Target behavior
- Use only supported context
- Reject insufficient evidence
- Return structured analysis
- Produce valid machine-readable output
- Attach precise citations
Open resources
The project connects model releases on Hugging Face with scripts, evaluation material and documentation on GitHub.