Why it matters
Recovery must be safe.
A pipeline can fail after it writes some data. A safe rerun does not add the same records again.
Pipeline reliability
One valid result after every rerun.
A repeated pipeline run produces the same valid result.
Visual model
Two runs. One result.
A failed job can require a rerun. The rerun must not create duplicate data or change valid totals.
First execution
Run 1
→
Controlled output
Valid result
←
Repeated execution
Run 2
Implementation
Use stable keys and upserts.
The load matches an incoming record to a stable business key. It inserts or updates the correct record.
Evidence
Compare both executions.
The second load produces the same row counts and financial totals.
Used in
Enterprise Spend
The spend pipeline uses this control for repeatable loads.
View project evidence →Control statement
A rerun must not change a valid result.