Hmm, for the docs, what about instead adding an east/west note on the timedelta object, something like:

'if the time being compared is east of UTC, timedelta is positive. If west of UTC, then timedelta is negative'

and keeping the east/west removal under the offset function? This way, other functions that refer to timedelta (like dst(), which doesn't have anything on east/west, but do in the docstring) can all refer to one spot.

Regarding the docstrings, I'm pretty indifferent and would be fine with keeping east/west as-is if it reads better there.