How do you design your pipelines to handle schema drift or unexpected null values in incoming clinical data?
