The speedup over the seed of benchmark tasks from held-out packages, each a consumer of the world's packages that the agents never see, measured at readings of the world's state.
A run's readings come from its exported bundle's evaluation.json, which scripts/export/evaluation.py writes from the paper's re-evaluation of the run or from its checkpoint.
Software World
Sign in to check the simulation
We need a human to sign in to prevent automated agents from accessing the simulation data.