An AI that criticizes your startup idea, not one that flatters it

The problem

Why an AI that criticizes, not one that praises

A single LLM answering in a chat tends to agree with whoever is asking. It's a known bias: without anyone tasked with attacking, the model drifts toward easy praise, even for a flawed idea. That doesn't help someone deciding whether to spend months, or savings, on it.

Tribunal of Ideas flips the design. Your idea is put on trial by multiple AI models in opposing roles: 3 prosecutors tasked with attacking, 3 defenders tasked with responding, 6 witnesses (3 deliberately chosen for, 3 against, to force both sides), and a jury of 12 models voting on 9 criteria, plus 1 judge who delivers the sentence. In total, that's 25 different AI models, one per seat, so no model judges alone or confirms its own bias.

The result isn't a score a chatbot decided to hand you to make you feel good. It's a calculated verdict, held to the same rigor as someone who has to defend the number afterward.

The rite

How the trial works

1. 1

   Real evidence

   Before any argument happens, a market search surfaces real pain points and named competitors, each with a source. That's what prosecution, defense, and witnesses cite, not guesswork.

2. 2

   Prosecution

   3 prosecutors, each a different AI model, attack the idea based on the collected evidence: risks, competition, weak points.

3. 3

   Defense

   3 defenders, also on different models, answer point by point, rebutting each risk the prosecution raised.

4. 4

   Witnesses

   6 witnesses, each a different AI model, deliberately biased: 3 escalated in favor do everything to sustain the idea, 3 against do everything to tear it down, each citing the same pain points and competitors surfaced in the search. That clash is what produces real cross-examination.

5. 5

   Jury

   12 jurors, each another AI model, vote on 9 equally weighted criteria. Each juror then reviews their own vote against the others' scores, shown anonymously, before locking it in.

6. 6

   Calculated verdict

   The final score doesn't come from any model's mouth: it's the normalized average of the 9 criteria. 80% and up is Innocent; 60% to 79% is Innocent with concerns; below 60% is Guilty.

What you get

What you get

A verdict across 9 equally weighted criteria (relevance, solution, differentiation, market, traction, monetization, timing, execution, and risk handling), each with the jury's score and reasoning. An idea that's still just an idea isn't penalized for traction or execution that only a running business would have.

The judge's sentence comes with an action plan anchored in the evidence gathered, and you also get a private counsel brief listing the risks and assumptions worth validating before you move forward.

Every citation used in the trial links back to a source: seats echo URLs from the search, they never invent data. And if the verdict doesn't convince you on the first try, every trial includes one appeal with cross-examination, at no extra cost.

FAQ

Does the AI only criticize, never approve?

No. The jury votes on 9 criteria based on real evidence, and the outcome can land as Innocent, Innocent with concerns, or Guilty. What changes is the process: instead of a single chatbot trying to please you, prosecutors, defenders, and witnesses argue it out before the jury decides.

Is it paid?

Yes. Running a trial on your own idea is always paid, through credits (card or boleto via Stripe). What's free on the site is the showcase: replays of full trials other people already ran.

How much does it cost?

Credit packs start at US$ 4.90. Check the pricing page for the full breakdown before putting your idea on trial.

How is the verdict calculated?

The 12-model jury votes on 9 equally weighted criteria, each score anchored in the real evidence surfaced by the search. The final score is the normalized average of those 9 criteria, converted to a percentage: no AI model sets the number, it's a formula.

How many AIs judge my idea?

25 in total: 3 prosecutors and 3 defenders attack and respond, 6 witnesses testify (3 deliberately chosen for and 3 against, to force both sides instead of a mild consensus), 12 jurors vote on the 9 criteria, and 1 judge delivers the sentence, each seat a different AI model. The multiplicity of models cuts down on hallucination and produces real back-and-forth, instead of a single model arguing with itself.

Can I respond to the criticism after the verdict?

Yes. Every trial includes one free appeal with cross-examination: prosecution and defense ask about the case's weakest points, you answer whatever you choose, and the same jury votes again in light of the full record.