Sieve Build data pipelines that are idempotent, schema-validated, and fail-as-value, enforcing raw and curated data separation.