Data Product
A curated, documented, owned dataset treated as a product with consumers, a service level and a lifecycle, rather than as the output of a pipeline.
The word "product" is carrying specific obligations, and stripping them out is how the term became diluted into "a table someone published".
A data product has an owner accountable for it, consumers whose needs shape it, a service level for freshness and quality it is measured against, documentation sufficient to use it without asking anyone, discoverability through a catalogue, and a lifecycle including versioning and a deprecation process with notice.
The properties usually summarised as discoverable, addressable, trustworthy, self-describing, interoperable and secure are a useful checklist precisely because most published datasets fail at least three.
The economic argument for the model is that it moves curation effort to where the domain knowledge is. A central data team building a table from a domain it does not understand will encode misunderstandings and cannot maintain it; the domain team can, and their reluctance is usually about incentives and funding rather than capability.
The failure to guard against is proliferation without governance: every team publishing lightly documented tables produces a larger swamp with better vocabulary. Interoperability standards and a functioning catalogue are what prevent that, and they have to be central even when the products are not.