Zhiqiangzhou

This release contains 5000 accepted scripted simulation demonstrations. All 1000 accepted episodes per task. Canonical indices follow the declared source-part order, then numeric original accepted episode directories. Episode JSON keeps original paths; task manifests map canonical names to the original manifest rows. Original run/env contracts are in sourceruns/. The additional/ directory preserves 400 further accepted episodes from the capacity collections. The balanced policy-training core remains 5000 episodes; these additional records do not enter that training budget. The robot uses recorded Gaussian appearance; table/wall RGB comes from captured photographs, while task props and…