Thanks for the response! I will definitely check out the video.
However making X a random variable and mixing the calibration set with the unknown set it’s problematic.
During the calibration step X can be considered not random, it’s value is know with very very high precision since those concentrations are by design. In the prediction step it’s the opposite, Xpred it’s unknown and pretty much needs to be treated as random variable.
I would need a way to establish very strict priors on the values of X for the calibration data and very uninformative ones for the unknowns.