I would using the AD5791 directly with a 7V reference.
Unfortunately the LTC2400 can not used with a 7V reference. I would use a capacitive divider to get 3.5V. It has 4ppm typ. and 15ppm max. INL. With the same 10uV calibrator resolution it's an error of 1.4 typ. and 5.25 max. "counts".
The LTC2400 has also the advantage of a much finer resolution.
Please correct me if I'm wrong or wrote complete nonsense.
I would use a capacitive voltage multiplier * 1.5 having a 10.5V maximum range.
For the LTC2400 I would use a voltage divider * 0.666 having a 4.6V reference and up to 5.1V measurement range.
See:
ppmgeekThe LTC2400 linearity is given as "best straight fit" so if you do a "end point calibration" of the range the actual value is practically doubled against the datasheet value. So 4 ppm are a 40uV missing to the 2.5 V mid range value.
(You can find this in carefully analyzing the assembler code of AN86).
The LTC2400 has a noise voltage of 10uVpp (= 0.3ppm, eff in best case). So the settling time to the final value below 10uV will need averaging many measurement values. So the much finer resolution is only valuable when you can use a integration time of 1 minute or longer.
So the settling time of the AD5791 is much better. And if the switching noise between neigboured output values is low enough one could have the idea to do some PWM with 10uV neighboured values.
With best regards
Andreas