Yes you are absolutely correct. As long as the kit is characterised then it does not need to be perfect.
Characterisation is as simple as sweeping the standard on a calibrated instrument.
There is however a caveat!
Due to the limitations in older gear (i.e. the stuff we all have) the characterisation data for those instruments is based on a simple model we all have seen (C0, C1, delay impedance etc). This model will not fit poorly manufactured standards very well over a wide bandwidth. So errors are introduced fitting the actual measurements to the model.
It is possible to work around this in two ways:
Do the correction calcs on a PC instead of on the instrument - i.e. if you give me any set of parts, along with the S1P or S2P data from a calibrated sweep, then I can use that to make calibrated measurements. I can also use it to 'calibrate' my own parts - with a loss of accuracy.
Use the instrument built in calcs as normal, but limit the bandwidth of the model.
So if the parts are not good enough for a 0 to 6GHz model, then we could instead create two models, one for 0- 3GHz and one for 3 to 6GHz. Or even finer to reduce the error introduced by model fitting.
Edit: Also working from first principles is possible using an RF modelling tool. I haven't tried this and don't have the software to do it, I believe Dr Kirkby has done work in this area as part of his cal kit development.