VISA: A Structured Description Protocol for Agent-Based Simulation Models Towards Machine Reproducibility

Agent-based modeling (ABM) is a powerful tool for studying complex systems, but it suffers from a persistent reproducibility crisis. Because models are typically described through prose narratives, platform-specific code, and implicit assumptions, researchers often struggle to reconstruct the same model from published documentation. This ambiguity makes it difficult to verify findings, compare models across platforms, or build upon existing research.

VISA (a structured, symbol-based description protocol) addresses these challenges by replacing narrative descriptions with a standardized, table-based format. The protocol organizes a model into eight interconnected tables across two levels:

• Agent Level: Defines the micro-level structure, including agent types, state variables, sensing mechanisms (information access), and internal behavioral functions.
• Model Level: Defines the macro-level configuration, including external data sources, input/output parameters, the temporal schedule of events, and validation criteria.

By adhering to the principle of 'minimality with completeness,' VISA ensures that every documented element is necessary for reproduction, while the use of standardized symbols allows for machine-parsing.

VISA goes beyond static documentation by providing nineteen executable consistency rules. These rules allow researchers to treat model validity as a checkable property, ensuring that the logic within the tables is internally consistent. Furthermore, the protocol includes three LLM-executable skills—authoring, checking, and code generation—that operationalize the entire reproduction loop. The author demonstrated the protocol's utility by successfully reproducing two cross-language models (NetLogo to Python) and documenting an industrial AnyLogic model, where the protocol helped clearly isolate proprietary dependencies that blocked full reproduction.