Data groups and schemas
What a data group is, and how its schema defines the fields an extraction produces.
A data group is one unit of extraction within a project's model: a named set of fields the extraction fills in for every document. The API calls them subtopics in a few summaries — same thing.
GET /data-groups lists them for a project; GET /data-groups/{id} reads one.
The schema
GET /data-groups/{id}/schema returns the fields the group extracts, the rows they are arranged in, and the prompt text that guides extraction. PUT /data-groups/{id}/schema replaces all three at once. Every PUT writes a new active prompt version; running sessions keep the version they started with, so a schema change never alters a run in flight.
The request body is validated against the same schema the product UI uses — a field that the UI would reject is rejected here with VALIDATION_ERROR and the offending field named in details.
Where the results land
Each document-and-data-group pair produces one set of data points. They surface first on the validation side (GET /extractions/{id}/validation), where they can be edited, and after promotion on the data-mart side (GET /extractions/{id}/data-mart). Both exports can be narrowed with ?dataGroupIds=.