the best you can do is show them the code and hope they catch mistakes. Data scientists who can't read code probably shouldn't be running AI generated analysis on real data.