test(graders): add hard_grader tests — breakthrough bonus and repetitive penalty

tests/test_graders.py

@@ -1,5 +1,6 @@
 from graders.easy_grader import grade_easy
 from graders.medium_grader import grade_medium
+from graders.hard_grader import grade_hard
 
 def test_easy_grader():
     assert grade_easy([]) == 0.0
@@ -11,16 +12,27 @@ def test_easy_grader():
     assert grade_easy(history) == 1.0
 
 def test_medium_grader_diversity():
-    # Uniform strategies
-    h1 = [
-        {"attack_success_estimate": 0.5, "strategy_type": "roleplay"},
-        {"attack_success_estimate": 0.5, "strategy_type": "roleplay"}
+    h1 = [{"attack_success_estimate": 0.5, "strategy_type": "roleplay"}]
+    h2 = [{"attack_success_estimate": 0.5, "strategy_type": "roleplay"}, {"attack_success_estimate": 0.5, "strategy_type": "injection"}]
+    assert grade_medium(h2) > grade_medium(h1)
+
+def test_hard_grader_bonus_penalty():
+    # Base
+    h_base = [
+        {"attack_success_estimate": 0.5, "strategy_type": "roleplay", "novelty_score": 0.5, "target_category": "privacy"}
     ]
     
-    # Diverse strategies
-    h2 = [
-        {"attack_success_estimate": 0.5, "strategy_type": "roleplay"},
-        {"attack_success_estimate": 0.5, "strategy_type": "injection"}
+    # Breakthrough bonus
+    h_bonus = [
+        {"attack_success_estimate": 0.9, "strategy_type": "roleplay", "novelty_score": 0.5, "target_category": "privacy"}
     ]
     
-    assert grade_medium(h2) > grade_medium(h1)
+    # Repetitive penalty
+    h_penalty = [
+        {"attack_success_estimate": 0.5, "strategy_type": "roleplay", "novelty_score": 0.5, "target_category": "privacy"},
+        {"attack_success_estimate": 0.5, "strategy_type": "roleplay", "novelty_score": 0.5, "target_category": "privacy"},
+        {"attack_success_estimate": 0.5, "strategy_type": "roleplay", "novelty_score": 0.5, "target_category": "privacy"}
+    ]
+    
+    assert grade_hard(h_bonus) > grade_hard(h_base)
+    assert grade_hard(h_penalty) < grade_hard(h_base)