Phase B.2 create-new: README card

README.md

@@ -0,0 +1,41 @@
+---
+license: cc-by-4.0
+task_categories:
+- robotics
+tags:
+- lerobot
+- evaluation
+- rollouts
+- imitation-learning
+- ur3e
+- bimanual
+pretty_name: "Eval of diffusion-multitask_12task_mix-multitask on uncap"
+---
+
+# eval-diffusion-multitask_12task_mix-multitask-on-uncap
+
+On-robot rollout evaluation of
+[`WetLabRoboData/diffusion-multitask_12task_mix-multitask`](https://huggingface.co/WetLabRoboData/diffusion-multitask_12task_mix-multitask) on the
+**uncap** task.
+
+| Field              | Value |
+|--------------------|-------|
+| Policy             | [`WetLabRoboData/diffusion-multitask_12task_mix-multitask`](https://huggingface.co/WetLabRoboData/diffusion-multitask_12task_mix-multitask) |
+| Target task        | uncap |
+| Episodes           | 20 |
+| Successes          | 13 |
+| Success rate       | 65.0% |
+| Robot              | UR3e bimanual (3 cameras) |
+
+## Contents
+
+- `grid_episode_###.mp4` (or `grid_episodes_###.mp4`): per-episode rollout
+  recordings laid out as a 3-camera grid.
+- `deployment_info.yaml`: exact deploy-time configuration.
+- `front.mp4` / `left_wrist_top.mp4` / `right_wrist_top.mp4`: aggregated
+  single-camera streams (not per-episode).
+
+## Provenance
+
+Reorganized from `WetLabRoboData/lerobot-data-rama-lbm_pretrain_all_ds5` on 2026-10-04. The original
+eval folder is preserved in the `old/` subfolder of that source repo.