In 20 minutes the 95% error is 2.1m
In 31 minutes the 95% error is 2.6m
Next I waived the unit around to break
any automatic position averaging
At 36 minutes, the plot has gone around
in a loop (green), 95% error is up to 2.9m
At 48 minutes, the plot has settled
down again at about the same place as the early run (red)
(Notice the satellite "trails" as they
move during the tests)