[torstenvl]

> Because TAI is not based on the Earth’s rotation, it’s not ever-so-slowly changing. It’s the measure of time against which UTC’s watch is occasionally correct. That correction is called a leap second: a 61st second that is sometimes added to particular minutes in UTC, like the very last minute of December 31, 2016. As of January 2019 there have been 27 such leap seconds inserted.

This seems backwards. Perhaps I'm misunderstanding what the author is trying to say, and another poster could help clarify?

If UTC is based on the earth's rotation and TAI is based on cesium-atomic timekeeping, and UTC is slowing down, and TAI is "the measure of time against which UTC's watch is occasionally corrected," then how can UTC have leap seconds? Delaying for an additional second on December 31st would only make it one second even slower, exacerbating the problem.

I don't see how adding a delay to the slower clock is going to put it in sync with the faster one.

Is it instead that UTC is TAI, plus leap seconds to slow it down to keep it in sync with the rotation of the Earth?

[rndntr]

It's the other way around - UTC is based on TAI plus an integer number of seconds offset, and this offset is adjusted (by the insertion of leap seconds) to keep UTC roughly in sync with the Earth's rotation.

[Asooka]

The delay is added to make UTC slow. The Earth is the slow clock, UTC ticks at the same pace as TAI, so sometimes we add a leap second to UTC to sync it with the Earth.