diff --git a/scripts/run_evmbench.py b/scripts/run_evmbench.py
--- a/scripts/run_evmbench.py
+++ b/scripts/run_evmbench.py
@@ -598,5 +598,10 @@ def run_benchmark(model, tokenizer, configs_dir, contracts_dir, use_processor=F
             )
             all_raw.append(raw)
-            all_findings.extend(findings)
+            # Findings are produced per source file, but the legacy result
+            # aggregated them without retaining that association.
+            for finding in findings:
+                finding_with_source = dict(finding)
+                finding_with_source["source_path"] = sol_path
+                all_findings.append(finding_with_source)
             if findings:
                 detected = True