Add dataset card for uncapped dev dataset

README.md

@@ -1,24 +1,21 @@
----
-dataset_info:
-  features:
-  - name: text
-    dtype: string
-  - name: employee_id
-    dtype: string
-  splits:
-  - name: train
-    num_bytes: 5321509
-    num_examples: 62854
-  - name: validation
-    num_bytes: 5276596
-    num_examples: 62875
-  download_size: 5037746
-  dataset_size: 10598105
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
-  - split: validation
-    path: data/validation-*
----
+# Enron Personalization — Dev (Uncapped)
+
+**Description**  
+Per-user Enron email text with original case & punctuation, **no word cap**.  
+Each user’s data is split 50/50 by sentence order into **train** and **validation**.
+
+- Users: 45
+- Split policy: first half of a user's sentences → `train`, second half → `validation`
+- Casing/punctuation: preserved
+- Build time: 2025-10-01 23:33 UTC
+
+**Columns**
+- `text`: string (one sentence per row)
+- `employee_id`: string (derived from filename)
+
+**Notes**
+- This dataset replaces the earlier dev set that had a 10k-word cap per user.
+- Splits are near-even by sentence count (difference ≤ 1 when the user has an odd number of sentences).
+
+**Citation**  
+Please cite the Enron Email Dataset and this repository.