ChatXmile : An LLM ‐Assisted Closed‐Loop Workflow for the System Dynamics Process

ABSTRACT This paper presents ChatXmile, a browser‐based method and web application (the ChatXmile system) that integrates large language models (LLMs) with deterministic, rule‐based software to streamline system dynamics modeling. In a closed‐loop workflow, an LLM generates or revises an XMILE model, which is then subjected to automated conformance and technical checks (schema and rule checks, units consistency, structural inspection, and executable trial runs). Diagnostics are fed back to guide targeted revisions. Once a model passes these checks, ChatXmile also supports generating structured narratives and HTML/JavaScript simulators to enable sensitivity analysis, extreme‐condition tests, time‐step adequacy checks, and scenario experiments. Overall, the approach establishes an auditable narrative–model–verification loop that supports reconstruction across sessions and improves the reliability and dissemination of system dynamics models.