Use 10k-row preview parquet as default config for HF viewer

README.md

@@ -25,8 +25,12 @@ pretty_name: UncGPT — Persona Pool
 size_categories:
 - 1M<n<10M
 configs:
-- config_name: personas
+- config_name: preview
   default: true
+  data_files:
+  - split: train
+    path: data/v3/personas_preview_10k.parquet
+- config_name: personas_full
   data_files:
   - split: train
     path: data/v3/personas.parquet
@@ -46,9 +50,12 @@ Part of the [UncGPT NeurIPS 2026 Competition collection](https://huggingface.co/
 
 | Config | Rows | What it is |
 |---|---:|---|
-| `personas` (default) | 1,000,069 | full v3 persona pool |
+| `preview` (default) | 10,000 | first 10k personas — for the HF dataset viewer to render quickly |
+| `personas_full` | 1,000,069 | complete v3 persona pool (3.5 GB; too large for the in-browser viewer) |
 | `names` | 15,000 | canonical name registry |
 
+For programmatic use, load `personas_full` directly with `datasets` or `pyarrow`.
+
 ## Persona fields (23 columns)
 
 - identifier + balance fields (gender, language, culture, age band)