Design a Data Contract Gate for a Feature Pipeline

A feature pipeline produces a table transaction_features consumed by a fraud model, with columns transaction_id, amount_usd, merchant_category, customer_tenure_days, and is_flagged_previously. Twice in the last quarter, a bug in this pipeline reached production undetected by CI: once when amount_usd started containing negative values after a currency-conversion bug, and once when merchant_category started containing an empty string instead of "unknown" for unmapped merchants, which the downstream encoder silently treated as a new, rare category.

1. Design a data contract for this table that would have caught both past incidents.
2. Explain where in the CI/CD pipeline this contract check should run, and why it needs to run in more than one place.
3. A teammate argues "we already have unit tests on the pipeline code, isn't that enough?" Explain specifically why unit tests alone would not have caught either incident.