fix: clamp scores to strictly (0.01, 0.99) — no exact 0.0 or 1.0

app/environment.py

@@ -41,6 +41,7 @@ class IncidentResponseEnv:
     def state(self) -> StateModel:
         grader_fn = GRADERS[self.task_id]
         score, _ = grader_fn(self.grader_state)
+        score = max(0.01, min(0.99, score))
         return StateModel(
             task_id=self.task_id,
             step=self.step_count,
@@ -72,6 +73,7 @@ class IncidentResponseEnv:
 
         grader_fn = GRADERS[self.task_id]
         current_score, _ = grader_fn(self.grader_state)
+        current_score = max(0.01, min(0.99, current_score))
 
         # check done conditions
         max_steps_reached = self.step_count >= self.meta["max_steps"]

app/tasks.py

@@ -25,6 +25,7 @@ def grade_easy(grader_state: Dict[str, Any]) -> Tuple[float, str]:
         score += 0.4
         reasons.append("applied correct fix (+0.4)")
 
+    score = max(0.01, min(0.99, score))
     return round(score, 2), "; ".join(reasons) if reasons else "no progress"
 
 
@@ -61,6 +62,7 @@ def grade_medium(grader_state: Dict[str, Any]) -> Tuple[float, str]:
         score += 0.2 * grader_state["postmortem_quality"]
         reasons.append(f"postmortem quality (+{0.2 * grader_state['postmortem_quality']:.2f})")
 
+    score = max(0.01, min(0.99, score))
     return round(min(score, 1.0), 2), "; ".join(reasons) if reasons else "no progress"
 
 
@@ -103,6 +105,7 @@ def grade_hard(grader_state: Dict[str, Any]) -> Tuple[float, str]:
         score += 0.20 * pm_quality
         reasons.append(f"postmortem quality (+{0.20 * pm_quality:.2f})")
 
+    score = max(0.01, min(0.99, score))
     return round(min(score, 1.0), 2), "; ".join(reasons) if reasons else "no progress"