Here i have some results. The diagrams show daily averages since June, 24. As one scanner round takes about 40 seconds, each average contains about 2270 * 2 = 4540 measurements of each voltage.
The setup includes three experimental voltage references: Two with LTFLUs, one with 2x JFET. The best agreement is between the two LTFLUs, the diagram is +/- 0.1 ppm now! Biggest day to day step observed was 0.06 ppm. The drift of 17.2 nV/day between the LTFLUs continues as expected from an exponential relaxation fit i showed some time ago in the LTFLU thread. It predicts 18.1 nV/day.
The 2x JFET JVR i made in February still isn't tired of drifting. 120 nV/day amounts to 4.4 ppm/year. Noise appears to be low, though.
Regards, Dieter
Edit: 120 nV/day * 365 days/a gives 44 uV/a, i.e. 4.4 ppm of 10 V. Not as bad.