Round-robin tournament

----|--------------------| | Self-play RL for LLMs | Multi-turn competitive/cooperative games | | Strategic reasoning evaluation | Games require planning, bluffing, memory, adaptation | | Theory-of-mind research | Multi-agent hidden-state interactions | | Tournament benchmarking | Standardized environment API across many games | | Reward-model / policy evaluation | Outcome-driven game scoring |

5. Example: Random / Heuristic Agent

import textarena as ta

class SimpleAgent:
    def __call__(self, observation: str) -> str:
        # Replace with parsing + strategy
        return "default_action"

agents = {0: SimpleAgent(), 1: SimpleAgent()}
env = ta.make(env_id="TicTacToe-v0")
env.reset(num_players=2)

done = False
while not done:
    player_id, observation = env.get_observation()
    action = agents[player_id](observation)
    done, step_info = env.step(action)

rewards, game_info = env.close()
print(rewards, game_info)

6. Tournament Loop

import textarena as ta

def play_match(agent_a, agent_b, env_id="TicTacToe-v0"):
    env = ta.make(env_id=env_id)
    agents = {0: agent_a, 1: agent_b}
    env.reset(num_players=2)