These are the central regions of various flats divided by the pre-2005 shutdown averaged flats. They are in chronological order. The scales are all the same and very tight (0.99 to 1.01): you can see the noise difference between the two averaged flats and the unaveraged dithered flats, and you can't see any variability in the depth of the snot.

The b1 flats show the same trend. But the features which appeared during the 2005 shutdown did not completely disappear, so neither the pixflat-00001 averaged flat nor the pixflat-53678 averaged flat match the current images terribly well.

MJD 53678 (post-2005 shutdown) averaged dithered flats over pre-2005 shutdown averaged dithered flats:

MJD 53921 (pre-2006 shutdown) dithered flats over pre-2005 shutdown averaged dithered flats:

Note that the features are really not substantially different from the 53678 averaged flats, it's just the noise which makes them look so.

MJD 53981 (post-2006 shutdown) dithered flats over pre-2005 shutdown averaged dithered flats:

MJD 53999 (last night) dithered flats over pre-2005 shutdown averaged dithered flats:

The bad column is from a broken fiber. That and the noise and the low-order extraction errors (the "folding" effect) show why 15-flat sequences and averaged flats are desirable. Oh, also the not-visible-here fringing, etc.