Anti-hallucination engineering harness. risk = confidence^2 x (1 - evidence). i^2 = -1, implemented. MIT.
ai-agents grounding llm hallucination-detection agent-evaluation anti-hallucination agent-harness harness2 fact-layer google-harness2
-
Updated
Oct 22, 2026 - Python