The Scale
Every clause gets a Rights Score from 1 to 10. Higher is better for you. A 6 is plain boilerplate. We only show clauses that stand out in either direction.
Well beyond industry norms. Perpetual content licenses, AI training with no opt-out, forced arbitration plus a class action waiver.
One-sided but common. Termination at their discretion, data shared with “partners,” liability capped at a token amount.
Things most companies get wrong, done right. No forced arbitration, you keep your content, no AI training. We credit these even when the company's overall score is bad.
What the AI Evaluates
Two models read every document. Jev, a classifier built for policy text, flags and scores the passages. Claude (Anthropic) quotes and translates them, judging each on five criteria:
How much do they claim over your content, data, or behavior?
Is it disclosed up front, or buried in a dense paragraph?
Do competitors demand the same? Unique grabs score lower.
What can this actually do to you? Biometrics outrank email preferences.
Can you undo it? Clauses that survive deletion score lower.
Clause Selection
5 to 10 clauses per analysis. Each one must pass this test:
“Wait, they can do THAT?”
If it wouldn't surprise a privacy-conscious reader, we leave it out. Priority: rights they take, things they do without asking, ways they limit your recourse.
Exact Quotes Only
Every clause is quoted from the source, never paraphrased on purpose. Our software then checks each quote against the document and labels the result.
EXACT MATCH means character-for-character. CLOSE MATCH means the AI smoothed a comma or a phrase, but the substance is clearly in the source. Either way we record a hash of the displayed text on Arweave, so anyone can audit what we showed.
Community Voting
The AI rating is a starting point. As votes come in, the community score takes over. Vote counts are always shown, and the crowd can correct the AI in either direction.
The Translation
Each clause gets a plain-English version, written like a text to a friend. It's an explanation, not legal advice.
The Crazy Graph
The Crazy Graph plots every company on two axes. Up and right is better.
Termination, arbitration, liability, payment. Can you leave, sue, get a refund?
Content license, data collection, AI training, privacy, sharing. What happens to your stuff?
The Algorithm
Each axis is a weighted Rights Score of the clauses in that category:
weighted_score = (average * 0.6) + (worst_clause * 0.4)
position = (weighted_score - 1) * 1.25
The 40% weight on the worst clause means one 2/10 can't be buried under moderate ones. A weighted 5 lands at center. No flagged clauses in a category scores a 10.
You keep your data and your rights
Your data stays, your rights don't
Your rights stay, your data doesn't
They take both
What We Don't Do
No legal advice. Awareness only. Ask a lawyer for decisions.
No opinions in the score. The criteria above, not how we feel about the company.
No pay-to-change. No company can buy a better rating or remove a clause.
No hidden methods. This page is the transparency we demand of everyone else.
Disclaimer: CrazyTOS is a consumer awareness tool, not legal judgment. Ratings start with AI and shift with community votes. Clauses are quoted from public Terms of Service; the original document is always authoritative.