A Round-Robin Tournament Dataset for Situational Assessment in a Real-Time Strategy Environment

This dataset contains the complete state trajectories and final outcomes of 3,150 adversarial engagements simulated within the microRTS environment, a standard benchmark for Real-Time Strategy (RTS) game AI research. It was generated to facilitate research in areas such as situational assessment, real-time outcome prediction, and multi-agent behavior modeling.The data was produced via a comprehensive round-robin tournament involving 10 distinct AI agents, each with a unique strategic or tactical profile. To ensure a robust and unbiased evaluation, we conducted 90 unique experimental setups. These setups were derived from the 45 possible pairings of the 10 agents (C(10, 2)), with each pairing being played from both sides to account for any potential positional or first-player advantages (i.e., Agent X vs. Agent Y and Agent Y vs. Agent X are treated as distinct experiments).For each of these 90 experimental setups, 35 independent matches were simulated, yielding a total corpus of 90 setups × 35 matches/setup = 3,150 game instances. Each instance in the dataset includes the sequence of game states represented as multi-channel feature tensors and a final label indicating the winning player. This rich dataset provides a powerful resource for training and validating deep learning models designed to understand complex, dynamic, and adversarial systems.