A number of changes have been made since last week.  In no particular order:

1: If there are any dead strips within the 2-1-2 group of strips immediately around the central strip, the track is ignored.

2: If there are any dead strips within the cluster surrounding the max strip, then this cluster is ignored.  The cluster is defined as the 2-1-2 group of strips around the maximum strip of the central group of strips.  Note that in this case the maximum strips is still used.

3: At Jan's suggestion, I switched from tower energy to track momentum.

4: Also at Jan's suggestion, to avoid having to deal with saturation effects I reduced the range of momenta i was considering to 1.8<p<4.

5. It turns out that I was using the relative gains wrong: I had thought that I had done them wrong originally and so was using 1/ the relative gains in the database, but actually those were the correct relative gains.

6. Because of a lack of statistics I reduced from 20 eta bins to 2.

7. I tightened up the requirements on the strip status.  Before, I was allowing the statuses that were less than good but not automatically fatal, but now only strips with status 1 are regarded as not dead.

8. Another suggestion from Jan: any strip with adc<3*sigma is removed.

As a result of these changes things look better: there is at least a hint of a linear trend.  The next step is to try to fit slices in p to Landaus.  Unfortunately, the slices don't look particularly Landau-like, see Figure 1.

Figure 1: Slices (from eta 5-strip cluster total adcs for |eta|<.5)

As Joe pointed out, though, you can't really expect to see a good Landau distribution because this data is pedestal-subtracted, so you're losing a good part of the distribution.  He suggested that a large number of Landaus added together might end up giving a Gaussian, but these slices don't look particularly gaussian either.  At high energy you might get a more Landau-like distribution, but we can't go up very high in energy because of the saturation problem.  I did try to look at the 4-5 range, but the statistics just weren't there for an improvement.

So, there are two ways forward from here:

1. Try to figure out how to compensate for pedestal subtraction in fitting these slices (possibly by some sort of Gaussian smearing?).  Suggestions welcome.

2. Get more data.  The entire pp500 run has been produced now, and that would increase the statistics considerably which might help with the fitting.  I'm not sure if Matt plans to run his electron tree maker on this data: if not, perhaps he can give me the scripts he uses and I can run it myself.