We added safety validators around the RAG path. These checks are meant to surface invalid inputs, unsafe states, and policy-sensitive paths before they become answer quality problems.
The implementation remains conservative: validation should block or warn with explicit reasons. Silent fallback is not acceptable for this layer.
A public-safe engineering status note for Cortagent Agentic RAG: routing, retrieval, evidence selection, memory, caching, safety, and evaluation remain the active surface.
We shaped Agentic RAG evaluation cases around follow-ups, weak evidence, source targeting, and complex multi-part questions.
We tightened the Agentic RAG path around avoidable work: repeated decomposition, repeated embeddings, and unnecessary retrieval fan-out.