Add dataset card

README.md

@@ -1,17 +1,45 @@
 ---
-dataset_info:
-  features:
-  - name: question
-    dtype: string
-  splits:
-  - name: train
-    num_bytes: 377362
-    num_examples: 1600
-  download_size: 221887
-  dataset_size: 377362
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
+license: cc-by-sa-4.0
+task_categories:
+  - text-generation
+language:
+  - en
+tags:
+  - math
+  - calibration
+  - rl
+pretty_name: CONCORD training prompts
+size_categories:
+  - 1K<n<10K
 ---
+
+# CONCORD training prompts
+
+1600 math prompts used to train the CONCORD models from
+
+> *CONCORD: Label-Free Calibration of Verbalized LLM Confidence via Rollout Consistency.*
+> Maximilian Schall, Gerard de Melo, Padhraic Smyth. **COLM 2026.**
+
+Sampled uniformly across difficulty levels from
+[`OpenDataArena/ODA-Math-460k`](https://huggingface.co/datasets/OpenDataArena/ODA-Math-460k)
+with seed 42. Only the prompt column (`question`) is retained — CONCORD is
+label-free, so reference answers are intentionally not included.
+
+- Source: `OpenDataArena/ODA-Math-460k`
+- Subset size: 1600
+- Columns: `question`
+- Sampling: uniform over `difficulty`, seed 42
+
+If you need reference answers, difficulty labels, or the full dataset,
+use the original source directly.
+
+## Citation
+
+```bibtex
+@inproceedings{schall2026concord,
+  title     = {{CONCORD}: Label-Free Calibration of Verbalized {LLM} Confidence via Rollout Consistency},
+  author    = {Schall, Maximilian and de Melo, Gerard and Smyth, Padhraic},
+  booktitle = {Conference on Language Modeling (COLM)},
+  year      = {2026}
+}
+```