Quote@THH You are completely right a 1% calibration error can statistically/mathematically disappear. A such small calibration error is just important if Your COP is within the same low percentage range...Recombination energies are "constants" with a slight dependece on Temperature. Thus the error propagation is linear...May be You can once explain us, in which formula you see an exponential error propagation. May be You consult the good old Wilkinson for an example...
This would require detail, specifics, and patience to answer. E.g. a thread looking, in detail, at the F paper, all its assumptions, which are known valid.
At this level of generality I'll address your assumptions:
(1) You assume some exponential error mechanism is needed to get large errors on total energy. That is untrue, I was not positing such, you assume it.
(2) You assume that because some of the known errors here behave linearly, that they all do. I can answer this (for one example) by supposing an error where lack of mixing alters recombination and is determined by non-linear changes in flow patterns of liquid within a cell. This error may not exist - but you cannot, atthis level of generality - show it does not. More productive would be to look at specifics where they will still be unknowns but more to work with.