Where a result lives after the job finishes
What reads from it
Page 2 of 4
GET /outputs on the application, with a dispatch token, is the ordinary way. It answers with what is
there for the tenancy the token names, and it is the same call whether the store behind it is a local
directory on a laptop or an S3-compatible bucket in a cluster — the scheme is resolved by the platform
rather than by the caller.
The uploads side is symmetrical: POST /uploads puts an input where the application will look for it,
under the same spine, and an operation then names it. So the round trip for a period is upload the
extract, submit the operation, poll the job, read the output — four calls, one identity, one path.