What an interesting comment! It actually ignited some reflection in me.
My son has a lean body right now, and we often have a hard time getting a sensor to last. When I think of accuracy, we have several cases that come up:
-
About half the time or more, our calibrations are within a very small number of units of actual. It often is associated with a specific sensor, where strings of calibrations always seem good—but not always.
-
About a third of the time, a calibration is about 15 off, and the calibration result will halve that. It does happen quite often that these come in strings too, where, everytime with calibrate a sensor it is 15 off. These strings appear to progressively drift away, but others don’t always drift continually with time (although the all do, just not with a steady rate)
-
A small number of times, we are way off. In that case, I will often do two calibrations within a half hour or so. It is quite rare to be that far off many times on a row.
I have experimented a few times, and forced a calibration on a sensor (twice the same number) when it is somewhat off, by averaging two or three fingerprint measurements and force-entering the result. It has several times resulted in a tack-sharp follow-up calibration. But my number if samples is too small to be reliable.