Metrics
Task success measures completion under the released task-specific rule. Grasp/place rates measure operation completion; long-horizon progress and operation disturbance are separate measures. Never replace missing physical evidence with CPU smoke results. See the scoring reference.
The scoring specification defines successful G/P operations, C = G + P, E = C + 1 − s, and the >10 mm disturbance threshold. Task-specific overrides are listed in its ruleset table.