The gap that resets your devices is the transfer window inside the hybrid inverter — the interval between the grid collapsing and the backup circuits being carried again by the battery, the panels, or both at once. It is measured in milliseconds, sometimes stretching to a small fraction of a second, and whether a particular device survives it comes down to a race between two numbers that almost never appear on the same page: how long the inverter takes to change source, and how long the device's own power supply can keep running with no input at all. Lamps, kettles and heaters lose that race and don't care. Routers, decoders, fibre ONTs and computers lose it and start over from nothing.
Come and stand in my study with me at twenty to eight on a wet Tuesday evening. Rain on the window, kettle just switched off, desk lamp on, the fibre ONT blinking quietly on the shelf, the router above it, the decoder under the telly in the next room. Everything on this side of the house sits on the backup circuit, so everything here is being fed through the inverter whether the grid is up or not.
Then the grid goes.
What you notice first is that you notice almost nothing. There's a small click from the inverter — the relay — quieter than its fan. The lamp dips, a brown-out so brief you'd swear it didn't move, and comes straight back. Ten metres away, though, the decoder's front light has gone red and the TV is showing a logo. The router's LEDs are running their start-up pattern. The internet is gone for the two or three minutes it takes that router to train up again, and every device in the house that was waiting on it has fallen over too.
What the box is doing while you're watching the lamp
The inverter is not waiting for a signal. It is watching the grid at its AC input, sampling voltage and frequency continuously, and it has a threshold below which it decides the grid is no longer there. Detection is not instant — it needs a cycle or two to be sure it isn't looking at a dip that will recover, which at 50 Hz is already tens of milliseconds gone. Then it opens the input relay so it isn't pushing power into a dead network. That anti-islanding behaviour is the same requirement that sits behind NRS 097-2-1 on the grid-tie side, and it applies whether you export or not.
Only after that does the inverter start or continue its own AC output onto the backup circuits, fed from the battery, from the panels, or from both. Detection plus decision plus transfer. The whole thing is over before your kettle has stopped boiling, and it is still long enough to matter enormously to some hardware.
The race between two numbers
Here is the thing that generic explanations of backup power skip. There are two clocks running during that window, and the outcome depends on which one is longer.
On one side, the inverter's transfer time. On the other, the hold-up time of whatever is plugged into the wall — how long that device's own power supply keeps its output alive on stored energy once its input disappears. A desktop computer's power supply holds up for something in the region of a couple of dozen milliseconds at full load; that figure is written into the ATX supply specification, which is why desktop PCs are more stubborn than people expect. A router's little wall adapter has almost nothing inside it — a small capacitor and not much else — so it holds up for a fraction of that. A decoder sits somewhere in between. A fibre ONT is more fragile still, because even when it has power it can need to renegotiate with the ISP after a break, which is a minutes-long problem rather than a millisecond-long one.
So the same thirty-millisecond gap that a desktop sails through can kill a router dead. The inverter isn't being unfair. The router just ran out of stored energy first.
Resistive loads have nothing to lose. A filament lamp, a kettle element, a heater bar, a geyser element — they don't hold state, don't run firmware, don't care that the voltage went missing for the length of a blink. They dim and they come back. That's why the lamp is lying to you about how bad the event was.
And note something about the devices that do survive: laptops, phones and tablets on their own batteries sail through the grid drop regardless of how fast the inverter is, because the inverter is not actually carrying them at that moment. Their internal battery is. That's not the inverter doing well. That's the inverter being irrelevant for a few seconds.
When it isn't a gap at all
Sometimes the backup circuit doesn't blink. It goes dark and stays dark for minutes, and the user blames the transfer time when the real cause is completely different.
If the instantaneous demand on the backup circuit is more than the inverter can deliver in that instant, it has two ways to fail and neither of them is a neat millisecond handover. It can shut down on overload and go quiet, or it can refuse the load and leave the circuit unpowered until it recovers. Either way the experience at the plug is identical — everything off — but the cause is a surge, not a transfer window. A fridge compressor kicking in at the same moment as a laser printer warming up and an aircon already running is the classic shape of this. If your household story is "the inverter tripped" rather than "the lights flickered", you were never in a transfer race at all.
Sun, cloud, and state of charge
A hybrid system does not have one personality. It has several, and they change with the weather and the time of day.
At ten in the morning on a clear day, the panels are likely carrying most of the backup load anyway. The grid drop is often a non-event in that state: the battery has headroom, the solar input is steady, and the inverter has plenty of room to absorb the transition. A cloud rolls over and production drops sharply, but the inverter simply takes the difference from the battery, so still nothing at the plug.
At ten at night there is no solar input at all. Everything on the backup circuit is coming out of the battery, and the question becomes whether that battery has both the energy and the discharge headroom for the load sitting on it. Near the bottom of its usable range, with a heavy backup load running, the inverter may simply not be able to hold the circuit through the transition — not because it's slow, but because there's nothing behind it to pull from. Same house, same inverter, different outcome depending on the hour.
Coming back is a second event
The grid returning is not the end of the story, and this is the part people find genuinely surprising.
The inverter does not trust the grid the instant it appears. It watches voltage and frequency and waits before tying back to it — NRS 097-2-1 requires a reconnection delay for grid-tied inverters, typically measured around the minute mark, precisely so a flickering network isn't hammered by every inverter on the street reconnecting into a fault. During that waiting period your backup circuits are still on battery, so you notice nothing at all.
When the check passes, the inverter resynchronises and hands the load back. Whether that handover is invisible or produces a second small blink depends on whether the inverter synchronises before transferring or simply switches. Some do it cleanly. Some don't. That's the second transition, and it's the one most people have never heard of until they watch their decoder reboot twice in ninety seconds.
The nastier version of this is a grid that comes back, holds for a few seconds, and goes again — auto-reclose on a municipal feeder does exactly that when a branch is lying across the line. If the second drop catches devices that were still booting from the first one, they can end up stuck mid-startup rather than simply restarting. A device loading firmware is often less tolerant of a second cut than it was of the first.
What actually decides it
The single thing that determines whether your evening goes smoothly is not the inverter's milliseconds. It's which loads were placed on the backup circuit when the system was designed, and how much headroom the battery has at the moment the grid fails. Those are decisions made months before the outage, by whoever drew up the essential load list. A circuit populated with lamps and a fridge will behave completely differently from the same inverter feeding a router, an ONT, a decoder, a NAS and a desktop.
There is also a language problem worth knowing about. Marketing copy saying a system provides uninterrupted supply does not always mean the output never drops — some inverters run the load from their own stage continuously so the grid is never the direct source, and nothing blinks on those. Many hybrids are simply fast rather than seamless. Fast enough for a lamp is not fast enough for a router, and no datasheet line tells you which one you bought.
I've stopped treating a reboot as a fault. If the lamp dimmed and the router came back by itself, that was a clean transfer and the hardware did its job. If the decoder is sitting on a logo screen and the fridge is humming louder than usual, something else happened — a surge, a shutdown, or a double dip — and the gap everyone talks about was never really the problem.
