Live · AI Court

Everyone flatters your idea. We put it on trial.

Dozens of different AI models examine your idea from every angle: they research real evidence and competitors on the web, prosecute, defend and vote. And you watch it, live.

Write your idea and take it to trial

Your trial will be conducted in English.Escrever em português

Inadmissible idea? Your credit is refunded.

Why it exists

Nobody tells you the truth about your idea.

Friends are too polite. Co-founders are optimistic. ChatGPT was built to encourage. You're left with an idea in your head and no honest feedback. That's what the Court is for: a prosecutor tries to convict it, a defender stands for it, and a jury of different AI models disagrees right in front of you. Authority — without the flattery.

The proceedings

A full trial, from charge to verdict.

Each step lights up in front of you, live. Then it's saved to replay.

1
Web search
Before the debate, a real-time web search anchors everything in real pain points and competitors.
Without it, the trial would be just opinion.
2
Evidence
An analyst separates real pain points from guesswork, each one linked to its source.
The factual base that blocks flattery.
3
Prosecution
Three prosecutors, each a different model, attack the idea in parallel.
Someone has to try to tear the idea down.
4
Defense
Three defenders read the accusation and rebut it point by point.
So it's a fair debate, not an execution.
5
Witnesses
Six witnesses testify: three for, three against.
Both sides, not just the fans.
6
Jury
Twelve jurors, each a different model, score nine criteria.
Many models dilute the bias of one.
7
Cross-review
The twelve re-score after seeing each other's marks — anonymously.
Blind: the argument wins, not the brand.
8
Verdict & sentence
The verdict is calculated by formula. The judge writes the sentence and action plan.
The score comes from math, not a model's mood.

See it in action

Three real ideas. Three fates.

Complete, unscripted trials: one idea condemned, one saved with concerns, one acquitted. Click a card to watch the full replay.

A social network exclusively for left-handed people to post photos of their daily routine. It targets the approximately 10% of the population that is left-handed and aims to monetize through advertisements.

This idea is found GUILTY of market delusion: it proposes building a standalone social network for left-handers with zero demand signal, no viable path to critical mass, and fatal competition from entrenched platforms that already serve this community.

3min 19s46 AIs
32Guilty
Watch the trial

A Mac menubar app that automatically tidies the Desktop and Downloads folders using customizable rules. It targets any Mac user overwhelmed by numerous files and offers a one-time purchase model without the need for accounts or cloud services.

A technically sound Mac utility addressing real desktop clutter pain, but entering a mature market without a defensible moat beyond pricing and simplicity claims that remain unvalidated.

4min 0s45 AIs
72Innocent with concerns
Watch the trial

A payment and credit infrastructure that connects manufacturers, distributors, and retailers in Brazil, facilitating order, payment, and credit processes. It leverages a two-sided network effect and proprietary data to sustain growth and profitability in the wholesale distribution market.

This idea is found INNOCENT: a profitable, high-traction infrastructure play with proven network effects and ERP lock-in, though vulnerable to commoditization if it fails to deepen its credit moat and defend against public rails.

3min 54s46 AIs
87Innocent
Watch the trial

Each card's polygon is the shape of the verdict: the 9 criteria scored by the jury.

Would your idea survive the tribunal?

The scale

Over 40 AI requests per trial.

Most AI evaluations are one model's opinion. Here, every seat is a different model, drawn from a bench of over 30 models across 20 AI companies.

Prosecution × Defense
3 models per side
6 witnesses
3 for, 3 against
12 jurors
12 different AI models
1 verdict
a 0–100 score

The 20 AI companies in the cast

OpenAI
OpenAI
Anthropic
Anthropic
Gemini
Google
Grok
xAI
DeepSeek
DeepSeek
Qwen
Qwen
Meta
Meta
Mistral
Mistral
IBM
IBM
MoonshotAI
Moonshot
Nova
Amazon
Azure
Microsoft
Minimax
MiniMax
Tencent
Tencent
NousResearch
Nous
Nvidia
NVIDIA
Arcee
Arcee
Cohere
Cohere
Upsate
Upstage
AI21
AI21

AI vs. AI

Not one AI's opinion. It's AI versus AI.

One model accuses. Another reads what it wrote and rebuts it point by point. Then twelve jurors judge the clash — and re-review each other blind, not knowing whose score is whose. That's what turns the score into the result of a contest, not the mood of a chatbot.

The defense rebuts the prosecution
The defender gets the full accusation and answers point by point — not two parallel monologues.
The jury weighs the debate
Witnesses and jurors read prosecution and defense before scoring. Nobody judges the idea in a vacuum.
Anonymous cross-review
Each juror reviews the other eleven's scores — unlabeled. Less herd effect, more truth.

The verdict

Calculated by a formula. Not decided by an LLM.

Nine criteria, all weighted equally. The average of the jurors' scores becomes a percentage, and the percentage falls into a fixed band. No model picks the verdict: it comes from a calculation you can audit. An idea that's still just an idea isn't punished for traction or execution only a running business would have.

0–59 · Guilty 60–79 · Innocent with concerns 80–100 · Innocent

The nine criteria, equal weight

Relevance
Is the problem real and important?
If the pain isn't real and frequent, no one pays to solve it. Without this, everything else is just theory.
Solution
Does it solve the pain well?
Solving it halfway keeps no one: the customer tries it, sees no value and leaves.
Differentiation
Distinct and hard to copy?
Without something hard to copy, a rival with more cash overtakes you. The moat protects the margin.
Market
Is the market opportunity big?
A small market caps your ceiling, however good the idea. It sets the size of the prize.
Traction
Any sign of real demand?
People already using or asking is the strongest proof of demand. Investors bet on signal, not promises.
Monetization
Can it make money sustainably?
Without a clear, sustainable way to charge, it's an expensive hobby. It separates a project from a company.
Timing
Why now? Is the moment right?
Too early, the market isn't ready; too late, they missed the wave. The right moment changes everything.
Execution
Feasible to build with what you have?
Anyone has ideas; building with what you have is the filter. It's where most get stuck.
Risk handling
Are the material risks plausibly addressed?
Every business carries risk; what matters is the profile: low or well-mitigated risk earns the top, severe and ignored sinks the ship.

The method

Not a gimmick. A method.

Frontier research on AI evaluation says two things: replace the single judge with a jury of diverse models, and don't let the AI set the final score. Most tools ignore both. The Tribunal was built this way from day one.

The bias of a lone AI judge
How the Tribunal corrects it
Favors its own style
A jury of models from different companies, none judges alone
Leans toward praise
Adversarial roles and scores anchored in real evidence
Prefers what it sees first
Anonymous cross-review, not knowing whose score is whose
Gives unstable scores
Verdict calculated by formula, not set by the AI

Based on multi-model jury research (LLM-as-a-Jury) for AI evaluation.

The bar

Scores here are earned.

Most evaluation tools approve almost everything. Here the verdict is calculated, convictions happen, and the distribution is public.

11.9% of ideas were convictedmedian 71only 25% reached Innocent
0-29
0%
30-39
3.6%
40-49
0%
50-59
8.3%
60-69
34.5%
70-79
28.6%median 71
80+
25%
0–59 · Guilty 60–79 · Innocent with concerns 80–100 · Innocent
Real distribution of first-instance trials, updated automatically.

Real evidence

Anchored in real pain, not guesswork.

Before the debate, we search the web for pain from people who have the problem and competitors that already exist — with a source link. Every argument is marked: real source or AI reasoning. And if the search finds nothing concrete, we tell you — instead of making it up.

with evidence AI reasoning

What you take home

The show ends. The value stays.

Forever replay
The full trial, revisitable forever. Private by default, but you can turn on the link and share it with a co-founder or investor.
Verdict card
An image with the score, verdict and your idea, ready to post. It's the trophy of the trial, made to share on social.
PDF report
The full sentence: verdict, action plan, the nine dimensions, the jury scoreboard and the evidence. Downloadable.

Reliability

If a juror falls, another steps in.

Dozens of models, with fault tolerance in every seat. If a juror doesn't respond, another model takes over its seat in its place. And if failures compromise the trial, your credit comes back. Live, in a few minutes, and you can replay it whenever you want.

Pricing

Start at $4.90.

1 credit = 1 trial, one appeal included. Credits never expire.

Single hearing
$4.90
1 credit(s)
$5 per trial
most popular
Triad
$11.90
3 credit(s)
$4 per trial
Full jury
$29.90
10 credit(s)
$3 per trial

Before anything is spent, the tribunal checks whether your petition is judgeable. If it is inadmissible, the credit is refunded.

The verdict is not the end. Your first appeal is included in the credit: the court cross-examines you on the sentence's weak points, you answer with facts, and the same jury re-votes the 9 criteria (or you just refine the text). The next ones cost 1 credit, each judged against the previous version.

FAQ

Frequently asked questions

Does this evaluate AI models, like "LLM as a judge"?

We use the same technique, a jury of many models, but aimed at your business idea, not at another AI's output.

Is the score decided by an AI?

No. The models assess criterion by criterion; the final verdict is a formula you can audit.

Is it infallible?

We don't promise perfection. The method reduces bias and is transparent: we mark what's real evidence and what's AI reasoning.

Is my idea public?

No. Private by default; you decide whether to turn on the link to share.

Why pay when ChatGPT gives feedback for free?

A single AI was trained to please you. What you pay for here is a process: over 40 requests to dozens of different models, web research with cited evidence, a real prosecution and defense, and a verdict that comes out of a formula. And if you disagree, you defend yourself: the court cross-examines you and the jury re-votes. It is a different product.

What if I disagree with the score?

You can appeal for free, once per judgment. The court raises the sentence's weak points and cross-examines you; you answer with facts and the same jury re-votes the 9 criteria. New substance moves the verdict; rhetoric doesn't.

Is there a referral program?

Yes, Refer and earn: every user has a unique link (account menu). When someone you referred makes their first purchase, you both earn 1 trial credit. Full rules live on the Refer and earn page.

Put your idea in the dock.

Write it in one sentence. The court handles the rest — live.