I created a skill for Claude Code to do a complex data analysis and create a 1500 word report from it. I prompted it to give a final review like an Editor in Chief and it looks amazing... but how would I know if there were hallucinations other than having a human review it? My human costs thousands for this work. I was thinking maybe plug the raw data into Codex and see if the findings come back the same? Any other tricks or software to verify the work?