Chaos Machina / CNCS

EXPERIMENTAL TRAINING PLATFORM / SIMULATION / REINFORCEMENT LEARNING

A simulation and reinforcement-learning platform for configuring vectorized environments, supervising training sessions, and inspecting policy behavior. It provides the experimental training and strategy-search layer of the Trinity architecture, separate from orchestration and worker execution.

Engineering

Python / PyTorch, PufferLib, and Gymnasium underpin environment control, actor–critic policies, PPO updates, and model checkpointing. FastAPI services and a real-time interface expose experiment parameters and metrics. Broader population-search and strategy-synthesis features remain experimental.