> How does posix know it’s a leap second? It needs to know one way or another.
Unlike with TAI it does not need to know. The time on the wall is leap seconds resolved (ignore for a second how it is done). Whereas TAI keeps ticking onwards without regard of them, UTC has them cleaned up. So if your end result is that you want to show a time in local time with TAI you need to know where all the leap seconds happened. With UTC you do not need to, you will have re-synchronized your clock plenty of times since to some time source which itself has resolved the leap seconds. You do not need NTP for this.
Even if you manually set your time once a year, you will use a reference source that has already dealt with the leap seconds. Since leap seconds are basically hidden historically too (they do not appear) you are good. You can always divide the seconds since 1970 down and you can pinpoint any date in the past without having to do complex leap seconds math.
