Some strips in module 103p have very few entries, e.g. strip 15407p:

Figure 1: Strip 15407p

The number of entries in the stat box is the total number of events, ~1.3M in the sample, but the peak is only at ~200.  Compare to strip 15408p, which has a peak two orders of magnitude larger:

Figure 2: Strip 15408p

Strip 15407p has only 1810 entries in total.  From strip 15391p-15419p, all strips with odd ids have far fewer entries than those with even ids, with peaks one or two orders of magnitude lower: see attached pdf for mod 103, pp. 15-16.  Note that this is 15 strips, suggesting a possible hardware problem.

One possibility was that the strips were simply off for part of the fill, but that does not appear to be the case:

Figure 3-5: Strips from runs 10091089, 10092031, and 10092049

As you can see, there is data present for the odd strips in each run, just very little.

Additionally, everything looks fine in the online monitoring: the pedestals for these strips for this fill do not appear to have any problems.  See the online monitoring plots for run 10091089 and run 10092049 (p. 62) and the pedestal qa plot for the associated pedestal run, 10091080.

I tried applying a cut that required the total integral of the distribution to be less than 5000 but that also had a significant effect on module 95p.  However, the situation in module 95p is different: while most strips in module 103p have an integral of 30000 or greater, most of those in module 95p have an integral that is less than 20000.  Thus while the above strips from mod 103p are anomalously small compared to other strips in mod 103p, similar strips in mod 95p compare will to the other strips in mod 95p.  See this plot of the average number of entries per strip by module for crates 7 and 8:

Figure 6: Average number of entries per strip

As you can see, the average number of entries is ~10000 for mod 95p and ~40000 for mod 103p.  Indeed, mod 95p has the lowest average number of entries per strip of all modules in this sample.  Perhaps this means that mod 95p is bad?  At any rate, a strict cut on the number of entries seems to be wrong.  One possibility is to put in a smaller cut -- say, nEntries>2000 -- which would eliminate the most egregious strips in mod 103p without touching mod 95p, but ideally all 15 bad strips in mod 103p would be removed.  Comparing to the overall average would probably work but that would require processing the strips twice, which would be a pain.  Instead, I calculate an approximate average by calculating the integral over the module and dividing by 150.  This is usually within 10% of the actual average, above.  Then I mark any strip whose integral is less than 1/8 of this average as bad.  This method is successful at removing the strips from mod 103p that I wanted to remove without killing any other strips.