Here I calculate what we expect for the RMS in Larry's q and u plots just
posted this week.  I find that he is getting errors about four times
bigger than those expected from photon noise.  A likely culprit is
intra-cycle-pointing-drift.

Lero's GIF of M82 that he just sent me shows
that the peak flux is 15 Jy per 9 arcsec beam, and the size to the FWHM is
about 18" by 45".  This is Larry's "source".  (Contours are 1, 3, and 10
Jy per beam, according to Lero.)  Thus, Larry's source is 4 x 10 pixels,
but since its not rectangular we expect it to be about 30 pixels.  Since
there are 30% dead pixels, we expect about 20 pixels which is about what
Larry gets (he gets about 23 on average according to his memo).

The average flux in Larry's source is somewhere between (15/2) Jy per beam
and 15 Jy per beam.  Lets use 10 Jy per beam.  A 9" beam is 64 square
arcsec, which is 3 SHARC-II camera pixels.  So we have 3.3 Jy per camera
pixel, on average, for Larry's source.

This is 7 times the brightness required for 1% SHARP polarimetry in 5
hours, when binning pixels in 2x2 (see Table of Specification posted to
instrument section of public SHARP web page).  Thus we should be able to
achieve 1% polarimetry in (5/49) hours = 6 minutes.  However, since we are
combining 20 pixels, not 4 pixels, this gets us to 1.2 minutes.

However, for tau = 0.05 weather (and a x 25 multiplier to get to 350
micron tau), EL = 40 is 1.7 times lower transmission than EL = 60, so we
should derate the sensitivity for this.  Then we need 3.5 minutes to get
1% polarimetry.  So in an 8 minute "cycle" we should be able to do 0.7%
polarimetry.

I took the standard deviation of the first 20 of Larry's values of q (they
look better than most of the others), throwing out two outliers, and I get
3.0%.  This is four times higher than what we should be doing based only
on photon-noise errors.

However, I am not too worried about this because there is another factor
to consider which is the errors due to pointing drifts.  According to my
estimate on the web-logbook (Dec 3), these are 10% per cycle.  By binning
20 pixels we reduce this a lot, but it would be surprising if we could
achive the 0.6% per cycle expected from photon noise only.

Still, this does mean that we have not yet proven that SHARP can achieve
its sensitivity spec.  For this I think we may have to analyze a fainter
source than M82 where the photon noise per cycle should dominate the
pointing-induced errors.