Several factors contribute to the uncertainty in the relative QE
measurements. Photon statistics contribute 
for the 1024
individual superpixel quantum efficiency ratios, but only
to the full CCD nominal value (assuming spatially uniform QE), where
the nominal value is taken to be the centroid of the Gaussian fit to
the histogram of 1024 values.
Reproducibility of relative QE measurements made after moving the CCDs out of and then back into the chamber is typically within 1%, but is observed to be as poor as 3% at Cu (8.04 keV), possibly due to relatively high uncertainty in the pileup factor for the reference CCD caused by a high count rate at high energy.
It should also be noted that the pileup correction parameters are generally similar from one FI CCD to another, but they do seem to be slightly different for one older reference CCD (w103c4) used to calibrate some flight CCDs. Because older electronics were used for this CCD, the depletion depth is thought to be smaller, which would exert some influence on the pileup rate. This difference has not yet been incorporated into the correction codes.
A further test of the accuracy of the relative quantum efficiency measurements is described in Section 4.7.3.