Epistemic Stress Test — Claude Sonnet 5 validated by MarCognity-AI

The article describes a validation of Claude Sonnet 5 using MarCognity-AI’s Skeptical Agent to expose the gap between textual confidence and actual verifiability, termed "Epistemic Fracture."

• The process involved passing model-generated responses through the Skeptical Agent.
• The goal was to reveal claims that cannot be traced back to any source, even in technically correct answers.
• Resources for MarCognity-AI are available on GitHub and Zenodo.