The comparison should begin with a correction. Established simulation software already provides capabilities I had been treating as generally absent:
- Guided explanation and exploration. STELLA can present a model’s structure in stages, relate it to behavior, and publish interactive interfaces. En-ROADS is used with groups to test policy, technology, and economic scenarios in an executable climate model.
- Collaborative model development. AnyLogic supports Git workflows and a multi-part model format intended to reduce conflicts between developers. Simulink provides model-aware comparison, code review, and three-way merge.
- Traceability and testing. Simulink’s Requirements Toolbox links requirements to models, code, and tests, finds gaps, and shows which artifacts are affected by a change. Vensim supports structural analysis, calibration, unit checking, model tests, and behavioral constraints.
- Several modeling paradigms. AnyLogic combines system dynamics, agent-based modeling, and discrete-event simulation in one environment. Simile supports modular submodels. Topos Institute and StockFlow.jl address formal composition rather than merely placing several diagrams in one file.
Two problems still look well supported by the evidence. First, reliably composing models made in different formalisms and tools remains difficult. Second, provenance across an entire study is usually spread among papers, interviews, datasets, spreadsheets, model files, experiment outputs, requirements, decisions, and later explanations. A tool may handle provenance within one part of that workflow without preserving the argument across all of it.
Other supposed gaps are still questions, not findings. I do not yet know how well current tools support review by technical and nontechnical participants together, competing interpretations of the same evidence, or using one maintained object for analysis, operational decisions, and public explanation.
The next useful comparison is one real workflow, not a larger catalog of features. Record what artifact each tool creates at each stage—problem formulation, evidence, executable model, experiments, decision, and explanation—then record what has to be copied or reconstructed between stages. That would answer a narrower question: whether a new thinking environment is needed, or whether the more useful work is connecting and presenting capabilities that already exist.