Also - Better diagnostics, output model being used - Prompt LLM that true content is being injected in <context> tag