test last rule only

VerifiableRewardsForScalableLogicalReasoning.py

@@ -123,7 +123,7 @@ def _evaluate_with_prolog(prediction, validation_program, eval_config, timeout=5
     # Extract configuration
     positive_pred = eval_config.get("positive_predicate", "eastbound")
     negative_pred = eval_config.get("negative_predicate", "westbound")
-    allow_multiple_rules = eval_config.get("allow_multiple_rules", True)
+    allow_multiple_rules = eval_config.get("allow_multiple_rules", False)
 
     # extract predicate from rule_to_evaluate
     rule_to_evaluate = extract_ilp_from_text_v2(prediction, positive_pred, allow_multiple_rules)
@@ -253,8 +253,8 @@ def extract_ilp_from_text_v2(text, target_predicate=None, allow_multiple_rules=F
     rule_pattern = re.compile(rf'({target_predicate}\([^()]*\)\s*:-.*?\.)')
     rules = list(rule_pattern.findall(text))
     if len(rules) > 1 and not allow_multiple_rules:
-        logger.warning(f"Found multiple rules in text, but allow_multiple_rules is set to False. Using only the first match.")
-        rules = rules[:1]
+        logger.warning(f"Found multiple rules in text, but allow_multiple_rules is set to False. Using only the last match.")
+        rules = rules[-1:]
     # Remove rules that are also captured as facts
     p_code = ''
     for rule in rules: