Earlier quoted context omitted.
What oscillator are they using? Things like the DS3231 (which are temperature-compensated crystal oscillators) only do about +/- 2ppm, which is about +/- 1 seconds per week. You can tune them to do better (mine has been running at about -0.007ppm for a while), but on the tin, they appear to be worse than what you reported. (And they go to great lengths to be that accurate.) My main conclusion is that I want a watch w…
I imagine the actual drift rate is normally distributed around 0 with 2 std deviations within 2 ppm or similar. Maybe I'm wrong. But I recall having not adjusted the clock for longer than a year and it only being about half a minute off. I just like to tell myself I got an accurate one out of the batch :P
Quite possible. Also possible that they are just doing a good job!