Clarify exact two-camera trimmed training snapshot

README.md

@@ -40,9 +40,12 @@ fixed; the policy continued to receive language instructions.
 
 The embedded `train_config.json` pins the motion-trimmed training data to
 [`justintiensmith/VLA_Reasoning_Training_Dataset_1200_Trimmed_Start_5_Frame@c1231a8`](https://huggingface.co/datasets/justintiensmith/VLA_Reasoning_Training_Dataset_1200_Trimmed_Start_5_Frame/tree/c1231a8f2b282b3f875fc898dced9ebf5574b903).
-The currently published dataset head is a content-preserving card update beyond
-that training revision. It contains 1,200 episodes, 570,386 frames, and 321
-instruction strings.
+That exact snapshot is a two-camera projection of the repository's five-camera
+trimmed release and is preserved by the `groot-frozen-training-snapshot` tag.
+The tabular trajectory file and middle/wrist videos are byte-identical between
+the two snapshots; episode metadata differs because the unused camera keys were
+removed. Both contain 1,200 episodes, 570,386 frames, and 321 instruction
+strings.
 
 The run records the base-model identifier but not its immutable source
 revision. The current public GR00T base revision at archival review is