The other thing with the temperature measurements is they don't have to be accurate, as long as they are consistent, because it is the delta t we are after. The real temps might be 40 and 35, the measured ones 37 and 32, but in each case the delta t is the same, 5 degrees. Because I am using the same instrument to take the readings for each source (IR for the independent, whatever Midea use for the Midea data), the chances are each instrument is reasonably consistent, and so it is not unreasonable to assume it gets the delta t right, even if the absolute number is out.

Another thought: I measure the primary flow and return temps as the primary circuit enters and leaves the plate heat exchanger, mainly because it is the least awkward place to take the measurement. But strictly speaking, it they are not the LWT/RWT, which should be (and are, but the Midea sensors) measured at the heat pump. But which is more valid? It's a bit like HP at the flywheel vs HP at the road wheels on a car. Assuming losses in the flow and return pipe, the heat pump may put out X at the heat pump, but only some of X ends up delivered to the house.

I've just taken another set of temp recordings, this time capturing the IR images while noting the Midea values. The latter were LWT 38, RWT 35, and here are the IR images: 

The pipes here are copper, with black insulating tape applied (the warm bars you can see) to get the emissivity at a reasonable level, primary inlet to the plate heat exchanger (and outlet from the secondary circuit) in the left hand image, primary outlet (and secondary inlet) in the right hand image. In each image, the red cross hair is on the hottest point in the image (in both cases the primary pipework), and as you can see, the primary in was 43.2 degrees, the primary out 41.3 (this red cross hair is partially obscured by the readout panel).

This time the Midea delta t was 3 (with the integer reading caveat) , the IR delta t 1.9. They may or may not be correct, but they are at least credible. This time there was a difference between the Midea and IR data, with the former being (as may perhaps be expected, see above) having the higher delta t.

The eagle eyed will also spot something else: on the left hand image, the centre cross hair (white) is very close to the max cross hair (red), yet there is a degree centigrade difference in temperature. I can't account for this, unless, and I am just thinking aloud here, it is an angular effect caused by the curve of the pipework, some of the heat misses the camera. The two red dots you can see are both on the centre of the pipe, ie perpendicular to the tangent of the pipe at that point, the point at which it 'points' most directly towards the camera.    

By the way, this IR camera has already paid for itself. Not only is it very useful for heat assessments and loss detection, it is also useful among other things for monitoring things like engines and drive trains. It can probably even tell you when your bun in the oven is ready to come out, but there are probably simpler ways to do that.                

I wonder what ball-park figure one might apply to the accuracy tolerance of the data supplied by default measuring equipment rather than by instruments installed for the purpose

I don't think, for my system, it is too bad, because I have some basic independent monitoring tools, and they broadly agree with the Midea data collected over modbus. For example, I calculate energy out from:

(1) Midea reported flow rate, which I can verify with an analogue flow meter installed in the circuit

(2) LWT/RWT delta t, which I can verify using an IR thermometer/camera

(3) specific heat of the circulating fluid, which is fixed for a given fluid mix

The problem is verification for (1) and (2) is based on spot readings, whereas the modbus/Midea data is continuously recorded at minute intervals. Nonetheless, a short while ago the Midea controller said the flow rate was 1.23 m^3/h (20.5l/min), while the analogue meter showed 20l/min, a ~2.5% difference, but the analogue meter has a very crude scale, maybe it was actually showing 20.5l/min. A few moments later, the delta t on both the Midea controller and the IR measurements was zero (because the heat pump is cycling, and it was in an off phase). A few minutes later, with the on phase underway, the delta t was 4 on both. Since the Midea numbers are integers, they may be a bit out, but not hugely so, and I should add that absolute values for the temps are not the same, the Midea numbers are about 3 degrees below the IR measured values eg Midea number is 38, IR number is 41, but it is not the absolute values we need, rather the difference (delta) value, which is the same. These readings (and others done in the past) are enough to satisfy me that the Midea data based calculated energy out is tolerably close to the real value, say within 5%. At least, that is what I think is the case... 

I haven't read your full system details, but perhaps the heat delivery is restricted (by pipe gauges and emitter dimensions, etc).

That is certainly possible for the pipework, though the radiators are matched, plus a bit, to the 12kW heat loss, ie they are unlikely to be a bottleneck, unless I'm missing something eg delta t isn't what it should be. I also have a (dreaded by some) plate heat exchanger and around 10% glycol in the primary circuit. The primary pipework is 28mm plastic, which being plastic may be too narrow. I should also mention heat output is calculated from flow rate, LWT/RWT delta t and circulating fluid specific heat, all from the Midea wired controller, not from independent  monitoring kit.

I haven't read your full system details, but perhaps the heat delivery is restricted (by pipe gauges and emitter dimensions, etc).

That's pretty difficult to do without noticing!

Unless it happens either (a) when you are asleep or (b) when you are out! Also bear in mind we only have to account for very roughly a doubling of the heat delivered per 24 hours which could for example, be achieved by going from 24 hour running to two six hour periods, more than enough on time to warm the house when the inhabitants are up and present. Perhaps I also lowered the thermostat a bit, and that further lowered the heat delivered, or added some loft insulation. Lastly, the 50% reduction over the noughties is a very approximate estimate, as can be seen in the chart below. This shows heating oil purchased, not used (which may explain some of the early spikiness), by year, and it seems likely there was a trend downwards over time. Lastly, let us not forget Occam's razor!   

One takeaway from all this for me is the rather obvious conclusion - to come back to the original point - that estimating heat loss from heat delivered derived fossil fuel use doesn't work as an observational study, too many variables involved, the details of which will have been forgotten in the mists of time. It might work as a more carefully done experiment, with as many variables as possible controlled, and if not controlled, then adequately recorded. Even then, we still have the apples vs oranges thing, a fossil fuel system is likely to be on a timer, a heat pump is usually not.

Perhaps a heat pump running 24 hours a day can be used to determine heat loss if the house is kept in a steady state, but using a heat pump isn't much use in the preliminary stages, before the heat pump is installed! And even that may not be reliable. Recall this chart (based on measured data), which suggests my house needs around 8kW delivered (per hour, kWh, which is what the chart shows) at -2 degrees outside, compared to the around 12kW suggested by the heat loss calculations. Which is it, 8kW or 12kW; or is it yet another hidden number?   

And yet another unexplained thing: at around -2 outside, my house starts to cool a bit, and the room temp drops by a degree or two. Yet the 14kW (nominal) heat pump has plenty in reserve, even if we adopt the published Midea output at lower ambient temps of around 11kW. Why is it stuck at a maximum output of around 8kW average, 9kW peak, when on paper it can get up to around 11kW in these conditions?    

when you tuned your oil heating down to reduce your bills that the temperature dropped incredibly when not on

I think that is very likely, but I didn't notice either because it happened overnight, or during the day when I was out or active. The other thing is that during the early period of higher oil use (mid 1990s onwards), I just think we didn't worry so much about keeping a fossil fuel boiler running much of the time. In the mid 1990s I was paying just over 10p a litre, which works out at roughly 1p per kWh. Adjusting the 10p a litre for inflation at today's prices comes out at about 20p a litre. Yet over the last three years standard heating oil has always been over 50p a litre, and on a few occasions in 2022 it went over £1 a litre. ONS (Office for National Statistics) data only goes back to 2011, but I did find this chart for for Brent Crude that goes back further and our prices probably followed a similar pattern ie prices took off in the mid noughties (the 2000s 'energy crisis'), which is exactly when I started reducing consumption: