This set is split into three groups.

Group a is from 2006-04-07 to 2010-10-30
  A new detector was installed in November 2010
Group b is from 2010-12-01 to 2012-04-06
  Group b is poor quality.  We found a way to improve quality in April 2012
Group c is from 2012-04-07 to 2012-08-28

We recommend using different range biases for the three sets.  We also have
explored error scalings, and believe that scaling our statistical
uncertainties (as represented in the normal point files) could be
accomplished by one of the following choices:

Group    factor for scaling   OR   RSS term 
  a             2.0                  17 ps
  b             6.0                  60 ps
  c             2.5                  20 ps

Thus the point:

5120101030123810000000024392228766593170610236   141648B 73152  8948 5320A 251A

which has a reported uncertainty of 14.1 ps would become 28.2 ps under
scaling, or 22.1 ps if applying the RSS value.

Batch data reduction using:

Common treatment:
threshhold = 1 (excluding all shots with > 1 photon received)
no first photon bias
no DAC correction
10 ps RSS term to account for DAC
channel offsets applied throughout, based on windowed time periods

Different treatment:
Group b uses the lun-fid correlation technique to form normal points.
Groups a and c use a functional fit method, except in cases where the fit
failed, in which case the correlation technique was substituted.

An extensive comparison between the two techniques revealed no net bias, at
the few picosecond level.