Last released Oct 23, 2026
Trustworthiness guard for LLM agent outputs — a triple-challenge checker (provenance / assumption / contradiction)