by fault.

We can do that inspection, or the client's own maintenance team can do it if it is written down how and when. What does not work is nobody having it on their list, which is what happens nearly always: it is the piece of equipment nobody remembers until the day the power goes.