The initial implementation assumes that each science exposure is followed by a bias. If this is too burdensome it should be possible to construct the correction from the science frame, but this requires a better model of the readout (in particular the wipe sequence) than we currently have. This would have better noise properties, but would also be more sensitive to temporal variation of the incident light.

I've now implemented the "analytic" version (isr.doBrokenRedShutter=True isr.brokenRedShutter.useAnalytic=True). This is necessary when we don't take enough biases and the illumination is changing — if it's different between the science integration and the readout we don't have enough information to make the corrections.

N.b. the correction is in the ISR, but it should be after CR repair.

For running constructFiberProfiles.py on SuNSS data, I think you want to use profiles.centroidRadius=3 profiles.profileRadius=3. This helps prevent bad profiles on the full-density fibers, which is what caused awful extractions.

(Wrong ticket! This was about PIPE2D-739)

https://github.com/Subaru-PFS/drp_stella/pull/153

I'm not totally happy with this, in particular it should run the analytic version after repairing CRs, and that's not done in the ISR. However, I think it's good enough to merge

Maybe I got it right this time? https://github.com/Subaru-PFS/obs_pfs/pull/90

Addressed comments, rebased to master, and pushed