Add README with explicit YAML schema to fix DatasetGenerationError (matched_pair_id string vs null)

README.md

@@ -1,3 +1,67 @@
----
-license: cc-by-4.0
----
+---
+dataset_info:
+  features:
+    - name: id
+      dtype: string
+    - name: text
+      dtype: string
+    - name: label
+      dtype: int32
+    - name: domain
+      dtype: string
+    - name: token_count
+      dtype: int32
+    - name: matched_pair_id
+      dtype: string
+    - name: split
+      dtype: string
+  splits:
+    - name: train
+      num_examples: -1
+    - name: test
+      num_examples: -1
+configs:
+  - config_name: default
+    data_files:
+      - split: train
+        path: "corpora/*/train_*.jsonl"
+      - split: test
+        path: "corpora/*/test_*.jsonl"
+---
+
+# PoolBench
+
+**PoolBench** is a benchmark for evaluating 19 hidden-state pooling strategies for activation steering across 17 text concepts and 7 language models.
+
+Each concept corpus contains up to 1,000 passages per class (700 train / 300 test) sourced from real public datasets on HuggingFace. No passages are LLM-generated.
+
+## Dataset structure
+
+Files are organised as `data/corpora/<concept>/{train,test}_{pos,neg}.jsonl`.
+
+Each JSON line has the following fields:
+
+| Field | Type | Description |
+|---|---|---|
+| `id` | string | Unique record identifier |
+| `text` | string | Passage text (300–500 tokens via LLaMA-3.1-8B tokenizer) |
+| `label` | int | 1 = concept present (positive), 0 = concept absent (negative) |
+| `domain` | string | Source domain (e.g. `academic`, `news`, `social`) |
+| `token_count` | int | Token count (LLaMA-3.1-8B tokenizer) |
+| `matched_pair_id` | string | Pair ID for matched-pair concepts; `null` for independently-sampled concepts |
+| `split` | string | `train` or `test` |
+
+## Concepts (17)
+
+`academic_tone`, `bureaucratic`, `causation`, `code_docs`, `conditionality`, `contrast`, `deference`, `depression`, `frustration`, `hedging`, `imdb_sentiment`, `legal_formality`, `narrative`, `negation_density`, `numerical_precision`, `planning`, `toxicity`
+
+## Citation
+
+```bibtex
+@misc{poolbench2026,
+  title={PoolBench: Evaluating Pooling Strategies for Activation Steering Vectors},
+  author={Anonymous},
+  year={2026},
+  note={NeurIPS 2026 Datasets and Benchmarks Track}
+}
+```