So time to publish some result.
It seems I almost sorted out all my problems with EMI and leads and power sources ... now no matter what I do, I do not have anymore super large drift of the output (almost 2ppm before) the max drift by changing swapping all this conditions is now usually 0.1 or 0.2ppm with the larger one of 0.3 ppm when I turn on a programmable power supply.
I will explain all I did in a further post, I think I can live with it for now.
Anyway here is last chart from yesterday when I applied heater caps ... (RMS noise went down of almost 100nV).
The image is very large ... so it's better you open in another window.
The red line at the bottom is RMS noise. 0nV at base of the chart and 250nV at every division, RMS is calculated on last 100 samples.
The tin green line that is my room temperature.
The green dots are the single readings (100NPLC with autozero sync)
The blue line is the median value of the previous 100 readings.
The orange dashed lines are the temperature of last ACAL and the internal temperature of the 3458A (temperature are chart with -10°K to make the chart more compact)
The last black line is internal reference temperature.
The chart is one 10uV large ... so 1.4 ppm.
PS: first drop in the chart is me changing leads.