Urgent.News

What's breaking now, across thousands of outlets.

Tech

dbt tests that actually catch bad data

Most dbt test suites I've inherited check not_null and unique and not much else. Those catch schema problems, not business problems. The tests that have saved us: accepted_values for status columns (catch typo'd statuses) A relationships test on every foreign key A custom test for "row count per day should be within 30% of last week" Freshness checks on source tables The custom tests took an…

We haven't written up this one. Dev.to has the full story — the link below goes straight to it.

Read the original at dev.to →

More in Tech

Why Your Telemetry Code Fails During Live Production Incidents

Most telemetry code gets written when you're calm, happy, and have zero idea how your system is actually going to fail. The result is predictable.

  • Telemetry code often fails to capture internal queue state during production incidents.
  • Effective telemetry requires emitting state vectors for observability systems.
  • Structured metadata logging enables querying patterns based on actual data.

Reranking Models: The Cheapest RAG Quality Upgrade

Last year I took over a support chatbot for a client with about 40,000 help center and internal wiki chunks in a vector database. Retrieval was pure dense vector search.

  • Reranking improves RAG accuracy from 61% to 84% for $40/month
  • Bi-encoders excel at recall, not precision, causing wrong chunk ranking
  • Hybrid retrieval with BM25 and dense vectors handles exact string matches better

More from Tuesday 27 October →