[evlatests] Closure oddity -- origin found! (Cancel Red Alert)

[Rick Perley]

    I reported last week on a small (~1%) but easily visible amplitude 
closure error in the 3/8 bit tests.  A brief summary:

    *  All six baselines common to antennas 1, 2, 20 and 23 show an 
excess 1% in amplitude following the usual antenna-based calibration.  
This calibration included bandpass calibration (which shows there is 
nothing unusual in the BP function for those four antennas).  The 
closure offset is the same in both polarizations, and is the same for 
all three data streams -- A1C1, A2C2, and B0D0.  It is also constant 
across the entire 2 GHz-wide bandpass, and is constant over the (short) 
observation length.    It *is* worth noting that these four antennas are 
the four most distant antennas on the north arm:  ea20 at N72, ea23 at 
N64, ea02 at N56, and ea01 at N48. 

    I should have noted the adjacent positions of these antennas last 
week ... for these clearly suggest that the problem is due to slight 
structure in the source. 

    Vivek ran the test again early this morning, this time taking out of 
the array some of the seemingly good antennas.  Quick calibration showed 
the same closure offsets, on the same antennas, despite having the 
correlator configured differently. 

    To prove that it's source structure which is responsible, I 
re-calibrated the data, using only short spacings.  The 'closure' effect 
disappeared. 

    All is good.