MORPHEUS: Skyfall AI's New Persistent RL Benchmark

Skyfall AI has released MORPHEUS, a new persistent enterprise simulation platform designed for continual reinforcement learning. This platform addresses a key challenge: most reinforcement learning benchmarks reset after each episode, but real-world operations never do. MORPHEUS is built on the Big World Hypothesis, which states that the world's complexity is too vast for any single agent to fully represent. This means the environment appears to be constantly changing, even if its underlying rules stay the same. To encourage continual learning, MORPHEUS includes three core features: persistence, non-stationarity, and operational complexity. Persistence means past decisions impact future dynamics. Non-stationarity ensures that any fixed strategy will eventually become less effective. And operational complexity means no single, perfect strategy exists. The platform introduces non-stationarity through a failure injection engine, which inserts disruptions like "missing_data" or "rate_limit" at various rates, from 5% to 30%. An independent configuration shift controller also changes failure presets and demand over time, preventing agents from predicting these shifts. Agents are rewarded based on a composite score combining failure event signals, financial ledger status, and resource throughput, with specific weights assigned to each. The bottom line is that MORPHEUS aims to make reinforcement learning more relevant to complex, ever-changing real-world scenarios.