-- Description for the plots --

(Basic idea) 
If the bunches behave normal, the number of counts in each detector at 
certain bunch should be proportional to the bunch luminosity, 
and the ratio of the number of counts between certain bunch and the 
comparison bunch would be the same (to be exact, the ratio is subjected to
sin phi function, here phi is a azimuthal angle of the CNI detectors).

The 6 ratios are fitted with a constant + sin(phi) curve. The chi^2/ndf of 
the fit for every bunches are calculated and plotted in the two forms.
 o projection 
 o function of bunch id
   - The plots can be identified with run number and fill-id. 
   - Apparently the color of the histogram tell us the ring.
   - Chi^2/ndf ranges 0-20, overflows are shown in the statistics box.
   - both files contain the full set of polarimeter measurements result 
     in 41-pages.   

(Detail)

-- run numbers
Only the runs with flag-0 in CNI polarimeter spread sheet are used.
Used runs are from RUN-222 after which spin bit pattern is archived on
the CNI data. There are three types of runs.
  (1) runs without 0-pol bunches (run 222-757) 
  (2) runs announced to have 0-pol bunches but actually they did not
     (run 759-828)
  (3) runs announced to have 0-pol bunches and they actually did
     (run 829-end)
The comparison bunches are chosen in different manner with respect to 
the different run types.

For run type (1), we need good comparison bunches for both signs. On 
selecting the comparison bunch, bunch sampling starts from (+) 21,25,29... 
(-) 40,44,48... taking the ratio between them then if the Chi^2/ndf is 
under 2 the bunch pair is adopted, otherwise trying different combination 
and iterate until the chi^2 satisfy the condition. Sampling proceed by +4
which reflects the different spin pattern in two rings. The ratio for plus
(minus) spin bunch is taken with minus (plus) comparison bunch.

For run type (2)(3), at first the good comparison bunch pair is taken from
unpolarized bunches (21th and 41th). The ratio is fit with flat function 
assuming they would not show angle dependence, and if the Chi^2/ndf 
evaluation failed, the polarized bunches are used as in the case (1).
This time the sampling starts from (+) 23,27,31... (-) 43,47,51... due to 
the change in spin bit patterns after inserting unpolarized bunches.   

-- Noise corrections
In order to suppress the effects from noisy strips, the background noise 
is subtracted from the detector counts assuming a flat background distribution
with respect to the bunch id. The size of the noise has been estimated from
the number of counts in abort gaps (average of 5 abort gaps).