Update dataset card

README.md

@@ -15,8 +15,6 @@ size_categories:
 
 The frozen **evaluation split** of VBVR-Pro. 100 parameterized reasoning tasks, 5 held-out instances each, in both a video (I2V) and an interleaved-image setting.
 
-Part of the **VBVR (Very Big Video Reasoning Suite)** project: <https://video-reason.com>.
-
 Companion resources: [VBVR-Pro-SFT-Video](https://huggingface.co/datasets/Video-Reason/VBVR-Pro-SFT-Video) · [VBVR-Pro-SFT-Image](https://huggingface.co/datasets/Video-Reason/VBVR-Pro-SFT-Image) · [VBVR-Pro-RL](https://huggingface.co/datasets/Video-Reason/VBVR-Pro-RL)
 
 ## At a glance
@@ -26,7 +24,7 @@ Companion resources: [VBVR-Pro-SFT-Video](https://huggingface.co/datasets/Video-
 | Tasks | **100** (`In-Domain_50` 50 + `Out-of-Domain_50` 50) |
 | Instances per task | **5** |
 | Total instances | **500** per setting (1,000 across both) |
-| Settings | Video (I2V) and Image (interleaved generation) |
+| Settings | Video and Image |
 | Resolution | **1024 × 1024** |
 | Video | 16 fps, MPEG-4 Part 2, as generated (no re-encode) |
 | Size | 83.9 MB (video) + 9.8 MB (image) |
@@ -52,7 +50,7 @@ VBVR-Pro-Bench-Video/
 │   │   │   ├── prompt.txt             # instruction
 │   │   │   ├── ground_truth.mp4       # reference video     (16 fps)
 │   │   │   └── final_frame.png        # last frame of the reference video
-│   │   ├── 00001/ … 00004/            # same five files
+│   │   ├── 00001/ … 00004/            # same four files
 │   │   └── …
 │   └── …                              # 50 tasks
 └── Out-of-Domain_50/                  # 50 tasks, same shape