> an instant can't occur on three days. [...]
> I think that it is always the case that there are no more than two dates in use at any one time. [...]
> for practical purposes, every instantaneous event occurs on exactly two different days. Am I thinking clearly?
Not really! Let's keep in mind that time zones span between -12:00 to +14:00. That makes room for three calendar days! Let's try to print the number of seconds since 1970-01-01T00:00:00Z (UTC) "the epoch" in a POSIX-compatible CLI terminal, thus writing:
date -d '2015-11-22T23:00:00-12:00' +%s
date -d '2015-11-23T12:00:00+01:00' +%s
date -d '2015-11-24T01:00:00+14:00' +%s
each of these will give the very same answer, namely
and that is the same moment in time, despite of the fact that there are three different "wall calendar" dates involved.
> the geographical location that you were measuring that time at, and that information tends to be abstracted into timezone offsets.
The geographical location is related to timezones, but geographical locations are different from timezone offsets.
> > refer to the period of time *between* the first instant and the last instant of year we call "1980", inclusive.
> [...] the interval notation would be used.....
> ..... is as close as you will get with the edtf spec: the interval beginning some time during 1-1-1980 and ending some time during 12/31/1980.
would be the duration between (1) some instant during the first second of the year 1980 and (2) the very last instant of that year. In other words, it would be between 31622399 and 31622400 seconds long, depending om "when" (during that second) within 1980-01-01T00:00:00 the measure was made. There were no leap seconds that year.
Of course, we could write
to get an exactly 31622400 seconds long period. Kepp in mind that 1979-12-31T24:00:00Z is the very last durationless instant at the end of 1979, which in this case comes AFTER the 1979-year's leap second (23:59:60 on 1979-12-31).