Suppose a data file consists of a sequence of time averages of an
observation, with an averaging interval dt (dt = 7 days, for weekly
averaging).

The averaged data have a timestamp, which could correspond to the
beginning of the interval, the middle of the interval, or the end of
the interval. I prefer the middle of the interval--then it would be a
centered average, which has some advantages.

If we have a model which uses the data in the file for a forcing as a
continuous function of model time, then we interpolate from the data (which
is averaged over interval dt).