Stories break flow and conversation behaves incorrect

are you using any tests? Setting up end-to-end evaluation would be a good start Evaluating Models