In reply to Marek:
> But how does a running watch do the equivalent? It doesn't actually measure power, it can only access some physiological numbers (HR, weight, cadence via accelerometers) and them guess how those parameters may translate in some typical person into 'power'.
Garmin's website gives an interesting level of detail on what factors it's trying to take into account:
https://support.garmin.com/en-GB/?faq=QRiQOEq5d09foNiH1DzUt5
Seems like it's not actually using heart rate but focussing more on the actual motion. Though I completely agree with you that it's ultimately a very indirect estimate that seems unlikely to be coming up with anything particularly accurate.
> Like I said in my first post, if it was tolerably accurate it would be accompanied with error specs and some calibration procedure.
I'm not sure this is entirely true, though. Products marketed at general consumers are very rarely going to prominently feature error specs and will always do their best to workaround or hide any need for a calibration procedure. Some of them will still be quite accurate despite that. My phone and my running watch both do an excellent job of tracking my location, well beyond tolerable accuracy. But neither have any published error specs or calibration processes as far as I'm aware. That's not necessarily because the data's bad. I suspect it's because lots of people won't understand error specs so there'd be no real upside to sharing the data and risk of people just seeing them as a bad thing. And any calibration that requires action by the consumer is just going to be a pain point.