Block the Pipeline, or Let It Flow?Quality · medium · ~12 minTwo incidents, two weeks apart, and the team has drawn opposite conclusions.Incident A: a currency-mapping bug produced garbage fx_rate values. The check on the table was *non-blocking* — it alerted, kept publishing, and three downstream teams consumed garbage for a day.Incident B: an upstream vendor file arrived 40 minutes late. The volume check on vendor_raw was *blocking* — it halted the whole nightly DAG, and six unrelated dashboards were empty at 9 a.m. for a data delay that fixed itself.Half the team now wants everything blocking. The other half wants nothing blocking. Design the policy that's actually right.Evidence

Config

Logs

Topology

Checkpoints

What's the principle that decides blocking vs non-blocking — the one both halves of the team are missing?

Incident B's real flaw wasn't the blocking flag. What was it?

The staging capability is sitting unused. How does Write-Audit-Publish change this whole debate for the billing path?

Write the check-policy standard you'd propose: how a check's mode gets chosen, defaults by table tier, and what changes for incidents A and B.