SOPRANO : Standard Operating Procedures for Reproducible Agent-based and Network-Oriented modeling

Ecco i testi pronti per i campi del deposito. Description SOPRANO (Standard Operating Procedures for Reproducible Agent-based and Network-Oriented modeling) is a normative framework defining the minimum methodological, formal and engineering requirements that a complex systems modeling exercise must satisfy in order to be scientifically defensible. Established practice provides strong instruments for describing agent-based models (the ODD protocol and its extensions), for validating simulation models, for analyzing network structure, and, separately, for conducting statistically sound reinforcement learning experiments. What it does not provide is a framework in which learning is treated as a dynamical level of the modeled system, subject to the same evidentiary standards as the state level and the topological level. When agents optimize their own policies during the simulation, a second optimization process runs inside the model, and behaviors produced by trained agents become, until proven otherwise, artifacts of the training procedure. SOPRANO closes this gap. Building on the classical treatment of dynamical networks and agent-based models, it introduces a nine-component formal ontology, a four-bit dynamic activity vector that extends Sayama's taxonomy of dynamics on, of and adaptive networks to systems with learning and demography, and an ODD+D+L documentation protocol that adds a Learning section to ODD+D. Its original contributions are five. First, an operational and falsifiable criterion for emergence, including well-mixed and shuffled-interaction controls. Second, mandatory classification of inheritance mechanisms, in particular the distinction between Darwinian and Lamarckian transmission of learned parameters. Third, control requirements for generative non-deterministic components, specifically language models participating in the simulation loop. Fourth, a protocol for structural erosion and critical transitions that imposes a positive control on any claim of absence of early warning signals. Fifth, compliance expressed as four binary verifiable levels rather than as descriptive guidance. The document specifies approximately 110 numbered requirements, a catalogue of twenty anti-patterns, a completable compliance checklist, and a YAML Model Specification Sheet template. It is intended for researchers and practitioners building simulations of social, organizational, ecological and evolutionary systems, particularly where reinforcement learning agents are involved.