Anticipating the social impacts of a policy prior to implementation requires institutional consistency: multi-agent reasoning must remain aligned with the rules specified in policy documents. However, existing LLM-based multi-agent simulations struggle to maintain such fine-grained institutional rules when policy documents are provided only as prompt context. In this paper, we propose TIGRIS, a Theory-guided framework for Institutionally Grounded policy Impact Simulation. TIGRIS (1) constructs a structured policy knowledge graph from natural-language policy documents, using the Institutional Analysis and Development framework to define institutional positions and Institutional Grammar 2.0 to encode behavioral rules; and (2) instantiates actor nodes as persona agents while anchoring them to a Theory of Change framework to map the causal pathways of policy impacts. Evaluated across four AI talent development policies and three LLMs, TIGRIS achieves the lowest macro-average MAPE, indicating that theory-guided institutional grounding supports domain-aligned reasoning for policy impact analysis.

As personality can be useful in many cases, such as better understanding people’s underlying contexts or providing personalized services, research has long focused on modeling personality from data. However, the development of personality detection models faces challenges due to the inherent latent and relative characteristics of personality, as well as the lack of annotated datasets. To address these challenges, our research focuses on methods that effectively exploit the inherent knowledge of Large Language Models (LLMs). We propose a novel approach that compares contrasting perspectives to better capture the relative nature of personality traits. In this paper, we introduce PADO (Personality-induced multi-Agent framework for Detecting OCEAN of the Big Five personality traits), the first LLM-based multi-agent personality detection framework. PADO employs personality-induced agents to analyze text from multiple perspectives, followed by a comparative judgment process to determine personality trait levels. Our experiments with various LLM models, from GPT-4o to LLaMA3-8B, demonstrate PADO’s effectiveness and generalizability, especially with smaller parameter models. This approach offers a more nuanced, context-aware method for personality detection, potentially improving personalized services and insights into digital behavior. We will release our codes.