I’m working with a group on an RL core with models as tool use, for explainable agentic tasks with actual discovery.