Hallucination Probability Index (HPI)

Overview

The Hallucination Probability Index (HPI) is a proprietary algorithm that estimates the probability that a model response contains hallucinated, fabricated, or unreliable content.

Risk Levels

HPI classifies responses into four actionable risk categories:

• Low (🟢): High confidence, reliable response suitable for automated deployment.
• Medium (🟡): Moderate confidence, potential minor inconsistency. Review recommended for high-stakes tasks.
• High (🟠): Low confidence, probable hallucination. Avoid direct deployment without human-in-the-loop validation.
• Critical (🔴): Confirmed hallucinated or highly contradictory content. Response should be blocked.