Good docs answer: what is this, who owns it, how do I run/fix it?
What to document
- Purpose and consumers
- Source systems and SLAs
- Schema / grain / primary keys
- Schedule, dependencies, retries
- Failure playbook (common errors)
- How to backfill
- Contact / on-call owner
Lightweight tools
README next to DAG, warehouse descriptions (dbt docs), data catalog notes, Confluence/Notion runbook, dashboard "data freshness" note.
Example outline
> For my internship pipeline I kept a one-page runbook: DAG ID, tables written, "if file missing → check S3 path YYYY/MM/DD", and a sample backfill command. New joiners could restart the job without pinging me.
Fresher tip
Document as you build. Future-you is the first user.
Interview tip: Mention you update docs when behavior changes (not only at project end).