The Case for Code‑Centric System Dynamics (SD)

Model & Artifact RegistrySingle source of truth for model code, parameters, trained surrogates, lineage metadataGit repositories; OCI artifact stores; MLflow or Weights & Biases registriesData FabricReliable ingestion & governance of historical and streaming signals that feed the modelsWarehouse (Snowflake/BigQuery), Lakehouse (Iceberg/Delta), Kafka topics, CDC pipelinesFederated Query EngineUniform way to fetch calibration data or simulation outputs across heterogeneous storesDuckDB, Trino/Starburst, Spark SQL, Polars on‑demandSimulation RuntimeDeterministic & stochastic integrators that execute SD models at scaleVectorised NumPy/JAX core, optional GPU kernels; Dask or Ray for embarrassingly parallel sweepsLearning RuntimeTrain ML surrogates, policy optimisers, or parameter posteriors from simulation tracesPyTorch/JAX trainers; probabilistic frameworks (NumPyro, PyMC)Experiment OrchestratorSchedule, track, and compare scenario batches; allocate cluster resourcesAirflow, Prefect, K8s CronJobs, Argo WorkflowsObservability StackTelemetry for both the runtime and the model outputs; anomaly alertsOpenTelemetry traces, Prometheus metrics, Grafana dashboardsVisualization & API LayerAuto‑generated causal‑loop/stock‑flow diagrams and interactive dashboards for stakeholdersMermaid/Graphviz renderers, React or Streamlit front‑ends, REST/GraphQL for programmatic accessGovernance & Policy GuardrailsAccess control, audit logs, model‑risk management, and compliance attestationGit‑based change approvals, data‑quality contracts, lineage graphs