A Job is a DAG, not a script. Each named Task declares its operation, inputs, options and dependencies. The scheduler can run independent branches concurrently and only releases downstream work when prerequisites succeed.
Think in a DAG
┌─→ thumbnail ─┐
import/source ─┼─→ pdf ───────┼─→ export/archive
└─→ text ──────┘Parallel work
Do not serialize independent outputs. If the same upload needs PDF, thumbnail and text extraction, make those Tasks siblings. ConverterHQ can schedule them in parallel subject to account concurrency.
Task boundaries
Use a Task boundary where you want independent retry, observability, billing attribution or fan-out/fan-in. Avoid creating artificial Tasks for parameters that belong inside one conversion operation.
Use the Job Builder
The interactive Job Builder renders the dependency graph, performs local cycle checks and produces the exact POST /v1/jobs body. Start from a template, then move to the generated JSON when you need full control.