datapackages

Datapackages are a simple, portable format for packaging a dataset together with its metadata. They aim to make data easy to share, publish, and reuse by providing a lightweight descriptor that accompanies the data files. A datapackage is described by a single JSON file at the root of the package, commonly named datapackage.json.

The descriptor includes metadata about the dataset, such as title, description, license, authors, and sources. It also defines one or more resources, each describing a data file or collection (for example CSV or JSON). A resource entry specifies a name, path or URL, and format, and can include size and mediatype. If a resource contains structured data, it can provide a schema that lists fields with names, data types, and optional constraints (such as required, unique, or allowed values), as well as optional key constraints like primaryKey or foreignKeys. The description may also include statistics, the profile, and other metadata.

Datapackages are designed to be self-describing and portable. They can reference local or remote data, and resources can be accessed independently of the descriptor. Validation tools can verify that data conforms to the declared schema and that required metadata is present. The specification is language-agnostic, so tools exist in multiple programming environments.

Origins and ecosystem: Datapackages are a product of the Frictionless Data initiative (and related to the Data Package concept) to promote interoperability and reproducible data workflows. They are used in data catalogs, pipelines, and publishing workflows to facilitate data discovery and reuse.