The power nets out, the energy does not

Kapfenberg holds sixteen Shelly devices: eleven permanently on Wi-Fi, three on battery, two over Bluetooth. Three of them are meters, and only together do they produce a picture — one I assembled wrongly twice.

Two traps in merely finding them

Too tight a sweep reports no error. The first run waited two seconds and tried once. It found eight devices. There are eleven.

Among the three that stayed silent was, of all things, the house meter — that is, precisely the device this is about. With six seconds and two attempts all eleven turned up.

Even the thorough sweep does not find everything. Three devices run on battery and sleep between their reports. They answer no request and are fundamentally not findable by address.

An address sweep is never an inventory. It is a list of the devices that happened to be awake and answered fast enough.

The error I put my finger on

The energy dashboard needed grid import and grid export. The obvious move was to reach for the energy counters of the three-phase house meter.

That was wrong. Measured:

power      −34.9 W    = 45.7 − 80.6 + 0           netted
energy    845.6 kWh   = 821.3 + 7.4 + 16.9        sum per phase

Same manufacturer, same device, same family of units — and two different ways of calculating.

The power nets the phases against each other: what is exported on one phase is subtracted from another. The energy only adds up.

The grid operator’s meter nets. And that is exactly why the device’s counters are useless for a bill:

Use them and every exported kilowatt hour also gives you an import on another phase — counted twice, plausible both times.

The dashboard therefore holds purpose-built netting values, not the device’s raw counters.

The two generation figures — and why they do not contradict each other

Two devices report solar generation, and they report different things. For a long time that looked like a fault.

Measured simultaneously:

Battery, its own solar power	338 W
of which to the output	148 W
of which into charging	190 W
Sum	338 ✓
Meter behind the inverter	145 W

145 against 148 watts — about two percent conversion loss. The meter hangs behind the inverter, not on a strand of its own.

panels → battery → inverter → meter
         ^                     ^
         reports 338 W         measures 145 W

Over the counter readings the same relationship becomes even clearer:

983.88 kWh   total yield, on the DC side at the panels
850.18 kWh   measured behind the inverter
-----------
133.70 kWh   difference, about 13.6 %

That is not a measurement discrepancy. That is the efficiency. Storage losses plus conversion losses, summed over months.

And the wiring explains why the battery reports the higher number at all: four panels deliver over a thousand watts on good days. Whatever the battery does not take bypasses it to the inverter — and that caps at around 800 watts.

What ended up in the dashboard

Field	Source
Grid import	the netted value, not the raw counter
Grid export	likewise
Solar production	the meter behind the inverter
Battery	none

The last row is deliberate and was the actual insight. The energy dashboard models a battery on the AC side. This one sits on the DC side, before the inverter — its energy is already contained in the solar production.

More on that in Measure fast, set slow: enter both and you count the sun twice.

What I’d suggest

1. Configure your network sweep generously before you turn its result into an inventory. Two seconds of difference cost me the house meter here.
2. Keep battery devices on a separate list. They are not findable by address, and that is not a fault.
3. With three-phase meters, check separately how power and energy are calculated. It can be the same maker, the same device, and still different arithmetic.
4. Compare your counter against the grid operator’s before you put it into a cost calculation.
5. When two devices measure the same thing and report differently, look for the chain, not for the fault. Two points on one line may well show different numbers — the difference is the efficiency.

So what?

The remarkable thing about this topic is how plausible both errors looked.

Eight devices found does not look like eight out of eleven. It looks like eight devices. And a counter reading 13 percent away from another looks like a measurement error, not like the efficiency of an installation.

In both cases the number was right and the interpretation wrong. And in the second case it was not Claude who noticed but me — because I knew that my meter nets and the devices in the basement do not.