Dealing with imputation of left censored multivariate normal distribution

Thanks @junpenglao for your valuable response!
But I’m a bit confused with:

In your use case, what I would do is treat the censored value as none-censored, and then account for the censoring using pm.Potential.

What does it mean? In my real case, censored values are censored and I know only they lie below the threshold. Also, I know that expectations are zeros but I don’t know the variance-covariance matrix…Could you please clarify a bit more?

Thanks in advance,
Alex