Evaluating RAG systems: building an eval set, recall@k, faithfulness, and the five places a retrieval pipeline quietly breaks.https://what-is-rag.net/https://what-is-rag.net/blogs/the-five-places-a-rag-pipeline-breaks/https://what-is-rag.net/blogs/the-five-places-a-rag-pipeline-breaks/A bad answer implicates one of five stages. A bisection procedure that finds the responsible one in about ten minutes.Sun, 26 Jul 2026 00:00:00 GMTdebuggingevaluationhttps://what-is-rag.net/blogs/measuring-faithfulness-not-just-answer-quality/https://what-is-rag.net/blogs/measuring-faithfulness-not-just-answer-quality/An answer can be good and ungrounded at the same time. How to check every claim against the retrieved context, and where LLM judges fail.Sat, 25 Jul 2026 00:00:00 GMTfaithfulnessevaluationhttps://what-is-rag.net/blogs/recall-at-k-and-what-it-hides/https://what-is-rag.net/blogs/recall-at-k-and-what-it-hides/Recall, precision, MRR and nDCG measure different things badly. What each one is blind to, and which to report.Sat, 25 Jul 2026 00:00:00 GMTmetricsretrievalhttps://what-is-rag.net/blogs/building-a-retrieval-eval-set/https://what-is-rag.net/blogs/building-a-retrieval-eval-set/Golden questions, labelled relevant documents, and the unanswerable cases. How to construct a set that predicts production behaviour.Fri, 24 Jul 2026 00:00:00 GMTevaluationretrievalhttps://what-is-rag.net/blogs/why-your-rag-demo-works-and-production-does-not/https://what-is-rag.net/blogs/why-your-rag-demo-works-and-production-does-not/The demo and production differ in five specific ways, all of which favour the demo. Each one is measurable before you ship.Fri, 24 Jul 2026 00:00:00 GMTevaluationdebugging