Skip to content

The Contract lifecycle

A Data Contract is a governed object, not a config flag. It moves through a single state machine, and the three dimensions (Meaning, Access, State) advance together through it.

draft → review → compile → deploy → active → audit
StageMeaningLegal next stages
draftAuthored, not yet submitted.review
reviewSubmitted for sign-off by the accountable owners.compile · draft (revisions requested)
compileSign-offs collected; the compiler materializes per-engine artifacts.deploy · draft (compile failed → re-author)
deployCompiled artifacts roll out to the engines (Trino, the SQL proxy, the catalog gateway).active · compile (deploy failed → re-compile)
activeThe Contract is live and enforced everywhere.audit · deploy (rollback)
auditPost-active review against audit evidence.active (clean audit) · draft (re-author)

Transitions not listed are illegal by construction — you cannot jump draft → compile and skip review, and you cannot edit a live Contract in place. Progress always moves forward through review; problems move backward to the stage that can fix them.

The terminal enforcement stage is named active. Earlier material may call it enforce; the state machine uses active.

Moving from review to compile requires sign-off from the accountable owners — typically the data owner (Meaning), the access owner (Access), and the governance owner. The review queue records who approved what and when; those approvals become part of the audit evidence the Contract later proves.

The lifecycle exists to serve three jobs, in order:

  1. Define — author the Contract (Meaning, Access, State) once, in draft, and shepherd it through review and compile.
  2. Enforce — once active, the Contract is enforced at every point in the data path (see Enforcement model): write-time at the catalog gateway, read-time at the SQL proxy, pre-write at the Spark transform, and as a post-commit detection backstop.
  3. Prove — produce audit-grade evidence that the Contract was honored: a tamper-evident hash-chain audit log, lineage showing which engines touched which snapshots, and detection records for anything that slipped through.

Define is the authoring loop; Enforce is the runtime; Prove is what you hand an auditor.

Governance that lets you edit a live policy in place has no defensible story for who changed what, when, and with whose approval. By forcing every change through draft → review → compile → deploy, Neksur makes every active Contract the output of a recorded, signed-off process — and makes rollback a first-class transition rather than a panic.