-
10 Nov 2026 · RAG evaluation
A small local judge passed its pilot. Fresh data killed it.
A small local model judged whether RAG answers stick to their source. It ranked six generators correctly in a pilot (Kendall tau-b 0.733) and failed on fresh data (0.048). The cascade idea survived on one dataset.
Read the article → -
24 Oct 2026 · RAG evaluation
I said GraphRAG beats vector RAG. Then I measured it.
A small pre-registered experiment: when a graph retriever and a tuned hybrid retriever get the same context budget, how much of the graph advantage is left? +19.0 F1 points on 2WikiMultiHopQA, +3.5 on MuSiQue, with the limits stated.
Read the article →