Legora tests agents in reinforcement learning environments containing data rooms with up to 10,000 files