Drop a file.
Get a job ID.
The master accepts the source over CLI, SDK, or NLE plugin. Every job lands in a single queue with a single ID — track it from any machine on the fleet.
No waiting room. No silent fail. The master answers immediately or refuses with a reason.