Add dataset card

README.md

@@ -1,31 +1,61 @@
----
-dataset_info:
-  features:
-  - name: text
-    dtype: large_string
-  - name: label
-    dtype: int64
-  - name: source
-    dtype: large_string
-  splits:
-  - name: train
-    num_bytes: 14055586
-    num_examples: 25648
-  - name: validation
-    num_bytes: 1744701
-    num_examples: 3298
-  - name: test
-    num_bytes: 1715329
-    num_examples: 3211
-  download_size: 2283925
-  dataset_size: 17515616
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
-  - split: validation
-    path: data/validation-*
-  - split: test
-    path: data/test-*
----
+---
+license: apache-2.0
+task_categories:
+- text-classification
+language:
+- en
+tags:
+- prompt-injection
+- jailbreak
+- ai-security
+- llm-security
+size_categories:
+- 10K<n<100K
+---
+
+# prompt-injection-binary
+
+Binary detection: **1 = prompt-injection / jailbreak attempt, 0 = benign.**
+
+Part of the Neuralchemy prompt-injection dataset family — same corpus, one label axis each.
+Each row is a prompt; `label` is its label; `source` carries provenance.
+
+## Splits
+
+| split | rows |
+|---|---|
+| train | 25648 |
+| validation | 3298 |
+| test | 3211 |
+
+## Class distribution
+
+| `label` | train | validation | test |
+|---|---|---|---|
+| `1` | 19181 | 2467 | 2402 |
+| `0` | 6467 | 831 | 809 |
+
+## How the labels were produced
+
+Labels were generated by a **14B open-weight model** (qwen2.5:14b class) prompted to classify
+each text, then cleaned (below). They are **model-generated annotations, not human gold
+labels** — good for training and benchmarking, but treat them as such; an LLM labeler is not
+perfectly consistent.
+
+## Cleaning (what makes this different from a raw dump)
+
+- **De-contradicted:** any text the labeler assigned *different* labels to in different places
+  was removed, so no label is ambiguous.
+- **De-duplicated:** exact-duplicate texts collapsed to one row.
+- **Group-aware split:** splits are assigned per unique text, so **no text appears in more than
+  one split** — verified zero train/validation/test overlap, so test metrics are not inflated.
+
+## Related
+
+`neuralchemy/prompt-injection-binary`, `-intent`, `-technique`, `-severity` — same corpus,
+different axis. `neuralchemy/Prompt-injection-dataset` — the original binary corpus.
+
+## Intended use & limitations
+
+Training / evaluating prompt-injection & jailbreak classifiers and guardrails. English only.
+Labels are LLM-generated. Only test systems you are authorized to test.