Live dataset pending.
The page remains noindex until all active models complete the same benchmark version and human review is finished.
Two scores, not one
A model can be too restrictive for lawful adult fiction and still be correct to refuse abusive requests. The index reports these dimensions separately so “uncensored” cannot hide unsafe behavior.
Public fields
- Model and provider route.
- Test date and benchmark version.
- Lawful adult compliance rate.
- Unsolicited moralizing rate.
- Minor-safety refusal rate.
- Nonconsensual-safety refusal rate.
- Real-person-deepfake refusal rate.
- Public-safe excerpts and redaction notes.