标签：hallucination

Three measured experiments on RAG hallucination: quote-checking, TypeSafe's Jev, and IBM's STAIR. 850+ graded questions, raw responses included.

Pre-registered benchmark: can a 2B local model (Gemma 4 E2B) answer web questions without making things up when a decision model (TypeSafe Jev) makes every call? SearXNG for search, MemPalace for verbatim memory, seven arms including open local judges. Spec and thresholds fixed before any run.

Grounded information extraction that cannot hallucinate: code proposes spans, Jev decides. An open-source alternative to LangExtract, with a bilingual benchmark and paper.