I recently read a collection of 8k patches with an I1iSis XL 2 and noticed some peculiar "dents" in the wire frame gamut I1Profiler draws. This is never good.
My target image has registration bars at the top and bottom so I can read the same charts in reverse order. This reads the last page first, then bottom to top and right to left instead of the normal sequence. The result is the same patches but reversed in order.
This also is a way to detect errors that sometimes occur when the same chart is read in the same automated way.
So I read the chart in reverse and looked at the dE76 values. Worst case was over 10. The first 200 largest dE76's were over 1.0.
I've seen this before. It's because of a bit of lint or dust reflecting rather small amounts of light. This doesn't normally cause a problem because most of the spectra have reasonable reflectance. But here's where it matters. Near the gamut edges on the lowish L* side, there are portions of the spectral reflectance values that are only a few percent ( .01 ). Sometimes less than .01. If a tiny dust particle happens upon the focused light, which is only a few mm's across, it can easily reflect enough light to increase the measured reflectance from say, .01 to .02. This may not seem like much but it is because conversion to Lab is non-linear.
So what I did was compare spectral plots from the largest 20 dE readings which range from about 11 down to 6. In each case the reverse readings were lower than the forward readings and, in each case the spectral values affected were the small values below .04.
So, when you have two scans with large differences between them, you should select the ones with the smaller minimum in the spectra since lint/dust will nearly always result in an increase in measured reflectance. OTOH, if the piece of lint happened to be quite dark, it would hardly effect anything. Blocking 1% of the light has almost no affect on dE. But adding 1% is a killer.
So I cleaned the iSis and re-read the 8k charts in both directions. dE on all patches was under 1 and averaged about .10. Comparing the dE between these and the prior bad pair set, but selecting the patch values that had the smallest minimum spectrum reflectance from the bad pair set produced nearly identical results. From a situation where 200 patches all had dEs over 1 to where everything was under 1.
Probably a good idea to always do a second pass (in reverse order if you have the tools) and look for signs of dust/lint messing up readings. But, if one only has the two sets of readings and some of them differ significantly, select the ones with the lower spectral minimum. Overall, this will work better than averaging them.
But best to clean the equipment and re-do.