Data Mesh Federated Governance
Central agreement on the rules that must be global, with domain autonomy over everything else.
The principle that keeps data mesh from becoming anarchy, and the one most often skipped when organisations adopt the vocabulary without the operating model.
Domain ownership without common standards produces datasets that cannot be joined: different customer identifiers, different date semantics, incompatible formats, inconsistent access models. That is not decentralisation, it is fragmentation, and it is more expensive than the centralised team it replaced.
Federated governance draws the line explicitly. Global: identity and how entities are keyed across domains, interoperability standards, security and access models, privacy classification, the metadata each product must publish, and the definition of shared cross-domain metrics. Local: internal modelling, technology choices, transformation approach, release cadence, and anything that does not cross the boundary.
The mechanism that makes it work at scale is computational governance — the global rules implemented as automated checks in the platform rather than as a review board. A data product that does not publish required metadata or fails a classification check simply cannot be registered.
The honest caveat: this model demands significant platform maturity and genuine domain engineering capability. Organisations that adopt mesh to escape a bottleneck without building either usually end up with the same bottleneck plus a distributed one.