The Eval Stack: Proving the agents are right instead of claiming It

Most AI research tools point a language model at the web and trust the output. Saarth Shah built Sixtyfour around the opposite instinct: grade everything, ship only what improves the score. Saarth Shah keeps a scoreboard.