feat: implement version 2 AI Red-Teaming framework with updated reports, server logic, and environment configuration

generate_master_dashboard.py

@@ -80,21 +80,18 @@ def parse_v1_markdown():
 
 
 
-def get_domain1_stats(evaluated: list) -> dict:
+def get_domain1_stats(v2_evaluated: list, v1_evaluated: list) -> dict:
     """Extract stats from Domain 1 (AI Red-Teaming) evaluated results, combining V1 + V2."""
-    if not evaluated:
-        return {"available": False}
-
-    v2_total = len(evaluated)
-    v2_succeeded = sum(1 for r in evaluated if r.get("attack_success"))
-    v2_defended = sum(1 for r in evaluated if not r.get("attack_success") and r.get("evaluation") != "SKIPPED")
-    v2_skipped = sum(1 for r in evaluated if r.get("evaluation") == "SKIPPED")
     
-    # V1 Hardcoded stats (46 tests) from reports/redteam_report.md
-    v1_total = 46
-    v1_succeeded = 1
-    v1_defended = 45
-    v1_skipped = 0
+    v2_total = len(v2_evaluated)
+    v2_succeeded = sum(1 for r in v2_evaluated if r.get("attack_success"))
+    v2_defended = sum(1 for r in v2_evaluated if r.get("evaluation") in ["DEFENDED", "LIKELY_DEFENDED"])
+    v2_skipped = sum(1 for r in v2_evaluated if r.get("evaluation") == "SKIPPED")
+    
+    v1_total = len(v1_evaluated)
+    v1_succeeded = sum(1 for r in v1_evaluated if r.get("attack_success"))
+    v1_defended = sum(1 for r in v1_evaluated if "DEFENDED" in r.get("evaluation", ""))
+    v1_skipped = sum(1 for r in v1_evaluated if r.get("evaluation") == "SKIPPED")
     
     total = v1_total + v2_total
     succeeded = v1_succeeded + v2_succeeded
@@ -104,9 +101,10 @@ def get_domain1_stats(evaluated: list) -> dict:
     asr = round(succeeded / max(total, 1) * 100, 1)
 
     return {
-        "available": True,
+        "available": True if total > 0 else False,
         "total": total,
-        "v2_results": evaluated,
+        "v2_results": v2_evaluated,
+        "results": v1_evaluated,
         "succeeded": succeeded,
         "defended": defended,
         "skipped": skipped,
@@ -149,9 +147,10 @@ def generate_dashboard():
 
     # Load all data
     d1_evaluated = load_json(DOMAIN_1_EVALUATED) or []
+    v1_tests = parse_v1_markdown()
     d2_results = load_json(DOMAIN_2_RESULTS) or []
 
-    d1 = get_domain1_stats(d1_evaluated)
+    d1 = get_domain1_stats(d1_evaluated, v1_tests)
     d2 = get_domain2_stats(d2_results)
 
     # Calculate totals
@@ -482,11 +481,6 @@ def generate_dashboard():
         # Let's clean them up to be safe (remove 'available' which is fine, but just standard json.dumps)
         
         # Ensure we don't have Path objects inside by just taking what we loaded
-        # Inject the V1 parsed tests so the HTML dashboard can render the 46 tests!
-        v1_tests = parse_v1_markdown()
-        if "results" not in d1:
-            d1["results"] = []
-        d1["results"].extend(v1_tests)
         
         inject_data = {
             "d1": d1,