Hi @zorgtool, I did some more analysis of the sensor data. I use Wolfram Mathematica for the analysis - but similar things could easily be done in Python.
Method: I loaded all raw values, determined min and max values and then scaled them to range from 0-1. This greatly amplifies minute differences (usually less than 1% at the very highlight end of the range, so very hard to distinguish in real-life scenarios).
Results for RED channel of the Bayered data per ISO:
Results for GREEN 1 channel of the Bayered data per ISO:
Results for GREEN 2 channel of the Bayered data per ISO:
Results for BLUE channel of the Bayered data per ISO:
The green channel seems the most uniform. The other channels have interesting patterns which are probably caused by the electronic layout of the sensor. If you look closely, the patterns actually shift somewhat along the shots. The shift is rather big, otherwise I would have somehow attributed this to the pixel-shift abilities of the sensor. Now I’m very unsure…
The LENR just introduces noise (dithering?) in particular for ISO > 1250 or 2500.
Despite this, I think I have enough information to update
camconst.json with proper white points.