Final

inference.py

@@ -219,10 +219,10 @@ async def run_task(client: OpenAI, task: str) -> float:
             if done:
                 break
 
-        # Normalise total reward → [0.0, 1.0]
+        # Normalise total reward → [0.0, 1.0] (Clamped to (0,1) for OpenEnv compliance)
         task_max = TASK_MAX_REWARD.get(task, 84_000.0)
         raw      = sum(rewards)
-        score    = float(min(max(raw / task_max, 0.0), 1.0))
+        score    = float(min(max(raw / task_max, 0.001), 0.999))
         success  = score >= SUCCESS_THRESHOLD
 
     except Exception as exc: