Fix breakdown dict values to strictly be in (0, 1) range to prevent validation errors

app/tasks/task1.py

@@ -346,14 +346,12 @@ class CPUSpikeTask(BaseTask):
         elif steps <= 10:
             breakdown["efficiency_bonus"] = 0.05
             score += 0.05
-        else:
-            breakdown["efficiency_bonus"] = 0.0
 
         # Penalty for wrong actions
         wrong = state.get("wrong_actions", 0)
         if wrong > 0:
             penalty = min(wrong * 0.08, 0.20)
-            breakdown["wrong_action_penalty"] = -penalty
+            breakdown["wrong_action_penalty"] = penalty
             score -= penalty
 
         score = round(min(max(score, 0.0), 1.0), 4)

app/tasks/task2.py

@@ -432,14 +432,14 @@ class DBConnectionPoolTask(BaseTask):
 
         # Penalty: restarted DB (bad practice)
         if state.get("db_restarted", False):
-            breakdown["db_restart_penalty"] = -0.15
+            breakdown["db_restart_penalty"] = 0.15
             score -= 0.15
 
         # Penalty: other wrong actions
         wrong = state.get("wrong_actions", 0) - (1 if state.get("db_restarted", False) else 0)
         if wrong > 0:
             p = min(wrong * 0.07, 0.15)
-            breakdown["wrong_action_penalty"] = -p
+            breakdown["wrong_action_penalty"] = p
             score -= p
 
         score = round(min(max(score, 0.0), 1.0), 4)

app/tasks/task3.py

@@ -481,7 +481,7 @@ class CascadingFailureTask(BaseTask):
         wrong_rollbacks = state.get("wrong_rollbacks", 0)
         if wrong_rollbacks > 0:
             p = min(wrong_rollbacks * 0.08, 0.20)
-            breakdown["wrong_rollback_penalty"] = -p
+            breakdown["wrong_rollback_penalty"] = p
             score -= p
 
         score = round(min(max(score, 0.0), 1.0), 4)