"Grey Lists" of IPAC archived references and Suggested Recipe
for Vetting Candidates
Here are the lists of suspect reference images in the IPAC archive that may
lead to unreliable difference-image detections or photometry:

g-filter suspect references
r-filter suspect references
i-filter suspect references

These references will not always lead to false positives (or inaccurate
photometry on true positives) because it depends where your target falls.
The cause of a suspect reference is due to the co-addition of several or
more science images with inaccurate flat-fielding. These flat-field residuals
follow large scale patterns and it's possible that your detection still falls
on a "spatially flat" region and is perfectly usable. "Bad regions" in a
reference image will impact *all* subtraction images (and detections therefrom)
generated at/around those regions at all epochs.

If your target is covered in any of the above lists, it is strongly recommended
you perform the analysis below. Usually step (1) should be sufficient, but you
will always want to run forced photometry on your candidates, at all epochs
where subtractions are available, then compare with the alert packet photometry.

 (1) When you have a list of prime candidates, visually examine their entire
     quadrant-based subtraction and reference images. Examine other "control"
     objects near your candidate and ensure there are no systematic residuals
     in flux on those objects. A flat-fielding error (or gain-mismatch) will
     manifest itself as a flux-dependent bias, i.e., the differential flux will
     be relatively larger on top of brighter objects. If neighboring objects
     show systematic excesses, toss your candidate. If unsure, perform step (2).

 (2) Run forced photometry on each candidate as well as on neighboring
     objects (i.e., your control sample selected in step 1). You will want
     several control objects that span a range of fluxes. Check for non-zero
     baselines in their historical raw-flux time series'. Sometimes, a non-zero
     baseline for your candidate of interest may imply the reference image
     is contaminated by the actual transient/variable flux, depending on
     how/when it was created. That's acceptable since you could perform a
     baseline correction according to the forced photometry documentation.
     However, if your neighboring control objects also show flux biases in
     their historical baselines, it implies there's a gain-mismatch and the
     candidate should be tossed.

*** For longer exposures (>= 90sec), resulting in higher S/N per pixel,
    systematics from imperfect instrumental calibrations (e.g., flat-fielding)
    become more important at fainter fluxes. These systematics are more
    likely to (i) yield false positives or (ii) dominate photometric errors.
    
---
F. Masci, 2020-03-24