data: add dataset preview

README.md

@@ -14,6 +14,11 @@ language:
   - en
 size_categories:
   - n<1K
+configs:
+  - config_name: default
+    data_files:
+      - split: test
+        path: preview/cases.jsonl
 ---
 
 # IndicBankBench — A Benchmark for Evaluating the Safety and Reliability of Language Models in Indian Retail Banking
@@ -30,6 +35,9 @@ Each case is a JSON file at `case_bank/<domain>/<tool>/<axis>.NNN.json`. Key fie
 `axis`, `domain`, `tool`, `target_behavior`, `setup` (persona + `login_context` +
 `prior_messages`), `tools_exposed`, `user_turns`, `mock`, `gold`, and `grading`.
 
+The Dataset Viewer reads the generated `preview/cases.jsonl`; the harness uses the canonical JSON
+case files.
+
 ## Running with the harness
 
 Clone the [IndicBankBench](https://github.com/npci/IndicBankBench) harness, then download the