HewlettPackard/sustain-cluster ? reverse-engineered prompt

Build me a runnable benchmark project for sustainable workload scheduling across multiple data centers.

I want an environment where I can simulate AI jobs arriving over time, see how they move between regions, and test an agent that decides whether to run now or defer. It should use realistic inputs like workload traces, electricity prices, carbon intensity, weather, and network transfer costs, so the scheduling choices actually feel grounded in the real world.

Please include a simple way to train and evaluate an RL agent, plus a few ready to run examples or notebooks so I can try it quickly. The app should make it easy to compare different goals like lower energy cost, lower emissions, and better SLA performance, and it should let me inspect the results clearly after a run.

If you need to look up current docs online while wiring things together, go for it.