Hi Frank,
Very nice data and testing, thanks for all the efforts! May you share few more details on linearity testing methods? You measured 5440 output with both meters and calculated difference between 3458A and 344xx readings? In return I'm interested of doing similar test with other meters, like K2001/K2002.
You might also use the 5442A directly, as it's very linear already. I tested mine to about 0.2ppm INL, between 1V and 10V.
It's an old device, so the negative polarity deviates in offset, gain and linearity. Everything is inside the 0.5ppm linearity specification, but it could do better.
Therefore, for high accuracy, I measure 3458A and DUT in parallel, both polarities, 0.1V interval, from -10.5 to +10.5V, so avoiding range switching of the 5442A.
Then, I take 3458A as X, DUT as Y, and let excel calculate a best fit straight line.
The regression parameters give a correction for the 34465A data regarding offset and gain factor.
The difference between these corrected data Y' and Y is the INL data, and its maximum by definition is the INL specification.
See xls file, including linearity calculations for 5442A also.
Frank