Catena 重建从‘链下感知(经 xAPI)→ 推理 → 链上交易’的真实因果链，用反事实干预检验 Agent 声称的理由是否真的导致了它的行为，并把整条 trace + 审计结论做哈希承诺封存上链，让审计本身不可篡改。
*A verifiable audit layer that reconstructs an AI agent's real `perception → reasoning → action` causal chain, uses counterfactual intervention to test whether its stated reasons actually caused its on-chain behaviour, and seals the verdict on-chain — so the audit itself is tamper-evident.0-1
-