Dataset Ownership Boundary

Without a boundary, a consumer reads a producer's table directly. The producer's internal schema has become a public interface nobody agreed to publish, so the producer cannot refactor and the consumer cannot rely on anything.

An ownership boundary makes the published surface explicit and separate from the internals — backed by a [[data-contract]] describing what fields exist and what they mean, what quality guarantees hold, how fresh the data is, how changes are handled, and who owns it.

What the published surface must carry

• Schema, with types and nullability, versioned with compatibility rules.
• Semantics. What each field means — which is where most misunderstanding lives. "Active" is a field; what makes a customer active is the contract.
• Quality guarantees: uniqueness, referential integrity, accepted ranges, and what happens when they are violated.
• Freshness: how current the data is and how that is measured, so consumers can reason about staleness rather than assuming currency.
• Ownership, so there is someone to ask and someone accountable.
• A change policy — notice periods, compatibility rules, deprecation process.

Why it enables the decentralised model

A data platform where the central team owns everyone's pipelines is a service desk whose capacity is the organisation's ceiling. Decentralising ownership to producing teams is the fix, and contracts are what make decentralisation safe — without them, decentralisation is fragmentation, and every team's internal schema becomes everyone else's problem.

Ownership decentralises; standards do not. The platform owns the contract format, the enforcement and the catalogue; domains own their data.

Implementation patterns

• Enforced as a build gate, with schema compatibility checked before a producer can publish. Registration in CI means a breaking change fails the producer's build, at the place the change is made.
• Quality checks running continuously against the contract's guarantees, with alerting on violation rather than discovery by a consumer.
• Consumers registered, so a producer can see who depends on what — which converts deprecation from an unbounded broadcast into a conversation with named teams.
• A view or projection as the published surface, so the producer can refactor underneath.
• Freshness published as a measured SLI, not as an aspiration.

Industry example

The recurring enterprise problem is that several teams write to the same core tables and change schemas independently. Nobody can state what is always true of a table, a NOT NULL addition breaks a batch job at 2 a.m. for a team the changer cannot name, and releases must be sequenced across teams.

Establishing single ownership per dataset and publishing a read contract is what breaks that — the owner can change internals freely as long as the contract holds, and consumers have something stable to depend on.

Failure scenarios

• A contract with no enforcement, which is documentation and drifts from reality.
• Semantics omitted, so the schema matches and the meaning does not.
• No freshness guarantee, so consumers assume currency and build workflows that fail intermittently.
• Consumers reading base tables anyway, because it is easier than the published surface — which means the contract exists and the coupling remains.
• No consumer registry, so the producer still cannot change anything safely.

Trade-offs

Contracts add process to publishing a dataset and constrain producers, who must now maintain compatibility for consumers they may not have wanted. That friction is real, and it is the price of consumers being able to build on the data at all.

For a dataset with one consumer inside the same team, a contract is overhead. It becomes valuable exactly when the consumer is someone the producer does not talk to daily — which is the situation a data platform exists to create.

Interview question

"A team changes a column type in their operational database and three downstream pipelines break overnight. What should have existed, and where should it have been enforced?"