thoughts on redshift completeness / SNR tests for BGS sample

[moustakas]

Note from David Weinberg [desi-bgs 902]:

  Here are some thoughts about the spectroscopic completeness tests,
after looking again over the informative systematic plots that Anthony has posted at
https://desi.lbl.gov/trac/wiki/BrightGalaxyWG/RedshiftCompleteness

  The questions that have the biggest impact on strategy decisions are
at least somewhat separable:

1. What is the spectroscopic completeness to r=19.5, r=20, as a function
   of some best (meaning most informative) choice of SNR?
2. What SNR can be achieved in a 3-pass survey over the 14k deg^2 footprint
   in the available time, with some margin (maybe 10%)?

The second question depends on survey simulations including moonlight,
but it doesn't actually require spectroscopic completeness simulations.
Instead it requires adopting some target SNR, and probably some maximum
exposure time (e.g., 1200 sec), then running BGS simulations to see
what SNR threshold can be fit within the available time.  An important
element is that we want constant SNR after accounting for Galactic
extinction, so varying with (l,b).  Probably Kirkby is the one to
do these simulations.

It is almost inevitable that we will live with this SNR no matter what;
the only available trade is to reduce the survey area to increase the
SNR if we decide that the spectroscopic completeness at 14k deg^2 is
too low.  The spectroscopic completeness simulations tell us what we
can expect to get at the achievable SNR, and they inform strategy decisions such as:
4. How do we set the separation between Priority 1 and Priority 2?
5. Do we further shape the sample beyond magnitude-limited, based on
    color, surface brightness, morphology, etc.?
6. Do we reobserve Priority 1 galaxies that do not yield a successful redshift?

Our current status based on the Moustakas-Tinker completeness simulations and Dawson strategy simulations is:
a. An SNR that corresponds to a 300-sec exposure under the fiducial
   bright-time conditions can be achieved in 3 passes, 5 years, with
   about 20% margin.
b. The spectroscopic completeness for this SNR is high:
   97% for r<19.5 and 92% for r=19.5 to 20, rising to
   99% and 96% if the exposure time is doubled (e.g., from reobservation
   of failed targets).

I have more to write on how I think the completeness tests can be
shaped to be most useful for addressing questions 1, 4, 5, 6,
but I'll send this now so it gets out ahead of today's telecon.
An important early question is what SNR choice is the most useful.