A problem in the AIPS calibration of EVLA data has been found.  Rick
Perley found the problem when he noticed baseline-dependent offsets
in the calibrated amplitudes of a strong calibrator.  He then repeated
the calibration omitting FRING and using BPASS to correct for the delay
errors.  That treatment did not produce the baseline-based offsets.

The bug was put in the code in September 2018 when the archive began
offering frequency averaging while downloading data.  If there is an
uncorrected delay error, such frequency averaging will cause a
reduction in fringe amplitude as described in AIPS Memo 90.  A
correction for any post-correlation frequency averaging has been
part of AIPS calibration since 1995 for the VLBA.  The EVLA was
added to this in 2018.  Unfortunately, one of the two corrections
applied to VLBA data does not seem to be correct for EVLA data.
This correction for "segmentation loss" in FX correlators is made
even when there has been no post-correlation spectral averaging.

This correction is larger for larger delays and amounts to a few
percent normally.  Unfortunately, this correction is a baseline-based
correction rather than antenna based and so is not undone by self-cal.
For data of high signal-to-noise, AIPS tasks BLCAL or BLCHN can be
used to correct the baseline-based error.  The EVLA these days
normally has rather small delay errors in 8-bit data, while 3-bit
data at some bands and some antennas can have delays of 10 nanoseconds
or more.  The bug produced an amplitude error of 6% for a baseline
between two such antennas.  This segmentation correction has been
removed from AIPS versions 31DEC22, 31DEC23, and 31DEC24.  A
"midnight job" ($HOME/do_daily.<hostname>) on any of these versions
will update your version of AIPS.

Note that the MNJ will also get a correction to VLBATECR and VLATECR
for TEC files to be used on data taken in 2024.