GRAND ARENA
Deterministic Simulation Replay
My Contributions
I designed and built a deterministic record-and-replay system for debugging, auditing, and reproducing AI-agent execution in Unity. While a session is running, the system captures the simulation state and serializes it into a CSV-based replay file.
The team can load that file in a separate, replay-focused Unity project and reproduce the game session exactly as it occurred. The replay is fully deterministic, with frame-stepping and time-scale controls for detailed investigation.
Variance in the original simulation was valuable, so the goal was not to make every live run identical. Instead, the replay platform preserved each unique run after the fact, giving the team a complete record of what happened and a reliable way to reproduce behavior for analysis and debugging.