We use PicoVNA 6 GHz, standard calibration with embedded kit, for same tasks

You need to understand your measurement system. Simply placing the sample between two horns is not easily going to give you useful information about the sample. Sample size will be a major issue as diffraction around the sample edges may dominate your measurements, especially for small samples or samples with low transmission. Suggest you look at 10.1109/TEMC.2009.2015147 .

For metasurface , you need to think about GRL or TRL calibration method. SOLT is normally at the end of the cable but in your case, you should calibrate upto the surface plane of your metasurface.

I think, you can calibrate transmission coefficient with no things in place of metasurface, and reflection coefficient by a PEC plane with same dimensions as the metasurface dimensions.

Using calibration artifacts is fine but they must be the same size and shape as the test sample to get equivalent edge leakage effects. If you can make a sample with known transmission coefficient similar to your expected results that would be better than no sample. You can use a metal plate with multiple small holes. Model it in CST.