...
Each histogram shows gaussian spread with RMS ~2.5 mm, but the first and the last segment have also substantial systematic bias ~4 mm:
The reason for this bias was the error in ground asymmetry calibration procedure, which was later fixed in the on-orbit code: the 10-bin histogram of longitudinal position (each bin corresponds to a longitudinal segment) for each crystal was filled using asymmetry value for each event as a weight. The average value in each bin was used as a calibration constant, but it works properly only if events are uniformly distributed within a bin. This is not the case for the segments 1 and 10, because the event selection required that particle track crosses both top and bottom crystal surfaces at more then 30 mm from the crystal ends, while bin size is 27 mm. Together with angular distribution of the particle tracks this selection makes part of the segments 1 and 10 empty and thus introduces systematic bias in average coordinate of events in these segments. For on-orbit code this problem was fixed by histogramming only the difference between actual asymmetry value and average linear function with slope 0.0022 per mm - this way the systematic bias was decrease by factor of 10.
In reality the influence of this bias in ground calibration is even more significant, because for segments 0 and 11, which cannot be calibrated directly (because of spread caused by the effect of direct light) the calibration constant was linearly extrapolated from segments 1 and 10 and the bias of 4 mm become 8 mm in segments 0 and 11.
The error mentioned above was the biggest, but not the only source of systematic errors in asymmetry calibration. When I compared the calibrations based on data collected in August 2008 and in October 2009, the difference shows systematic behaviour with position along the crystal: