Life of a Datum: Record, Derived Copies, Deletion Diagram

Record First

New data lands in the system of record before anywhere else
Each fact is stored exactly once and is authoritative
On conflict, the record is correct by definition

Derived Copies

Caches, indexes, views and analytical copies come from the record
They are redundant but make reads fast
Lose one and you rebuild it from the record

The Cost

Every change must propagate to each derived copy
Data pipelines carry the change; consistency becomes work
Read speed is bought with write complexity

End of Life

Erasure must reach the record and every derived copy
Immutable append-only logs make deletion genuinely hard
Data minimization: some data is better never stored