The solar inverter system meters everything crossing the property boundary, in fifteen-minute intervals, and uploads it to Enphase's cloud: how much the panels generated, how much the whole site consumed, and how much crossed the meter as import and export.

Those four are not independent — they must balance, and that gives us a free integrity check on every interval.

Measured on real data at an average of 0.013 Wh out. When this identity fails it points at the solar meter, not at ours, which is why it is reported separately on the Data quality page.

A Shelly Pro EM-50 sits on the granny flat's own circuit and is read every minute — far more often than the solar data arrives. Channel 0 is the whole flat. Channel 1 is the EV charger, which lives inside that circuit.

The two channels are never added together. The EV is already counted in the flat's total; summing them would bill that energy twice.

There is no meter on the house. Its consumption is what is left when the flat is taken off the whole property.

This is the single most important thing to understand about the system, because everything awkward downstream follows from it. Two independently calibrated instruments are being subtracted, so the house carries the error of both. On a quiet interval the result can even come out slightly negative, which is impossible — so it is clamped to zero and the interval is flagged.

That clamp is the Neg. house column on the Data quality page. It runs at roughly 3% of intervals and accounts for about a kilowatt-hour in total.

Each dwelling is entitled to a share of whatever the panels made in that quarter-hour. Whatever one cannot use — because it simply was not drawing that much — spills to the other rather than being treated as exported.

Then a cap, because the two meters disagree. Self-consumption is exactly what was generated and not exported, and the Envoy measures both of those itself — so it is the trustworthy number.

Without that cap, meter drift credited the dwellings with solar that had actually been exported — 1.5% of generation, which is money.

Whatever a dwelling used beyond its solar came off the grid. The flat's import comes from its own directly measured consumption. The house's is then whatever is left of the real meter reading.

Anchoring the house to the measured import rather than to its derived consumption is deliberate: it guarantees the two shares always add up to what the meter actually recorded, so the retailer's kilowatt-hours are never invented or lost.

A time-of-use plan charges differently through the day, so each quarter-hour is matched to the band in force at that moment — the plan that covered the date, then the window that covered the time and the day of the week.

This is why the Savings page prices interval by interval instead of using an average: a kilowatt-hour used during an expensive band is worth more than twice one used at midday, and averaging would credit the panels for the weather rather than for when they generated.

The retailer's bill is the authority. Each charge line is divided by each dwelling's share of the import measured during that band's intervals.

The flat's side is rounded and the house's is taken as the remainder, so the two always sum to the bill exactly, to the cent — nobody is ever short or over by a rounding error.

All figures in this system include GST, on both the bill and the plan, so the derived total matches what was actually paid.