So here’ my attempt at calculating n from measurements.
The general idea is that radiator heat output is proportional to (T_Flow-T_room)^n = deltaT^n, with n the radiator coefficient. The temperature decay during the compressor-off time is then given by the differential equation d(deltaT)/dt = -k * deltaT^n. If n = 1, the solution is the well-known exponential decay.
We can just plug this into Wolfram Alpha (ChatGPT o1-preview also does a good job here) to get the solution (can we get LaTeX or some form of Markup equations here in the forum? 
)
deltaT(t) = (deltaT_0^(1-n) + (n-1)kt)^(1/(1-n))
I then raised my flow temperature for a bit to get some proper data. Note that the solution depends both on an accurate flow but also an accurate room temperature at the location of the radiators. This becomes more important if room and flow temperature are very close.
Here’s flow and whole-house indoor temperature (I have temp sensors in every room and calculate a weighted average house temperature). I download the data as csv from Emoncms and process it in Python fit_n.py.txt (3.5 KB). This will directly work with emoncms exported csvs.
The fit works nicely and I get an n of approximately 1.18.
As stated before, this is very sensitive to the room temperature measurement. Adding/subtracting 0.5°C and redoing the fit yields 1.29 and 1.07 for n, respectively. I think the value is quite ok for my house, as I have mostly Type 33 and 22 radiators and not all are equipped with fans. Also my pipes are not insulated so there’s some UFH-type characteristic in there too.