>>> OK will add a count of readings to ignore at the beginning and also a count of readings to average.
B...x this dumb forum software lost my post....here's the abbreviated version. DCV only, AZERO on, 1s dwell between changing source and making first reading.
# ignored # averaged nplc cal headroom %
0 1 100 85
1 1 100 85
1 5 100 74
1 10 100 74
1 10 10 89
1 30 10 85
1 100 10 83
1 1 1000 71
(actually I'll post this and add the other results as they arrive)
Headroom of 90% means device is 10x better than spec, 100% means perfect device, 0% means on spec (on average over tests). This is with the averaging being done in software rather than in the device.
So, with this data at least, the benefit is marginal except for 10 samples of 10 nplc (rather than 1 of 100 nplc as in the cal manual). Looking at the detail, the barrier to a much tighter spec seems to be the low voltage offset - sometimes 10uV which throws off the low voltage sourcing. As the cal algorithm is a black box (e.g. it doesn't seem easy for the user to just adjust the low voltage offset) I'm not sure where to go from here.
I guess you could wrap an optimisation algorithm around it and let it run for days but....
My simple test set-up (below) - no low emf stuff. Will release this version as soon as I have tested it. This is suggesting to me TiN that your LTZ mod is having a good effect - I see the LM399 has ~15x noise of LTZ1000 so maybe that's it.
>>> we don't know if there is any firmware-implemented tempco correction
Well it doesn't seem to work too well if there is