DeadlyLover , Well yes,
R6581 has the LTZ1000 implemented correctly, at 45°C, no wonder that it has 5ppm/yr. instead of 8ppm/yr., and so easily beating the 3458A in this aspect. (Could even be spec'ed much better, anyhow)
The spec. reveals the ACAL function, as the ranges have constant difference in uncertainty, e.g. 1V for all times constantly 1ppm more uncertain than the 10V range.
This 1ppm difference requires, that the 10:1 ratio transfer must have been made with an uncertainty of 0.1ppm of input. From the linearity curve you presented, that fits to the INL of 0.08ppm (of Range) for its A/D.
The 100mV is always critical, due to thermal voltages; 10ppm is about 1µV, only.
Often, you can not null these offsets in real measurement situations, so you have to include them in the spec.
For example, if you measure a Weston standard cell, you can't short the DUT for cancellation, so you will at least measure all thermal voltages at the jacks, internally to the DUT, also on the cables.
In my post about the 34465A/470A, also ACAL instruments, I later found out, how to verify 100mV DC on the '465A vs. the 3458A, more precisely.
The trick was, to null both instruments at zero output of the calibrator, and then applying the 100mV or 1V.. The ratio of both readings gives the gain calibration for the '465A only, and these 10 or 15ppm additional error (due to thermals) is eliminated. After that procedure, the 100mV range (gain) was measured to be uncertain to < 1ppm.
Maybe that's a specific advantage of ACAL instruments, as they also eliminate offsets during the ACAL procedure, so that external thermal offsets really can be nulled.
Stability (24h, 2h, 10min) and noise measurements would also be interesting.
Frank