Search the archive, browse topics, and jump into the newest pieces below.
Matching posts
6 matching posts
Wrong RAG answers: diagnose retrieval, context, and generation
Trace a wrong RAG answer through source coverage, retrieved candidates, final context, and generation to choose the right corrective experiment.
Oct 21, 2026 • Ranker.dev
Read more6 RAG evaluation mistakes: retrieval, answers, and refusal
Build a layered RAG scorecard for evidence recall, context quality, groundedness, correctness, unanswerable questions, and over-refusal.
Oct 21, 2026 • Ranker.dev
Read moreRetrieval benchmarks: what BEIR and MTEB tell you
Use retrieval benchmarks to shortlist models, then check task, judgments, domain, contamination, and transfer with a local evaluation.
Oct 21, 2026 • Ranker.dev
Read moreHybrid search explained: lexical, vector, and rank fusion
Learn when lexical and vector search complement each other, how reciprocal rank fusion works, and which costs to measure before combining them.
Oct 21, 2026 • Ranker.dev
Read moreDo you need a reranker? Quality, latency, and cost tradeoffs
Decide whether reranking can fix your search failure, with candidate-coverage checks, batch-latency arithmetic, and a cost decision table.
Oct 21, 2026 • Ranker.dev
Read moreHow to choose an embedding model: evaluate your own corpus
Compare embedding models on held-out local queries, relevance judgments, latency, index size, and migration cost rather than leaderboard rank.
Oct 21, 2026 • Ranker.dev
Read more