experiment (multiple times) or even a preloaded experiment (never been
swapped). Well, if the first thing you see in an interval is a
swapmod, then you have no idea if the experiment was actually swapped
in at that time, and counts as an experiment using its nodes! So,
have to go back before the interval and look at the previous stat
records. As Mac points out, a better solution might be to store this
info in the stat record itself, but I want to think about this first.