I built an AI that judges startup ideas. So I put its own idea on trial. It demoted me.

Every validation tool I tried just encouraged me. Friends are polite, co-founders optimistic, ChatGPT is built to agree. So I built the opposite: different AI models prosecute and defend your idea, a 12-model jury scores it, and a formula calculates the verdict, no single model decides. To test if it was actually honest, I ran its own idea through it. It scored 81. I appealed, refused to show my numbers, and it dropped me to 78, with the judge calling my defense "clever but circular." Genuine question for this crowd: do founders actually want an AI that tells them their idea is weak? Or do we secretly prefer the encouragement, even knowing it's useless?