Quality metrics to detect bad (and good) flats and cals

[bberkeyU]

Beyond a meta-data inspection, are there any quality metrics we should compute to decide if flats or cals are high quality?

This question is motivated by the fact that the grid images are bad flats to the human eye, but the pipeline line sees them as ok flats. We have a couple of known fixes for the grid images: don't process epochs for the past data, and plan to add meta-data to track when we ask for the grid to be in the beam. However, is it worth developing detection methods to find other similar data quality issues?

If we wanted to find something like the grid images, we would compute an FFT on the image and compare how the power spectrum matched a nominal good flat or cal image.