How many tokens did it waste building that website scraper, when all it had to do was parse some html/js?