https://www.youtube.com/watch?v=PIsp1ckK14A

I hope these experiences are of some help.

They are, and thank you for the detail of what you do. One of the many useful functions of this forum is that it provides a repository of different ways to tackling various setups.

I doubt we will ever know exactly how AAA or Homely for example work, that would take complex reverse engineering way beyond what I am capable of let alone willing to do given these are commercial products, but I do think it is very valid to consider the general principles eg what are the factors that determine the actual room temperature, and then see what we can do to reduce costs while maintaining comfort, which at the end of the day is what most of us want.

On the question of recovery after a power cut (which we have a fair number of, rural location with overhead lines) - I have deliberately made my setup as resilient as possible. The mini PC that runs the show is set to reboot after a power cut, and then various scripts including dare I say it a residual Home Assistant installation auto-restart. The main failures I get at the moment are HA logging conking out with "Unknown error OverflowError('int too big to convert')" the cause of which I have not been able to find (I can track down the code that logs the error, but not what triggers the error), which I fix by having a HA automation detect the error, and when it occurs, it reboots HA.

The reason I still have HA (for those who know my general views on HA) is purely historical. It's been on my system from day one, installed as part of the initial installation, and although I have increasing moved away from it, it does still have some useful functions, and rather than re-invent the wheel, I have let HA carry on doing what it does perfectly well. 

@transparent I'm sorry, I don't know. Others have said this on this forum in the past. I haven't noticed anything after a power cut, but I haven't checked.

Also AAA does not follow a preset curve like WC. Presumably the "learning" is something to do with the observed rate of change of Room Temperature, but that's just guesswork on my part. Others have said it takes days to settle down after a restart to the system.

Please clarify.

Are you suggesting that a power-cut would lose the data which the AAA control system has 'learned'?

If so, then DESNZ need notifying.
They're proposing a Demand Side Response mechanism, which compensates householders who allow their heat-pumps to be turned off by a DSR-Agent using remote control.

I do already have a basic auto-adaption setup, a python script that checks the difference, if any, between the actual room temp and desired room temp, and then, if there is a difference, it adjusts the flow temp by moving the ends of the weather compensation curve accordingly. Generally it does work eg if I run with an overnight setback, and the actual room temp drops by a few degrees, then the auto-adaption script adapts the weather compensation curve to achieve a post setback recovery boost, and removes it when the room temp is close to where it should be. The code isn't complex, it is a series of 'if this then do that' statements. I am now turning my thoughts to what other auto-adaptions I can include, hence my interest in what other auto-adaption systems do. The two obvious candidates are firstly the weather and secondly the buildings thermal characteristics.

The weather of course is unpredictable in the detail, and I know from other work that I have done that converting weather data (actual or forecast) into something a computer can use is far from easy. Take solar gain for example. It would need something like the predicted (or actual) number of hours of sunshine to be useful eg if predicted number of sun hours > 5 then lower the weather comp curve a bit. Ditto for rain but wind is always harder to quantify eg how do you turn cyclonic then SW veering NW 5-6 decreasing 4 later into something a computer can use? Calibrating the building's thermal characteristics might not in fact be that hard, and probably only needs to be done once (unless you have the builders in). Most if not all of the key data is already available, energy delivered to the building, rate of change of indoor and outdoor temperature etc. Without even doing any sums, I know that my building is slow to heat up, but equally it is slow to cool down.  

That said, most of the time my weather compensation plus auto-adaption script works just fine, the building stays at a constant indoor air temp, unless I have an overnight setback, when there is an overnight dip, but of course that is meant to happen. What matters then is managing the recovery boost, which is done by my auto-adaption script. The times when my indoor air temp does fail to stay where it should are almost always in severe cold weather, and the failure isn't down to a failure of control, it is down to a failure in heat pump output (see the Bleak Midwinter thread for details). It may be that the system I already have is already 'good enough' and any further tweaking will move into the realm of diminishing returns.

I do realise that not everyone will want to do the sort of tweaking I do, but at least I have established it is do-able, for those who want to give it a try. I also realise I haven't solved the getting the weather curve/control logic right with minimal effort problem, I still had to tweak mine over weeks if not months to get it right, and I still find myself adjusting it by very small amounts every now and then!

@sliderule you are quite right to post here about this, even tho' it's clearly unrelated to thermal losses in the DHW supply at @econonsense 's site.

When you next respond, could you please two steps back and start with an overview of what issue you think the AAA feature is intended to solve?
... then tell us why you're mentioning a zone pump - preferably with a diagram if Mitsubishi have one you can paste here

... and then describe that DIP switch in the Flow Temperature Control box (photo?)
what it's for
and why it needs moving.

Links to online references on Mitsubishi's own site would be even more welcome 🤗 

That will leave this topic to remain on-target to address the heat losses and excessive costs for @econonsense

Many thanks.