The claim
A green workflow tells us that code ran, dependencies resolved, and outputs were written. It does not tell us that the output represents the same business reality as the systems around it.
The recurring pattern
The difficult incidents often begin with a harmless difference: a source closes its business day later, a hierarchy changes before a dependent system receives it, or an adjustment is represented in one domain but not another. Every individual pipeline remains valid.
A better response
Operational health and decision health need separate controls. The first asks whether the machinery ran. The second asks whether the evidence agrees strongly enough to support the intended action.
The question to carry forward
For every important data product, what could be wrong even when all scheduled jobs finish successfully?