In checking on the reference cavity transmission this morning, I noted that it took a
dive over the last day in particular, and over the last three days in general.  Also
attached is a plot of the IMC locked status.

Looking over the trend data, it appears to me that the reference cavity transmission
suffers when there are long periods of time when the IMC is trying to acquire lock.
My hypothesis is that the IMC trying to acquire lock means that the VCO frequency of
the double-passed AOM swings from one extreme to the other.  This might result in
heating of the AOM, which in turn changes the alignment of the AOM with respect to
the input beam.  We don't observe an alignment change on the nearby EOM because the
EOM is too close for a small alignment change to be noticed.  Of course we observe
the alignment change on the iris located in front of the reference cavity, since that's
some distance away from the AOM (~2 m).

Also attached is a plot of the reference cavity transmission and VCO frequency.

One might ask, why didn't you just fix it by re-aligning the AOM previously.  And the
convenient answer is that the AOM alignment is quite sensitive.  This doesn't necessarily
fix the problem if overheating of the AOM is the culprit.