Three months of receipts — what solar-aware EV charging actually returned
In August we published how we taught a cheap EV charger to follow the sun. This is what it returned: 6,792 km, every kilowatt-hour metered at the wall, 4.5c per kilometre against 26c for the diesel it replaced — and the four silent faults we found along the way.
In August we published how we took a cheap Tuya EV charger off the manufacturer’s cloud and taught it to follow live solar surplus, amp by amp. The reasonable question from anyone weighing up whether to do the same is the one that write-up could not answer: what does it actually return?
Three months later there is a metered answer. From 1 August to 7 November the vehicle covered 6,792 km, every kilowatt-hour of home charging passed through a dedicated meter, and every public fast charge was reconciled against the receipts. This is what the numbers say.
This is a follow-up to Charging the car on sunshine, which covers how the control system was built. There is a longer and considerably blunter version of the results — the arguments at the barbecue, the mistakes and the full configuration — on Angry Dad Automates, the community side of what we do.
How it was measured
The credibility of a result like this rests entirely on the instrumentation, so it is worth being specific. Four independent sources:
- A metered DIN breaker on the charger circuit. The vehicle charges from a 15 A portable EVSE on its own circuit. Every watt delivered from the house passes through that meter — not an estimate, not a derived figure.
- A clamp meter on the main board. A Shelly EM3 on the grid feed reports import and export minute by minute, which is what allows each kilowatt-hour into the car to be attributed to the roof or to the grid.
- The tariff clock. Every minute of charging is tagged with the rate in force at the time, so cost is calculated rather than apportioned.
- The vehicle odometer, for distance.
Why the vehicle’s own consumption figure was not used
Over the same ninety days the manufacturer’s app reported total energy use of 8.7 kWh — against nearly seven thousand kilometres of driving. Vehicle telemetry of this kind is a dashboard feature, not a billing record, and it is not reconciled against anything.
If a number is going to inform a purchasing decision, meter it at the wall. This is the single most common reason we are asked to re-do someone’s energy reporting.
The result
Energy delivered and what it cost, across the full period:
- Home charging, metered: 1,439 kWh for $170.74 — 21% of it from the roof
- Home charging in the first week, before the meter was fitted (estimated from state of charge): about 57 kWh, about $17
- DC fast charging — five sessions at roughly 70c/kWh: about 166 kWh, about $116
- Total: about 1,662 kWh for about $303
Against a diesel 4WD covering the same 6,792 km, assuming 10 L/100 km — a fair real-world figure for a 2.8 litre ute that works for a living — and $2.60/L, a mid-range Brisbane price across the period:
- Electric: $303, or 4.5c per kilometre
- Diesel: 679 litres, $1,766, or 26c per kilometre
A difference of $1,463 over three months, or about 83% cheaper per kilometre, at around 22 kWh/100 km for a three-row vehicle.
Where the saving actually comes from
This is the part that gets lost in the general argument about electric vehicles, and it is the part that is engineering rather than purchasing. Charging is not one price. It is whatever the tariff clock says at the moment the current flows. The same 1,439 kWh of home charging split like this:
- Super off-peak (midnight–6am), 8c: 779 kWh
- Solar surplus, which would otherwise have exported at 2c: 308 kWh
- Shoulder, 30c: 202 kWh
- Off-peak, 25.6c: 83 kWh
- Peak (4–8pm weekdays), 39.9c: 66 kWh
Roughly three-quarters of the energy came from either the cheapest window or directly off the roof, which is why home charging averaged about 12.5c/kWh all-in. The same energy drawn on a flat 30c tariff would have cost around $430 instead of $170.
The public fast charging is the mirror image: 10% of the energy and 38% of the cost. At 70c/kWh it is still around a third cheaper per kilometre than diesel, so it is not a problem — but it is a travel cost, not an operating model, and any projection that assumes regular fast charging will look very different.


The vehicle is not where the saving is. The saving is in what time of day the current flows, and that is a control problem.
Commissioning versus steady state
The first two months included delivery, longer trips and a control system still being tuned, so the last 30 days are a better guide to what the arrangement settles at. Comparing 1 August–8 October against the most recent 30 days:
- Home charging cost: 13.2c/kWh → 11.0c/kWh
- Share supplied by solar: 19% → 24%
- Share bought at shoulder or peak: 19% → 15%
- Consumption: 22.2 → 21.2 kWh/100 km
- All-in cost: 5.1c → 2.7c per kilometre
In fairness to the sceptics, most of the headline improvement is simply that there was no fast charging in the recent period. But the home charging genuinely improved on its own terms as well — 17% cheaper per kilowatt-hour, with more sun and less peak.
Month by month, home charging only
- August (from the 7th): 554 kWh, $88.54, 16.0c/kWh, 20% solar, 46 kWh at peak
- September: 345 kWh, $25.23, 7.3c/kWh, 20% solar, zero peak
- October: 475 kWh, $54.29, 11.4c/kWh, 18% solar, 20 kWh at peak
- November (first week): 65 kWh, $2.68, 4.1c/kWh, 62% solar, zero peak
August is the control group, and it is worth dwelling on, because it is exactly what an electric vehicle does out of the box on a time-of-use tariff: it is plugged in on arrival home and charges straight through the most expensive four hours of the day. September is the same vehicle, the same tariff and the same roof with the control system left alone. The difference is 16.0c against 7.3c per kilowatt-hour, and forty-six kilowatt-hours of peak against none.
What went wrong, and what that says about commissioning
We publish the faults because they are the part of this work that is genuinely hard, and because every one of them failed silently. None produced an error anybody saw.
A timezone fault that charged straight through peak
The stop logic needed elapsed time since the charger last changed state, to stop it oscillating. The calculation subtracted a timezone-naive timestamp from a timezone-aware one, which the template engine refuses to evaluate. The error was swallowed, the stop branch never matched, and the charger ran into the evening peak on grid power.
The arithmetic fix is one line. The design fix matters more: there is now an unconditional override — after dark, or during peak, stop immediately, with no debounce timer in the path. Any controller whose failure mode is “keep running” needs a stop condition that does not depend on the logic that failed.
A renamed sensor, and a load that quietly took the wrong turn
An update renamed one of the vehicle’s sensors. The other solar-consuming loads were still interrogating the old entity to find out whether the car was plugged in, and an entity that no longer exists never reports that it is. They concluded the car was never charging and helped themselves to surplus that should have gone to it. No error, no alert — just a worse outcome for a couple of days until the data was reviewed.
Every entity an automation depends on needs availability monitoring, not just the ones in the headline. A dependency that silently reads as “no” is more dangerous than one that fails loudly.
Equipment that is online but not complying
Two separate instances. The charger dropped off the network after being re-paired in its own app. And for a stretch the vehicle declined to AC-charge at all, because a setting in the manufacturer’s app had been switched off — so the control system asked, the car refused, and the pair of them did this indefinitely without complaint.
Monitoring for “offline” does not catch this. The system now alerts on the charger dropping off the network, and the broader lesson is to alert on commanded-but-not-complying: energy expected and not delivered is the condition worth watching.
The most expensive fault was a person
Every peak-rate kilowatt-hour in October traced back to someone pressing the manual Charge Now override on a weekday evening. Three times. The same person who built the system.
That is not a defect, and the answer is not to remove the button — sometimes the car genuinely needs charging at 5pm and that is a legitimate call. The answer is to make the expensive choice visible and deliberate at the moment it is made, and to log it so it shows up in the review rather than as an unexplained bump in the bill.
Orchestration: what happens when the car is not there
A vehicle with a working day is frequently not home while the sun is up, and when it is home it is sometimes already full. Surplus with nowhere to go exports at 2c.
So the EV is not the only load following the sun. A small chiller and a pool filtration pump both have work that must happen anyway, and both are better done on surplus than bought at night — the chiller case is covered in the forecast-driven chiller write-up.
They run from an explicit priority order, exposed as a single selection on the dashboard. The default is vehicle, then chiller, then pool. Each load may only start on solar when nothing above it still wants the energy, and the allocation layers: once the vehicle is pinned at its 15 A ceiling and surplus remains, the chiller may take the remainder, and then the pool.
The ordering is not arbitrary. The modulating load belongs at the top. The vehicle is the only one of the three that can adjust its draw an amp at a time, so it sits on top and absorbs the noisy, cloud-dependent part of the surplus; the fixed on/off loads sit underneath and take the steady block beneath it. Put a fixed load on top and it spends the day tripping on and off behind passing cloud.
Changing the order is a dropdown, not an edit to the configuration. On a handover that distinction is the whole point — a system the owner cannot safely re-prioritise is a system that will be worked around.
What it projects to
With 235 days to the end of the financial year, at the October home-charging rate (which ignores the summer sun, so it is deliberately pessimistic) and allowing one fast charge a month:
- At the average pace so far, 70 km/day — about 23,200 km: roughly $980 electric against $6,040 diesel. About $5,060.
- At the recent pace, 58 km/day — about 20,400 km: roughly $900 against $5,310. About $4,410.
For sensitivity: every 40c/L movement in the diesel price shifts the comparison column by around $900 over the year, and any more fast charging than one session a month lifts the electric column. The gap is measured in thousands either way.
A necessary note on scope and safety. These are energy costs only — no registration, insurance, servicing, tyres or capital. That is a different and much longer analysis. The diesel comparison is a modelled estimate, not a metered one. Installation of EVSE, dedicated circuits and fixed wiring is work for a licensed electrician and must comply with the relevant standards; changing charge current in software does not alter what a circuit is rated to carry, and the automation must never be the only thing preventing an overload. The charger here is driven over a local protocol through a community-maintained integration, and the vehicle through an unofficial one that uses the manufacturer’s cloud — appropriate for a personal installation, and something we would scope differently, with supported equipment, for a commercial deployment.
What this means if you are considering it
- Most of the saving is scheduling, not the vehicle. The same car on a flat tariff, plugged in at 5pm, captures a fraction of this. The difference between August and September here is the whole argument.
- Three things are required: a tariff with a spread worth chasing, a charger whose current can be controlled, and a true import/export measurement at the switchboard. Beyond the charger itself, the hardware is a clamp meter.
- The feed-in tariff is doing quiet work. At 2c export against 30c retail, self-consumed solar is worth roughly fifteen times what the retailer pays for it. That ratio is what makes diverting surplus worth engineering at all.
- There is no battery in this installation — a 6 kW array and nothing else, so the roof can only help while the car happens to be parked at home in daylight. The 8c overnight window does the heavy lifting. Adding storage changes the shape of the problem substantially, and is a separate piece of analysis.
The full configuration
The complete Home Assistant packages — the two-minute decision loop, the fifteen-second current modulation, the priority ordering and the fault fixes described above — are published in full in the Angry Dad Automates write-up linked at the top, and in the original build notes. They are there to be copied.
If you would rather have it designed, installed and monitored properly — including the alerting that catches the silent failures above — that is what we are for.