A minor clarification:

wrote:
I am both of those and I assure you that the time of sunset is well-
defined and readily predicted for any location to a tiny fraction of
a second.

Well, not exactly to the second. Even the USNO only reports sunset time
to the nearest minute (see the "Accuracy of rise/set computations"
section in the link below). And their calculator only accepts Lat/Lon
to the nearest 1/10th of a dagree. So the error in the calculations are
+/- a minute or more. This is a technical point, but it is important to
remember that all measurements have some error.

You may have difficulty observing the position of the sun. Like,
it might be cloudy, there might be a mountain in the way, or
you might be indoors. But that is not relevent.

You are correct that the time of actual sunset is moot, since nobody
regularly observes and records this. The only relevant time is the
official prediction. This prediction is normally in the pilot's favor,
since the sun sets earlier than predicted at high surface elevation and
high temperature due to recuced refraction in the less dense air.

Any regulatory requirement should (and I daresay will) be based
on that well-defined time, or the equally well-defined times of civil,
nautical, or astronomical twilight.

http://en.wikipedia.org/wiki/Astronomical_twilight
http://aa.usno.navy.mil/faq/docs/RST_defs.html#top

Obviously you may be safer flying a few minutes after sunset in
a clear sky than a few minutes before sunset in a cloudy sky.
But that is a practical consideration.

You are correct that these definitions are based on certain
assumptions, but the point is that a rule that says you are
supposed to be on the ground by sunset, the end of civil
twilight, or whatever, should be a rule that uses the defined
time, not some local observation. That gives you a clear,
unambiguous requirement.

--

FF