Also, if the cal is done offline in software, then changing coefficients (i.e. recalibrating using a different set of coefficents) is purely software and doesn't require another sweep. So it would be technically possible to do a full sweep and change cal definitions on the fly. The definitions would match the real cal standards very closely.
If you have a VNA where you can access to the calibration coefficients, like the 8753 and, I think, most of the ones not targeted at hobbyists, you can simply use something like scikit-rf to do the calibration using the calkit S parameters directly (no need of building any model of them) and then write the computed calibration coefficients back to the VNA. I don't know if this is what the METAS VNA Tools can actually do or if it allows only a post processing of raw data acquired from the VNA.
Then another issue is the calkit repeatability, with cheap stuff you never know how it will behave tomorrow
Yes, very true. In this way we can remove the model entirely.
With the cal kit repeatability issue, I think that is where buying high quality open/shorts and characterising them should produce a better result than DIY.
Also, the proposal I mentioned a while back, was to not use an open standard at all when the VNA has a female SMA port.
As long as it is characterised then is there any point in connecting a standard to it, aside from a slight noise reduction? This is one less connection to make when calibrating.
I need an s1p file from a calibrated VNA which has a female SMA port left open. Since we have optimisation software, I will try to create a model for it.