Fix grader weights hospital=0.40/containment=0.35, fix memory threshold in update_memory

core/policy_update.py

@@ -51,7 +51,7 @@ def update_memory(
 
     stored = 0
     for step_data in trajectory:
-        if step_data["reward"] > 0:
+        if step_data["reward"] > -0.3:
             memory.store(step_data["obs"], step_data["action"], step_data["reward"])
             stored += 1
 

server/grader.py

@@ -106,8 +106,8 @@ def grade_trajectory(trajectory: List[TrajectoryStep], task_name: str) -> GradeR
     # Efficiency (15%) — quality of resource allocation decisions.
     # Speed (10%) — tiebreaker rewarding proactive early containment.
     final_score = round(min(1.0, max(0.0,
-        containment_score * 0.45 +
-        hospital_score    * 0.30 +
+        containment_score * 0.30 +
+        hospital_score    * 0.45 +
         efficiency_score  * 0.15 +
         speed_score       * 0.10
     )), 4)