What is federated computational governance in a data mesh, and which decisions stay global while others stay with each domain?

## Two words, two ideas - **Federated**: decisions are made by a **federation** — domain data product owners and data platform product owners together — rather than by a central governance office. Domains keep autonomy and decision rights over their own data. - **Computational**: agreed rules are **executed by the platform automatically**. A policy that personal data must be classified, or that products must publish discoverability metadata, is enforced in the platform's templates, pipelines and access layer, not by a review meeting. The 2020 article describes the federation's hard job as keeping an equilibrium between centralisation and decentralisation: which decisions are localised to each domain and which are made globally. Global decisions have one purpose — interoperability and the network effect of composing data products. ## Global versus local | Decision | Global or local | Why | |---|---|---| | How a shared entity (customer, listener, product) is identified | global | products must be correlated across domains | | Addressing and naming conventions | global | consumers reach any product the same way | | Access control model and classification scheme | global | security must be consistent | | Discoverability metadata required of every product | global | the mesh must be searchable | | The domain's own data model and semantics | local | the domain knows it best | | Internal pipeline design and tooling choices within the platform | local | implementation detail of the domain | | Service-level objectives for a product | local, within global reporting rules | the owner commits to what consumers need | ## The worked example from the source Dehghani's example: the **semantics and syntax** of the "podcast audienceship" data model are left to the podcast domain team. But **how to identify a podcast listener** is a global concern: a listener is a member of the population of users, who appear in other domains such as stream plays, and unified identification lets the organisation correlate the same person across domains. ## Contrast with traditional governance Traditional data governance centralises decisions and seeks one **global canonical model** with little support for change. Federated computational governance **embraces change and multiple interpretive contexts**: each domain may model a concept its own way, as long as the global rules — identity, interoperability, security — hold. ## What it looks like in practice 1. A governance group with rotating domain representatives meets to agree global standards. 2. The platform team turns each standard into **defaults and checks**: templates that emit required metadata, classification scans, access policies. 3. Compliance is **measured automatically** across products, not audited by hand. ## Why interviewers ask it It is the least understood of the four principles. A good answer explains both words, gives concrete **global versus local** examples, and shows that enforcement is **by the platform**.