Building Chronicle in the open

Record-and-replay for agent decision graphs: reproduce a prod failure as a committed test, replay your fix with no live LLM calls. If a post here saves you a debugging session, a star tells us to keep writing.