Quality and change safety

Tests that gate promotion, and a preview of what a plan would change before it ships.

#Quality

An agent profiles each dataset and drafts a test suite (null, uniqueness, range, schema and freshness checks, with a severity). You adopt suites through the Inbox. When a suite runs, a blocking failure halts promotion of that connection; a non-blocking failure is reported but does not stop it.

Two named metrics are used everywhere: Checks passing and Datasets passing, so the same word never means two things.

#Change safety

Models are built with SQLMesh. Before a change reaches production, Change categorization shows a dry run of the plan: which models would be added, removed or modified, and whether each change is breaking, non-breaking, forward-only or metadata-only. The page also lists the environments the engine has recorded and the impact of changing a model: what it reads and everything downstream.