Thinking through this again, and trying to place fixes at various different levels, I came to what is likely the best solution for now:

 constructDark.py /project/shared/BOT \

        --calib /project/shared/BOT/rerun/cslage/PTC_LSSTCAM_New_12606/CALIB \

        --rerun /project/czw/DM-27158/run1016a \

        --batch-type=smp  --cores 1 \

        -c isr.doCrosstalk=False isr.overscan.fitType=MEDIAN_PER_ROW isr.overscan.order=1 \

           isr.doEmpiricalReadNoise=True \

        --id expId=3020100800045 detector=166

Setting `doEmpiricalReadNoise` forces ISR to ignore the camera defined read noise levels and independently estimate it from the data.  This provides a better variance to the CR code, and it then flags a much smaller number of pixels (132 CR found on 3020100800045).

Once DM-27010 is complete, we will be able to supply a PTC dataset to ISR, and pull gain and read noise values from there instead of from the camera definition.  I think this means that creating initial calibrations for a camera is going to need to iterate:

1. Initial bias, dark with `doEmpiricalReadNoise=True`.
2. Initial PTC to generate more reliable read noise estimates.
3. Second dark with `doEmpiricalReadNoise=False` supplying the initial PTC as the source for the read noise.
4. Possible second PTC using second dark if the empirical read noise and PTC measured read noise values are significantly different.

The change itself is trivial, and it looks good if it helps constructDark. And I also see that you outline above an iterative process that is self-contained within the DM code to deal with this via doEmpiricalNoise. The more general question is, given that we have gain, full-well capacity at saturation, and noise values from the latest BOT runs (both using eoTtest and DM tools), can't we just update all these YAML files with those more realistic values? Maybe this is a more general discussion.

I ran run 12639 with doEmpiricalReadNoise=True.  It hasn't finished with all of the PTC analysis yet, but the problematic CCD 166 successfully generated the dark, flat, and defect calibration files.  I also took the resulting empirical noise values that were extracted and compared them to the noise values measured at EOtest.  They correlate well (see plot) although not with a slope of 1.  Interesting that the E2V and ITL detectors have slightly different slopes.  Both measures agree that detector 166 amp C14 is the worst one.

I've updated the `obs_lsst` readnoise for this chip as well, using an estimate from a few exposures.  That should solve the issue without any extra effort, at least until we can fully update the gain/readnoise/saturation values.

Run 12639 has now finished with the PTC analysis, and CCD 166 successfully ran all the way through.  I attached a log-log PTC plot, and you can see the high noise, but at least it ran.  I agree this issue is resolved.  We can just use the doEmpiricalReadNoise=True option until we have valid values populating the camera.