I've attached an image that shows the locations of those pixels that record a number of counts = (2*n-1)*128.
The image is the sum of 200 binary images where pixels are given values of 1 if their number of counts = (2*n-1)*128 and 0 otherwise.
The excess of counts is clearly coming from the left hand tap. This is good news because the two taps have independent ADCs and it suggests that it is only a malfunctioning ADC on the LHS that is giving us this problem.
I plotted a histogram of the total intensity of the Hartmann sensor when illuminated and found that the 128 count problem extends all the way up through the distribution. This isn't unreasonable since that digitizer is going to be called on mutliple times.
First things first, the value of 128 equals a 1 in the 8th digitizer, so for a 16-bit number in binary, it looks like this: 0000 0000 1000 0000 and in hex-code 080
0000 0000 1000 0000
The values of the peaks in the attached distribution are as follows: