I ran an updated script. One fase (tuning) of a measurement cycle measures only current voltages, the last fase measures also Vbe voltages. This way there's less channel switching.
I would have like to use only one trigger model, which I would the steer with my own script. This seems almost impossible. There is a way to sync with the dmm: let it wait for a *TRG command. But that cannot pass info. And there's no way to way to sqeeze in info some other way, except for lan events. These can have 8 different id's, so they could be used for steering.
Did some research and with some more effort lan events (broadcasting a udp packet) could have worked.
However I went the route of swapping between different trigger models.
The funny thing is that without the ratio measurement, I still get the jumps some times.
The solution is to have large delays between measurements. I think I went for 1 s, but that might not even be enough. Will experiment further, because it seems to depend on what is measured and at what range. This makes me also wonder of secondary measurements, in what way is are they different to what my trigger model is doing? Why would it result in either a jump, or no jump, but not much in between. (Except sometimes there seem to be 3 levels)
Fascinating..