Skip to content

Topic

Data pipelines without silent errors

The worst data bug is the one that doesn't crash. The load “completes”, the report opens, and the number is wrong — a schema that changed in silence, a join that inflated the total, a record dropped by an INNER JOIN, money typed as a FLOAT. This is where I collect what actually breaks trust in a production pipeline and how to guard against it: data contracts, validation that fails loud instead of silent, quarantine before the report, and real reproducibility. It's the engineering side of “can we trust this number?” — written by someone who has fixed it in production, in finance and international operations.

Does this sound like your problem?

A 30-minute call, no commitment. I'll tell you where the risk is and what to fix first.