Last released Oct 29, 2026
Layered defense framework (verifier, metadata-filtered RAG, negative prompting, alignment) for mitigating LLM hallucinations, with a reproducible ablation harness.