RiverBase Software Advisory
Vendor CSV contracts that survive a column shuffle
How data engineering teams lock vendor CSV column contracts so ETL pipelines fail safely when suppliers reorder or rename fields.
Supplier files are a frequent source of quiet warehouse corruption. A column inserted in the middle of a CSV, or two columns swapped, can load for nights before an analyst notices prices on the wrong SKUs.
Prefer named headers over positional reads
If a vendor can emit a header row, require it in the contract and validate names before any type cast. Positional parsers should be treated as legacy debt with an explicit retirement date.
Staging before business tables
Land the file as text, check column names and null rates, then promote to typed staging. Promotion — not the SFTP drop — should be the gate that wakes on-call.
When vendors cannot change
Some older suppliers will never add headers. In those cases we fingerprint the expected column count and a sample of headerless “sentinel” values, and we version the parser beside the contract date in the runbook.
RiverBase covers this pattern inside the ETL Pipeline Engagement when vendor feeds are in scope.