Revac-8 wins MindGames Arena social deduction AI track
Revac-8, an AI agent built for the Mafia-style Social Deduction track of the MindGames Arena competition, took first place by combining memory-based player profiling, social-graph analysis, and dynamic tone selection.
Score breakdown
Practitioners building agentic systems for adversarial or collaborative multi-agent environments can draw on Revac-8's architecture — combining persistent memory, relationship-graph reasoning, and adaptive communication — as a blueprint for agents that must operate under deception and incomplete information.
- 01Revac-8 is an AI agent designed for social deduction games like Mafia, where deception and incomplete information are central.
- 02The agent won first place in the Social Deduction track of the MindGames Arena competition.
- 03Revac-8 evolved from a simple two-stage reasoning system into a multi-module architecture.
Mihir Shriniwas Arya, Avinash Anish, and Aditya Ranjan introduce Revac-8, an AI agent built to tackle the unique challenges posed by social deduction games such as Mafia. Unlike deterministic board games where perfect information or exhaustive search can dominate, Mafia requires players to reason under uncertainty, interpret intentionally misleading information, evaluate human-like communication, and make strategic elimination decisions — capabilities that demand inference, memory, and adaptability in the face of deception.
The agent's architecture grew from an initial two-stage reasoning system into a more sophisticated multi-module design.
Revac-8 was developed specifically for the Social Deduction track of the MindGames Arena competition, where it achieved first place. The agent's architecture grew from an initial two-stage reasoning system into a more sophisticated multi-module design. Its core components include memory-based player profiling, social-graph analysis that tracks patterns of accusations and defenses among players, and a dynamic tone selection mechanism for shaping its own communication. The paper's findings highlight that structured memory and adaptive communication — rather than raw reasoning power alone — are the critical ingredients for high performance in high-stakes social environments.
Key facts
- 01Revac-8 is an AI agent designed for social deduction games like Mafia, where deception and incomplete information are central.
- 02The agent won first place in the Social Deduction track of the MindGames Arena competition.
- 03Revac-8 evolved from a simple two-stage reasoning system into a multi-module architecture.
- 04The architecture includes memory-based player profiling to track individual players across the game.
- 05Social-graph analysis is used to model patterns of accusations and defenses between players.
- 06Dynamic tone selection allows the agent to adapt its communication style strategically.
- 07The authors conclude that structured memory and adaptive communication are key to strong performance in deceptive social environments.