[Wavytone]

Long ago in the days of old-school chronometers you only correct a clock against a known time standard once at a known time and date, then let it run free for months (possibly a year).

Subsequently at a later time you then compare the unknown clock against the standard to work out the rate at which it is drifting, and repeat this numerous times, eg weekly or monthly.

When using the unknown clock to make a real measurement of an event, you then apply a correction (the offset at the last check, plus the drift rate times delta T) and again some time later compare the clock against the time standard to check the drift rate is still the same.

This avoids having to manually try to synchronise clocks which is virtually impossible to better than 1 second, whereas relative comparisons can be made by audible ticks or visual blinks to much better than 0.1 second and possibly 0.02 second.

The same should be done for computer and portable digital devices between synchronisations to say NTP because most are not all that good - it’s not uncommon to see an iPhone or iPad off by a few seconds when compared to atomic time as broadcast by radio, or the 1194 time service.