Describe the release as RoboDojo evaluation records

Presentation-only update; links, reader, recorded episodes and original validation receipts are unchanged.

README.md

@@ -11,7 +11,7 @@ tags:
 - pi05
 - gpt-6-astra
 - evaluation
-pretty_name: Astra as an Embodied Policy — RoboDojo Paired Rollouts
+pretty_name: Astra RoboDojo Evaluation Records
 size_categories:
 - n<1K
 configs:
@@ -21,10 +21,13 @@ configs:
     path: episodes.jsonl
 ---
 
-# Astra / RoboDojo paired rollout dataset
+# Astra RoboDojo Evaluation Records
 
-Recorded evaluation data for **GPT 6 Astra as an Embodied Policy**, by Jiayi Su,
-Yixin Zheng, Mi Yan, Li Yi, Zhizheng Zhang, and He Wang.
+Rollout records from the evaluations in **GPT 6 Astra as an Embodied Policy**,
+by Jiayi Su, Yixin Zheng, Mi Yan, Li Yi, Zhizheng Zhang, and He Wang. This archive
+includes action proposals, executed actions, observations, robot states,
+model-provided explanations and reasoning summaries, and metadata for reproducing
+the evaluation settings, together with a reader and documentation.
 
 - [Report](https://anonymous-report-421.github.io/public-website/)
 - [Public controller source](https://github.com/anonymous-report-421/eval-of-gpt-6-astra-as-policy)
@@ -34,7 +37,7 @@ Yixin Zheng, Mi Yan, Li Yi, Zhizheng Zhang, and He Wang.
 The release contains the **100 selected final evaluation records**: ten tasks,
 five fixed cases per task, two methods. `hybrid` is π₀.₅ + GPT 6 Astra;
 `direct` is GPT 6 Astra（direct）, with no π₀.₅ inference. All records use
-`gpt-6-astra` and `xhigh`. This is an evaluation set, **not a training split**.
+`gpt-6-astra` and `xhigh`. These are recorded evaluation runs, **not a curated training split**.
 It does not include discarded retry attempts, earlier pilot runs, RoboLab,
 or a separately executed π₀.₅-only baseline.
 
@@ -76,7 +79,7 @@ root = snapshot_download(
 ```
 
 Keep `local_dir` set: it produces a self-contained directory, while the reader
-intentionally rejects file links that escape its dataset directory.
+intentionally rejects file links that escape its download directory.
 
 Run `python example.py /path/to/download --method hybrid --task fold_clothes`,
 or import the reader (put the downloaded directory on Python's import path):
@@ -147,7 +150,7 @@ candidate-prefix execution, native/adjudicated outcomes, paired seeds/layouts,
 and recalculates the reported success and Score values. The security check rejects
 known credential patterns, private host paths and internal gateway addresses.
 With all raw observation bundles downloaded, add `--full` to verify every raw
-frame hash and its state-array agreement. This reads the entire image dataset
+frame hash and its state-array agreement. This reads all archived observations
 and takes substantially longer. The release receipt states which checks ran.
 
 Token totals are the controller-recorded usage, including reused cached input.
@@ -189,9 +192,9 @@ budgets or chunk lengths. Five cases per task limit statistical precision.
   author = {Jiayi Su and Yixin Zheng and Mi Yan and Li Yi and Zhizheng Zhang and He Wang},
   year = {2026},
   url = {https://anonymous-report-421.github.io/public-website/},
-  note = {RoboDojo paired rollout dataset: https://huggingface.co/datasets/YuMoool/astra-robodojo-rollouts}
+  note = {RoboDojo evaluation records: https://huggingface.co/datasets/YuMoool/astra-robodojo-rollouts}
 }
 ```
 
 Please also cite RoboDojo and π₀.₅/OpenPI when using their benchmark or policy.
-Pin the dataset revision in derived results and report any excluded records.
+Pin the archive revision in derived results and report any excluded records.