Update README.md

README.md

@@ -2,6 +2,13 @@
 license: other
 license_name: gg800-subset
 license_link: LICENSE
+configs:
+- config_name: default
+  data_files:
+  - split: train
+    path:
+    - train/metadata.csv
+    - train/**/capture.mkv
 size_categories:
 - 1K<n<10K
 task_categories:
@@ -41,11 +48,15 @@ Each clip follows a fixed format:
 
 - resolution: **720p**
 - frame rate: **12 FPS**
-- clip length: **360 frames** (30 seconds)
-- `events.jsonl`: **exactly 360 lines**, one per frame
+- clip length: **384 frames** (32 seconds)
+- `events.jsonl`: **exactly 384 lines**, one per frame
 
 So the alignment invariant is strict:  
-**frame `i` in video <-> line `i` in `events.jsonl`** for `i = 0..359`.
+**frame `i` in video <-> line `i` in `events.jsonl`** for `i = 0..383`.
+
+Practical note: the first ~2 seconds (~24 frames) may include scene/asset loading artifacts in some clips.  
+For model training/inference pipelines, a common practice is to skip these initial frames and use the final **360** frames.  
+Even when you apply this trim, action logs remain perfectly synchronized with frames by index.
 
 ---
 
@@ -54,8 +65,8 @@ So the alignment invariant is strict:
 Each row corresponds to one clip directory with three assets:
 
 - `video` (`.mkv`)
-- `events_srt` (`.srt`)
-- `events_jsonl` (`.jsonl`)
+- `srt_file_name` (`.srt`)
+- `action_file_name` (`.jsonl`)
 
 In `events.jsonl`, each line includes:
 
@@ -88,8 +99,8 @@ ds = load_dataset("lucidml/GG800-Subset", split="train")
 sample = ds[0]
 
 video_path = sample["video"]          # path or HF file handle (depending on loader setup)
-srt_path = sample["events_srt"]
-jsonl_path = sample["events_jsonl"]
+srt_path = sample["srt_file_name"]
+jsonl_path = sample["action_file_name"]
 
 with open(jsonl_path, "r", encoding="utf-8") as f:
     first_event = json.loads(f.readline())