From Silent Data Failures to Enforceable CI/CD Gates

Modern data pipelines often fail quietly: a schema drifts, a late CDC event lands out of order, a Spark job produces plausible but wrong output, or a model starts consuming stale features before anyone notices. This talk walks through a practical approach for turning those silent failures into explicit checks that run in CI/CD and production release gates. I’ll cover data contracts, CDC-aware validation, Spark and lakehouse testing patterns, and a few lessons from building reliable data systems where correctness has to be treated like a release requirement rather than a dashboard afterthought. Join us on [slack](https://forgeutah.tech) Schedule: 6:30pm-6:50pm pizza 6:50pm announcements 7:00pm speaker 7:45-8:30pm mingle How to find us: Enter by the East entrance on 400W (across form the Delta Center). If the door is locked ring the bell and you can be let in. Sign in at the security desk. **Park in the Gateway North Garage. Parking validation is available by QR code at the security desk. Don't use the QR code in the garage (garage QR code requires a credit card, QR code at Recursion's security desk does not). Validation is not available for the South Garage**