The problem is a lot easier to think about if there *is* a right height -- then the right estimation methods are the ones that include/exclude changes/differences so that you end up with the right height. To get back to the coastline problem for a moment, this may be something you already know: you can design a function that is not smooth at any point (so it's not differentiable at any point), like a fractal. In that case, the "length" of a bounding fractal curve may not be easy to calculate -- however, the function may still be integrable so that the area enclosed by the function is finite (sorry, terrymorse). That's kind of sort of related to what I do: I calculate the area bounded by the curve and that's well-defined even if the curve itself causes nightmares.
You're exactly right, and that's why I said you need an accurate power meter, speedometer, total mass, and a few other things (that's why I warned about "hella wind"). What y'all have been mostly talking about is using one sensor (either a barometer or a GPS receiver); I use multiple sensors. I think (but am not sure) that people thought this wouldn't work because they thought that each different sensor introduces errors, so they thought you'd end up with worse estimates due to all the other errors. The surprise (?) was that the data from different sensors can be combined in ways so that they provide a check or constraint on the other sensors' data so that the combined estimate behaves much better than the estimate based on any single sensor.
Sorry. I *think* I fixed it and it should work now. That said, it's not key -- I just linked to it because it had a few visual examples from long ago.
[Edited to add] Dammit. The link still doesn't work. Hmmm. If you want (not that you need to) you could try copying the link and then dumping the "https:" at the front, but I understand if no one has time for that kind of nonsense.