A directed acyclic graph (DAG) models processing stages whose outputs become inputs to later stages, with dependencies flowing one way and no cycle. section 9.1
Data pipeline or training run
Map each processing stage as a node and connect it to the stage that consumes its output. The resulting DAG shows execution order and dependencies without treating the system like a multi-service web application. section 9.1
Use the graph to establish the system boundary before investigating implementation details:
- Classify the system first: a data pipeline or machine learning training run uses a DAG of processing stages, while a web application usually uses a network of distinct programs. section 9.1
- Follow outputs into downstream inputs to expose dependency order and avoid hunting for a multi-service graph where a single-process or pipeline model is more appropriate. section 9.1
















