OK, finally, some rules?
Anyhow, one single point does not tell so much, I have to point out again.
See the noise diagram of the 3458A below, how the results may be presented in a systematic and clear manner.
you will get similar graphs for every DMM, I suppose, and also, the relationship to the different DC ranges might also be expressed similarily.
Therefore, several NPLC points should be taken, from 0.01 to 100 ..1000 NPLC, if possible, and in a more intelligent way, maybe.
To gather these data, I think, it's not necessary to blindly log hours of data.. The statistics will converge sufficiently quickly for higher NPLC numbers, I assume.
It will be sufficient, to gather 10.000 samples for NPLC 0.01, down to 100 samples for NPLC 100, to get a reasonably stable statistics, and to reduce the measurement time.
That would be a maximum of 6 measurements per range, less than 20 for all three ranges of interest (10V, 1V, 100mV), and about 20 minutes per range.
Then, all the measurements of the participants must be properly and completely specified, which range, NPLC, etc., so that a proper normalization can be done!
In the case of the 3458A, HP normalized to ppm relative to full scale.
Btw.: As the input is shorted brutally, there will absolutely be no difference between 10 MOhm or Hi-Z mode of the front-end.
The 10MOhm divider will be in turn shorted, as will be its noise.
Also, if the measurement time is kept short, the actual offset voltage will not play a role, neither its drift.
Therefore, on the short NPLC numbers, AZERO may be turned off, to get a faster result.
You won't draw any conclusion either from hours of offset / Null stability monitoring, especially if you only want to measure the input noise .
Frank