Re: Fwd: [Aavso-photometry] pipeline

[Stupendous Man <richmond@stupendous.cis.rit.edu>]

  Tom wrote:

> One thing that I have learned is that if you try to be perfect, nothing
> will come out of the end of the pipeline. ...

  I'm working on reducing some of the accumulated Mark IV data right
now, and one thing _I've_ learned is that it's (almost) impossible to have
too many cross-checks.  One of my goals is to figure out which
of the data was taken under the best sky conditions, so that I can
concentrate on it (for some purposes -- for other purposes, one
might want to look through all the measurements).  I _thought_ that
I'd found a good way to characterize each night, or portion of a
night, as "good" or "not good" (see section 2 of TN 97).
Earlier today, however, when making a graph of sigma-vs-mag from
a section of "good" data, I discovered that a bunch of measurements
were way, way off the rest, by 0.3 or 0.4 mag!  Good thing I 
made that little sanity check.  After a little digging, I found
that all the bad measurements came from the I-band camera on
one particular night (the latter half of JD ...2817, June 26, 2003).
A little more checking showed that my "night-quality" tool 
actually _had_ shown this night to be abnormal, but I had
not noticed and accepted it as "good" anyway.

  Double-check, triple-check, and check again.  I think this is 
especially important when dealing with data from an automated 
instrument, or, in fact, any data which one has not gathered
oneself.  

  I will be going through the data Tom collected in the summer and fall
of 2003, looking for systematic errors in photometry as a function
of position in the focal plane.  My first efforts on TOM1 for June, 2003,
show the same sort of pattern I noticed in earlier data (see TN 97, again).
I'll be checking TOM2 and TOM3 for such effects as well.  I'll
try to make some empirical correction for the errors, then re-reduce the
data, and see if the scatter at bright magnitudes diminishes.

  It will take a while -- my poor old computer requires about six hours
to chug through the main bit of analysis for a month worth of data
from one unit -- but I'll try to give little updates as I go.  

                                           Michael