Evaluation

This project aims to evaluate how well structured reasoning improves reliability of outputs.

Current Status

• Qualitative examples demonstrate behavior
• System identifies unsupported or exaggerated claims

Planned Evaluation

• Dataset of claims (news / public statements)
• Comparison with baseline LLM outputs
• Metrics:
• Credibility accuracy
• Detection of unsupported claims
• Quality of reasoning

Goal

To measure whether structured reasoning reduces misleading or unsupported outputs.