AI idea validation: how the trial actually works

The mechanics

How an AI actually validates a startup idea

Ask a random chatbot if your idea is good and you'll get a loose opinion, no source, no one arguing the other side. AI idea validation on Tribunal of Ideas works differently: before any score is given, a real market search surfaces documented pain points and named competitors, each with a linked source. That evidence, not a gut read, is what feeds the entire trial.

That evidence becomes ammunition for prosecutors and defenders to argue over, and for witnesses to testify citing the same pain points and competitors. In the end, no single model does the scoring: 12 jurors, each a different AI model, vote on 9 criteria based on what was surfaced and argued, not guesswork.

The final score doesn't come from any of those models' mouths either. It's a calculated verdict: the normalized average of the 9 criteria, converted to a percentage. A chatbot decides the answer it gives you; here, a formula decides the number, and no single seat can swing the outcome alone.

The process

The pipeline, step by step

1. 1

   Admissibility

   Before any argument happens, the system checks whether the idea has enough information to go to trial. Without that, prosecution and defense have nothing to argue over.

2. 2

   Market search and evidence

   A real search surfaces documented pain points and named competitors, each with a source. That's the material, not a loose opinion, that feeds prosecution, defense, and witnesses from here on.

3. 3

   Prosecution and defense

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

4. 4

   Witnesses

   6 witnesses, each a different AI model, deliberately biased: the 3 for do everything to defend the idea, the 3 against do everything to attack it, each citing the same pain points and competitors surfaced in the search. That built-in bias is what gives the debate real back-and-forth.

5. 5

   The jury votes the 9 criteria

   12 jurors, each a different AI model, vote on relevance, solution, differentiation, market, traction, monetization, timing, execution, and risk handling, all equally weighted.

6. 6

   Anonymous cross-review

   Before locking in a score, each juror reviews their own vote against the others' anonymous scores, correcting isolated bias without letting a single juror swing the outcome.

7. 7

   Calculated verdict and sentence

   The final score is the normalized average of the 9 criteria, converted to a percentage: 80% or higher is Innocent, 60% to 79% is Innocent with concerns, below 60% is Guilty. The judge then writes the sentence with an action plan, and you also get a private counsel brief.

FAQ

Is this just an AI giving an opinion, like a regular chatbot?

No. A regular chatbot answers instantly, without searching anything and with no one to disagree. Here the idea goes through a real market search before any argument happens, then gets debated by prosecutors, defenders, and witnesses before it reaches the jury. No single model's opinion becomes the result on its own.

Where does the evidence used to validate my idea come from?

From a real market search run on your idea. It surfaces documented pain points and named competitors, each with a linked source. Prosecution, defense, and witnesses cite that same evidence throughout the trial, and the judge's sentence includes an action plan anchored in it.

Who decides the final score, one of the AIs?

No single AI decides it. The jury's 9 scored criteria run through a fixed formula that normalizes and averages them, then turns the result into a percentage. Each model's job is to grade its own criterion; the arithmetic, not any LLM, is what sets the final number.

What happens if my idea doesn't have enough information?

The trial's first step checks exactly that: without minimum information about the problem and the proposal, prosecution and defense have nothing to argue, and the idea doesn't move on to the merits.

Why multiple AIs instead of one deciding everything?

A lone model tends to rubber-stamp its own first take, with nobody around to push back. The cast spans 25 different AI models, one per role: 3 prosecutors, 3 defenders, 6 witnesses (3 deliberately chosen for and 3 against, to guarantee both sides), and 12 jurors, plus 1 judge who delivers the sentence. That mosaic of voices genuinely disagreeing, rather than one AI echoing its own logic, is what keeps hallucination in check.

Is AI idea validation on Tribunal of Ideas paid?

Yes. Putting your own idea on trial always draws from your credit balance, topped up by card or boleto through Stripe. No charge applies to browsing the showcase, where full trials run by other people are available as replays.