Update README.md

README.md

@@ -24,6 +24,15 @@ tags:
 - synthetic-data
 size_categories:
 - 1K<n<10K
+configs:
+- config_name: default
+  data_files:
+    - split: train
+      path: audit_log.jsonl
+- config_name: signed
+  data_files:
+    - split: train
+      path: audit_log_signed_adaptive.jsonl
 ---
 
 # Streaming PHI De-Identification Benchmark
@@ -41,6 +50,7 @@ It supports:
 
 ## Quick Start
 
+
 ```python
 # Install dependencies
 # pip install phi-exposure-guard datasets pandas
@@ -63,14 +73,18 @@ adaptive = [e for e in events if e["policy_run"] == "adaptive"]
 text_events = [e for e in events if e["modality"] == "text"]
 ```
 
-Or load via Hugging Face datasets library:
-
+Or load via the Hugging Face datasets library:
 ```python
 from datasets import load_dataset
 
+# Base audit log (default config)
 ds = load_dataset("vkatg/streaming-phi-deidentification-benchmark")
+
+# Signed adaptive audit trail (includes _signature, _record_hash, _chain_ts)
+ds_signed = load_dataset("vkatg/streaming-phi-deidentification-benchmark", "signed")
 ```
 
+
 ## Motivation
 
 Automatic de-identification of healthcare data is critical for safe secondary use in AI research and clinical analytics. Existing PHI datasets have significant limitations for streaming and multimodal evaluation: