The timing and bias scans for the EPD in run 2019 need to be repeated, because the new algorithm loaded into the QT32Bs had a bug.

Even when that is fixed, we know there will remain a mysterious 20% "gain" reduction in both the QT32Bs and QT32Cs.  Given how nontrivial it was to find the QT32B bug, if this remaining mystery is likewise subtle, we may not find it.  We want as much information as possible, however.

Furthermore, the 20% reduction might be due to gate timing non-optimization.  In our original scan, the beam was such crap that we basically said "it looks a lot like the 2018 timing curve, so let's just use the 2018 timing settings."  Those were GateStartDelay=32 and 23 for the QT32Bs and Cs, respectively.  With that in mind, let's look at the 2018 timing curves for two channels:

Figure 1 - MIP peak position as a function of StartDelay for one tile read out through a QT32B and one through a QT32C.  The timing parameters used in 2018 are indicated.

Clearly, the curve's structure is such that a 10-ns slip can cause an effective gain change of ~10-20%.  If the 2019 curve is a bit different than the 2018 curve (and we couldn't tell very clearly because of the crap beam), and we use the 2018 parameters, this might be the cause of our 20% "gain loss."  However, if this is the explanation, there is no reason that channels in QT32Bs would have the same fractional loss (20%) as those in the QT32Cs, except for "coincidence."  And that's kind of an explanation-of-last-resort.