The details will largely depend on the amount of flux that we are able to get from all of the lamps. The plan is to take short exposures at the start of the run with different lamps in order to estimate the fluxes from all of the lamps. 

In case that every 10th fiber is illuminated, we would be able to defocus the spot by moving the slit by +-2.5 mm. This would give spots that are around 30 pixels wide. That should be sufficient for analysis, especially if we have larger spots from dcb data.
Taking the exposure time function from (https://github.com/Subaru-PFS/pfs_utils/blob/cd4274d68bd5c02437946f9e883fa13bdbebe878/python/pfs/utils/ncaplar.py) it seem that we need around 30x more exposure at that particular value of defocus, compared to the exposure time in focus. So, for two images from each side of defocus, it is roughly 60x more than exposure time in focus. 

Flux levels of the calibration lamps at Subaru are presented in flux_of_lamps_red.png.

Unfortunately, the levels of flux are quite faint. Around 30 seconds are needed for Argon to reach 10k to 11k of flux. For Neon, the relevant line reaches 14 to 15k of flux in 15 seconds. Krypton level is similar to Argon.

Taking images at +-2.5 mm requires 34x more time than in focus. This pushes one exposure of Argon to 15 minutes. This is in my mind acceptable, and I would avoid longer times anyway due to cosmic ray contamination. Assuming 45 seconds of overhead, one set of images taken at (+2.5, focus and -2.5) will take 0.61 hours for Argon and 0.325 hours for Neon. This is just under 1 hour for one set of Argon and Neon images. I suggest not taking Krypton images in order to save time. Argon and Neon span the whole detector and the cost of taking Krypton images is prohibitive.

Continuum is also relatively faint, but because we only need to take it in focus, it is feasible to get good quality in a finite amount of time. 5 minutes exposure should give fiber fluxes with roughly 25k of flux. 

Assuming that we take 6 continuum observation of 5 minutes, and require 10 fiber configurations to take all the defocused data, this experiment lasts 9.875 hours. This assumes 45 seconds overhand per observation and does not take into account time to move black spots.
As mentioned above, I hope that this experiment can be done in 10 hours, which (I believe) is within the scope of one night of observation. 

In the future, in case one wants to measure just changes of frd, I believe only taking one side of defocused Neon data is sufficient (cca 8 minutes).  We should probably also do an experiment of changing illuminations while taking data at lower levels of defocus to test how well can I/we do.

Few comments on fluxes in blue (flux_of_lamps_blue). Neon flux seems acceptable. I am confused that the relationship of the flux with time for HgC seems linear. Continuum in blue is faint. For the exposure time that I recommended above (300 seconds), blue continuum reaches peak at around 3.5k counts in the red part of the blue detector. 

I have added `flux_of_lamps_red_lin_Sep18.png' and `flux_of_lamps_red_log_Sep18.png' which show the measurements of flux with data taken on Sep 18 and Sep 17. Full lines and points are measurements from Sep 18 (visit_id from 67707 to 67796) and dashed lines are fits to the data from the previous night (visit_id from 67627 to 67650). 

1. In general, the measurements are brighter by cca factor 2. I am getting that maximum fluxes measured in Argon, Krypton and Neon are between 2 to 2.4 times brighter, the continuum is 3.3 times brighter and HgCd is 1.7 times brighter.

1.b) example_ratio_flux_Sep18_to_Sep17.png shows the example of the ratio of flux for a section of Argon data, both taken with 15 seconds. We can see that there is a wide range of values of how much has the flux changed in each fiber, and this value is in general somewhat lower than the value just reported above (maximum flux for Argon changed 2.2x times)

2. Continuum fluxes for exposure which are 1 second and 2 seconds long are almost exactly the same in both datasets. Is this a possible indicator that might be helpful to understand this behavior?

3. Continuum has roughly 300 to 400 counts per second maximum on red detector, in pixels which are summed along the horizontal direction. If I understand correctly, that is the value that Jim and Erin are plotting in continuum_counts.pdf and which should be around 6000. If this is all true, it seems that we are around factor 15 below the level projected.

This ticket has been superseded with PIPE2D-902. The future data acquisition plans for the next run (November 2021) to be placed in a new ticket.