New Regex Implemented in final_check.py (Also improved inference)

final_check.py

@@ -144,7 +144,7 @@ class FinalIntegrityCheck(unittest.TestCase):
             self.assertTrue(len(step_lines) >= 1, "No STEP lines found")
             for sl in step_lines:
                 step_match = re.match(
-                    r'^\[STEP\] step=\d+ action=\S+ reward=\d+\.\d{2} done=(true|false) error=\S+$',
+                    r'^\[STEP\] step=\d+ action=.*? reward=-?\d+\.\d{2} done=(true|false) error=.*$',
                     sl
                 )
                 self.assertIsNotNone(step_match, f"STEP line doesn't match regex: {sl}")
@@ -152,7 +152,7 @@ class FinalIntegrityCheck(unittest.TestCase):
             # Verify END tag format
             end_line = lines[-1]
             end_match = re.match(
-                r'^\[END\] success=(true|false) steps=\d+ score=\d+\.\d{3} rewards=[\d.,]+$',
+                r'^\[END\] success=(true|false) steps=\d+ score=-?\d+\.\d+ rewards=(?:-?\d+\.\d{2}(?:,-?\d+\.\d{2})*)?$',
                 end_line
             )
             self.assertIsNotNone(end_match, f"END line doesn't match regex: {end_line}")

inference.py

@@ -115,7 +115,7 @@ def log_step(step: int, action: str, reward: float, done: bool, error: Optional[
 
 def log_end(success: bool, steps: int, score: float, rewards: List[float]) -> None:
     rewards_str = ",".join(f"{r:.2f}" for r in rewards)
-    print(f"[END] success={str(success).lower()} steps={steps} score={score:.3f} rewards={rewards_str}", flush=True)
+    print(f"[END] success={str(success).lower()} steps={steps} score={score:.2f} rewards={rewards_str}", flush=True)
 
 
 def build_user_prompt(step: int, features: list[list[float]]) -> str:
@@ -294,7 +294,7 @@ def main() -> None:
         # Ensure absolutely no element is exactly 0.0 or 1.0 or outside the valid range.
         for i in range(len(rewards)):
             rewards[i] = float(max(0.01, min(0.99, rewards[i])))
-            
+        
         log_end(success=success, steps=steps_taken, score=score, rewards=rewards)