fix: class-based graders with grade() method per Meta instructions

openenv.yaml

@@ -22,41 +22,41 @@ tasks:
     name: JSON Config Debug
     difficulty: easy
     num_bugs: 2
-    grader: server.graders.grader_api:grade_task1
+    grader: server.graders.grader_api:Task1Grader
     has_grader: true
   - id: task2_yaml
     name: YAML Config Debug
     difficulty: easy
     num_bugs: 2
-    grader: server.graders.grader_api:grade_task2
+    grader: server.graders.grader_api:Task2Grader
     has_grader: true
   - id: task3_dockerfile
     name: Dockerfile Debug
     difficulty: medium
     num_bugs: 3
-    grader: server.graders.grader_api:grade_task3
+    grader: server.graders.grader_api:Task3Grader
     has_grader: true
   - id: task4_compose
     name: Docker Compose Debug
     difficulty: medium
     num_bugs: 4
-    grader: server.graders.grader_api:grade_task4
+    grader: server.graders.grader_api:Task4Grader
     has_grader: true
   - id: task5_k8s
     name: Kubernetes Config Debug
     difficulty: hard
     num_bugs: 5
-    grader: server.graders.grader_api:grade_task5
+    grader: server.graders.grader_api:Task5Grader
     has_grader: true
   - id: task6_github_actions
     name: GitHub Actions Debug
     difficulty: hard
     num_bugs: 5
-    grader: server.graders.grader_api:grade_task6
+    grader: server.graders.grader_api:Task6Grader
     has_grader: true
   - id: task7_nginx
     name: Nginx Config Debug
     difficulty: very_hard
     num_bugs: 6
-    grader: server.graders.grader_api:grade_task7
+    grader: server.graders.grader_api:Task7Grader
     has_grader: true

server/graders/grader_api.py

@@ -1,6 +1,6 @@
 """
-grader_api.py - Bulletproof grader wrappers.
-Returns FLOAT ONLY. Never throws. Always in (0.01, 0.99).
+grader_api.py - Class-based graders for OpenEnv validator.
+Validator expects classes with a grade() method returning float in (0, 1).
 """
 
 from server.graders.json_grader import grade_task1 as _r1
@@ -12,34 +12,73 @@ from server.graders.github_actions_grader import grade_task6 as _r6
 from server.graders.nginx_grader import grade_task7 as _r7
 
 
-def _safe_float(fn):
-    def wrapper(*args, **kwargs):
-        try:
-            if not args and not kwargs:
-                return 0.5
-            submitted = args[0] if args else kwargs.get("submitted_config", "{}")
-            result = fn(submitted)
-            if isinstance(result, (tuple, list)):
-                reward = float(result[0])
-            else:
-                reward = float(result)
-        except Exception:
-            reward = 0.5
+def _safe_score(fn, env, *args, **kwargs):
+    try:
+        config = ""
+        if env is not None and hasattr(env, 'state'):
+            state = env.state
+            if hasattr(state, 'last_action'):
+                config = state.last_action
+            elif hasattr(state, 'current_config'):
+                config = state.current_config
+        if not config and args:
+            config = str(args[0])
+        if not config:
+            config = "{}"
+        result = fn(config)
+        if isinstance(result, (tuple, list)):
+            reward = float(result[0])
+        else:
+            reward = float(result)
         return max(0.01, min(0.99, reward))
-    wrapper.__name__ = fn.__name__
-    wrapper.__qualname__ = fn.__qualname__
-    return wrapper
+    except Exception:
+        return 0.5
 
 
-grade_task1 = _safe_float(_r1)
-grade_task2 = _safe_float(_r2)
-grade_task3 = _safe_float(_r3)
-grade_task4 = _safe_float(_r4)
-grade_task5 = _safe_float(_r5)
-grade_task6 = _safe_float(_r6)
-grade_task7 = _safe_float(_r7)
+class Task1Grader:
+    def grade(self, env=None, *args, **kwargs) -> float:
+        return _safe_score(_r1, env, *args, **kwargs)
+
+
+class Task2Grader:
+    def grade(self, env=None, *args, **kwargs) -> float:
+        return _safe_score(_r2, env, *args, **kwargs)
+
+
+class Task3Grader:
+    def grade(self, env=None, *args, **kwargs) -> float:
+        return _safe_score(_r3, env, *args, **kwargs)
+
+
+class Task4Grader:
+    def grade(self, env=None, *args, **kwargs) -> float:
+        return _safe_score(_r4, env, *args, **kwargs)
+
+
+class Task5Grader:
+    def grade(self, env=None, *args, **kwargs) -> float:
+        return _safe_score(_r5, env, *args, **kwargs)
+
+
+class Task6Grader:
+    def grade(self, env=None, *args, **kwargs) -> float:
+        return _safe_score(_r6, env, *args, **kwargs)
+
+
+class Task7Grader:
+    def grade(self, env=None, *args, **kwargs) -> float:
+        return _safe_score(_r7, env, *args, **kwargs)
+
+
+# Function aliases so app.py /grader endpoint and task_registry still work
+grade_task1 = Task1Grader().grade
+grade_task2 = Task2Grader().grade
+grade_task3 = Task3Grader().grade
+grade_task4 = Task4Grader().grade
+grade_task5 = Task5Grader().grade
+grade_task6 = Task6Grader().grade
+grade_task7 = Task7Grader().grade
 
-# Aliases for /metadata endpoint which generates grade_{task_id} names
 grade_task1_json = grade_task1
 grade_task2_yaml = grade_task2
 grade_task3_dockerfile = grade_task3