Abstract for: OpenCLD: A Scalable Python Library for System Dynamics Modelling

System Dynamics modeling remains anchored to proprietary, monolithic tools that conflict with open-source, reproducible, and programmatic workflows common in the Python ecosystem. Licensing costs restrict access in education and research, while GUI-first binary formats are incompatible with version control, containerized deployment, and machine-learning pipelines, creating integration, reproducibility, and collaboration gaps that no existing framework fully addresses. OpenCLD is introduced as a lightweight, open-source Python library built on object-oriented principles, where each SD primitive — Stock, Flow, Auxiliary, Parameter, Table — maps directly to a Python class with a single, well-defined responsibility. The library enforces dimensional consistency via Pint, performs automated structural analysis for feedback loop detection, and exports results as pandas DataFrames, ensuring seamless compatibility with the scientific Python ecosystem. Three canonical SD models (Lotka-Volterra, SIR, Bass Diffusion) validate the library's core pillars of transparency, dimensional consistency, and extensibility. Three applied hybrid SD–ML case studies demonstrate end-to-end integration: a flood control system coupling SD with LSTM forecasting, a supply chain surrogate achieving R²=0.9927 across 640,000 simulations, and a microgrid PPO agent reducing energy costs by up to 45.6%. OpenCLD establishes a "Model-as-Code" paradigm that positions System Dynamics as a first-class citizen within the scientific Python ecosystem. Key limitations include reliance on synthetic validation data, absence of stiff solvers, and a Python-programming prerequisite that may exclude non-technical users. Future directions include differentiable programming via JAX, cloud-native digital twin deployment, and LLM-based natural language interfaces for model construction. Generative AI for proofreading