It would be interesting to try in practice how well one can calibrate against skew, amplitude, phase, etc. types of differences that exist between the two paths of the signal (talking about the setup used in the cross correlation method shown at minute 35:35 in the video).
Intuition tells me that the most important would be the stability of the parameters. If the differences (between the two paths of the signal) are stable in time, then it might be possible to identify the exact values of those differences (at calibration time, like it is done for example in a VNA calibration) then apply corrections to the sampled signal.