...and we don't necessarily always have a clear picture of the limitations of the proxy...or even of the proxy of the proxy of what you are really after. A consumer scale may not be corrected for gravity's variation, but a commercial/scientific one is.
@jbriggs444 I'm not sure if I agree with your examples though because I consider internal functions of the device to be a separate issue from what "thing" is being applied to it to measure. A scale that reads out in mass is nevertheless *actually* measuring force, regardless of what the readout units are. We often use such scales for experiments to measure force and then back-convert if needed (if English, no conversion needed), for that reason.
Let's explore your wind example. The wind vane, if I understand them correctly, is directly sensing wind speed. The rotation rate of the vanes is motion and even if converting motion from linear to rotational, that's not really changing anything relevant. My understanding is that because it is a direct measurement, it is not subject to variation due to density (though I'm not certain).
But consider instead a pitostatic tube. It measures (is connected to a device that measures) pressure, not speed. In my industry, sensors often come with multiple input devices to the same readout device and include multiple modes for converting proxies into what you really want to know in the readout; pressure, velocity, or volumetric flow rate can all be read from the same pressure measuring device.
But even for HVAC, where we do most of the final work in terms of volumetric flow rate, that's still another proxy for what *really* matters most of the time: mass flow rate. That can be a problem if we rarely think about that second-level proxy and then have to deal with a change in air density.
My understanding (not totally certain) is that they deal with the issue in airplanes from the opposite direction, by accepting that the proxy - airspeed - is more a reading of mass flow rate than speed, even though it is called a "speed".
There are other fluid speed measuring devices that use doppler shift to measure the speed of suspended particles. We can set those aside for now though because their principle of operation may contain other proxies...
My point either way is that we don't always think about these issues when not in a professional context, but when we are, these issues can matter a lot.
Edit: ...I suppose the internal vs external proxy thing can be a bit arbitrary of a distinction though: my understanding is that multimeters use some internal proxies, such as voltage drop across a known resistor for current.