You already know: the lifecycle stages and why lineage matters (ML lifecycle); graphs and topological order (D1 Data structures); transactions and idempotent writes (D1 Databases); journals and retries (D9 Advanced agent engineering).
Why this matters for Atlas. A pipeline turns the lifecycle into code a machine runs: ingest, validate, build features, train, evaluate, register, as a graph of steps with explicit dependencies. Done well, it is reproducible (same inputs, same outputs), incremental (only recomputes what changed), validated (bad data stops the line instead of reaching the model), and recoverable (a failed step retries or resumes without redoing everything). This lesson builds those properties from scratch, then maps them to the orchestrators teams actually use.
1. Pipelines are DAGs
The question. Why did running the notebooks out of order produce a broken model without any error?
Start from nothing. Each step reads some artifacts and writes others. Those read/write relationships form a directed acyclic graph (DAG). An orchestrator does three things a person running notebooks can't do reliably: