The default query result is inline JSON. There is no
Content-Disposition: attachment header. A client can save the body if it wants a file. The same query with format: 'csv' changes only the representation; source selection, cells and missingness rules remain identical.
For an uncached SILO request, /api/v1/data/requests returns 202 and a request ID while acquisition runs; see the request lifecycle. Once it is ready, the values are served by the same direct query. A direct query never starts an acquisition, and nothing converts a query into a paid job: paid requests will require an explicit quote, and the client must enforce its own spending limit.
Streaming and alternate representations will be advertised through capabilities only when implemented. Current pagination uses a stable cursor and an immutable dataset version. Clients should follow nextCursor rather than constructing offsets or assuming that one page is complete.