Off-chain verify/recompute layer for the trustless-ai boundary chain: commit-before-outcome, anchor to public sources, verify any layer's claims trusting no one.