SocialFiVis: A Visual Analytics Sandbox for LLM-Grounded Multi-Agent Simulation in Social Finance

Social Finance (SocialFi) communities function as complex socio-economic systems where digital assets, such as NFTs, are tied to social identity and governance rights. These communities manage a "digital commons" defined by two interdependent pillars: social capital (trust and participation) and financial health (liquidity and asset value). Governing these systems is high-stakes; interventions like changing royalty fees or access rules are often irreversible and can trigger systemic collapse. Current tools for community management are largely reactive and lack the ability to simulate how specific governance rules influence the non-linear, emergent behaviors of diverse community members.

To address this, the authors developed SocialFiVis, a visual analytics sandbox grounded in the Institutional Analysis and Development (IAD) framework. The system operates through a three-part architecture:

1. Quantification Module: It uses a soft-penalty geometric blend model to measure social capital and financial health, providing a baseline for the community's state.
2. Two-Phase Simulation Engine: This engine populates a virtual environment with heterogeneous agents. Each agent is assigned an LLM-derived persona based on empirical data and follows a Perception-Reasoning-Action (PRA) runtime to simulate realistic decision-making under specific governance rules.
3. Visual Analytics Dashboard: A hierarchical interface allows operators to inject counterfactual policies (e.g., changing a token-gating rule) and observe the cascading effects from the macro-level (community metrics) down to the micro-level (individual agent reasoning).

The system enables "what-if" analysis, allowing operators to backtest strategies without risking real-world assets. Case studies demonstrate that SocialFiVis effectively reveals how governance shocks affect different cohorts differently. For example, the system can visualize the structural decoupling of social capital from financial health, helping operators identify why certain members remain resilient during market downturns while others abandon the community. By tracing macro-level outcomes back to individual behavioral rationales, the tool provides an interpretable bridge between policy changes and community response.