Anatomy of a pipeline

A pipeline is composed of four core elements:

• Definition — the DAG itself. Authored in Python via the Vantage SDK and uploaded as a versioned artifact.
• Versions — every upload of a definition file becomes a new pipeline version. Old versions stay runnable.
• Runs — concrete invocations, each with parameters, a status, and per-step logs.
• Schedules — cron-style triggers that kick off a run on a recurring basis.

The list view

Each pipeline row shows the definition name, the latest version, when the last run finished, last run status, and the next scheduled run (if any).

Click any pipeline to see all its versions, all runs, and the schedule list.

Run detail

A run page shows the DAG visually — circles per step, lines per dependency, color-coded by status. Click any step to see its inputs, outputs, container image, and logs.

Common patterns

Pattern	What it looks like
Train + evaluate + register	Train job → eval task → conditional registration into the model catalog if accuracy passes a threshold.
Nightly retrain	Cron schedule → ingest fresh data → train job → diff against last week's model → notify on regression.
Data pipeline	Pull from sources → transform → write to feature store. No training step.