I would like to actively contribute.
To add to the discussion:
There are two types of metadata - (A) on a specific row (I converted kilograms in this row from lb in the source system).
And (B) metadata on a dataset (e.g., count of persons that were removed due to missing birth year) during source to target (target being OMOP CDM) data transformation.
For (B) type, the fact_relationship "hack/overload" would not work. (or would be "stretched" a bit)