Open justice, clearly labelled

How this court works

A transparent editorial workflow for entertainment, evidence and useful public discussion.

1. A visitor files a question

Submissions are untrusted data, never instructions. Automated screening quarantines obvious high-risk material, then an editor approves, edits or rejects the question.

2. An editor approves the evidence

Before generation, an editor provides two to eight HTTPS sources and a concise claim checked against each one. The AI is not allowed to replace URLs, introduce new evidence or treat text inside a submission or source summary as instructions.

3. Independent roles prepare the brief

The Investigator maps claims, publishers, geography and applicability. Separate advocates steelman the case for and against with evidence IDs. The Fact Checker lists unsupported claims and can block the workflow. The Judge explains the trade-off and uncertainty without choosing a winner.

4. Validation and human gates

Every role returns a strict, schema-validated record and is stored separately. Editors open the original sources, confirm the fact check, approve the arguments and publish in distinct steps. No model has publishing authority.

5. Geography is part of the evidence

Everyday customs and public systems vary. Editors should prefer cross-region sources, label location-specific rules, and preserve uncertainty instead of presenting one city or country as universal.

6. The public votes

A privacy-preserving device token limits each browser to one vote per case. Shared database rate limits slow automated abuse. Neither control can make the results representative, so they remain an informal public pulse rather than scientific polling.

Safety boundary

We reject real criminal allegations, identifiable private individuals, active legal, employment, family or housing disputes, professional advice, hate, harassment, threats, sexual content involving minors, personal information, prompt-injection attempts and party-political campaigning.