...
Pixels do not switch from gn0 to gn1 when they should. If I look at my results (text) files for fsi10, I notice that at low light, there are a lot of pixels in gn1, which is dubious. At high light, there are a few pixels at gn0 which is ok. The behaviour of the gain-choice seems to depend on the rowgroup too, as if the wiring for a given rowgroup is faulty and is not receiving VrefDB. It may be worth altering IFBiasN (as AM says) or ICBiasP to alter this behaviour.
This is an open issue; I Sedgewick has seen it and has said it's what we expect to happen so it's a closed issue too.
The script wn_multigain_graph.py also prints the intercept of the fit-line with time=0, which is the theoretical dark-frame, aka Pedestal. The extra features CDS and CMA will require different Pedestals.
...