Add dataset card

README.md

@@ -1,24 +1,42 @@
 ---
-dataset_info:
-  features:
-  - name: text
-    dtype: string
-  - name: label
-    dtype: string
-  splits:
-  - name: train
-    num_bytes: 164294
-    num_examples: 1029
-  - name: test
-    num_bytes: 35007
-    num_examples: 259
-  download_size: 124567
-  dataset_size: 199301
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
-  - split: test
-    path: data/test-*
+language:
+  - pt
+license: apache-2.0
+task_categories:
+  - text-classification
+tags:
+  - mteb
+  - portuguese
+  - brazilian-portuguese
+  - classification
+pretty_name: FactckBrClassification
 ---
+# FactckBrClassification
+
+Classify the veracity of a native Brazilian-Portuguese claim fact-checked by Brazilian agencies (Lupa, Publica, Aos Fatos), into 3 classes: falso, impreciso, verdadeiro. The input is the claim text only (not the fact-check title, which would leak the verdict). News / fact-checking domain; heavily imbalanced toward falso.
+
+Part of **[MTEB(por)](https://huggingface.co/mteb-pt)** — the native Brazilian-Portuguese MTEB sub-benchmark. **Task type:** Classification · **Language:** Brazilian Portuguese (mined from real-world sources) · **Domains:** News, Social, Written.
+
+## Dataset structure
+- **`default`** — columns: `text`, `label`
+
+## Source
+https://github.com/jghm-f/FACTCK.BR
+
+## License
+`apache-2.0`
+
+## Citation
+```bibtex
+@inproceedings{moreno2019factckbr,
+    title     = {{FACTCK.BR}: a new dataset to study fake news},
+    author    = {Moreno, Jo{\~a}o and Bressan, Giovanni},
+    booktitle = {Proceedings of the 25th Brazillian Symposium on Multimedia and the Web (WebMedia)},
+    year      = {2019},
+    doi       = {10.1145/3323503.3361698},
+    url       = {https://doi.org/10.1145/3323503.3361698},
+}
+```
+
+---
+*Repackaged for MTEB(por); pinned by commit SHA in the task for reproducibility.*