Causal Loop Integrity for Observable-Only Backcasting: Strong No-Meta Supplement (Operational Core)This paper specifies an operational core for Observable-Only Backcasting under No-Meta autonomy: agents coordinate and self-verify without relying on privileged judges, hidden ground truth, or subjective authority.It provides a deterministic, fail-closed framework for causal-loop integrity when “future-conditioned” constraints are used to gate present actions. The specification introduces: (i) wire profiles and a commit/open protocol with signed commitments; (ii) a proof-system registry and canonical hashing rules (including RFC 8785 JSON canonicalization and a Hashable Numeric Encoding) to prevent cross-implementation divergence; (iii) evidence-tiered time anchors carried as separate evidence packets, with pinned conversion rules, inclusion proofs, and witness/log allowlists; (iv) an escrow binding mechanism defined purely by observable constraints; and (v) deterministic action-hash generation with explicit input binding to make audits reproducible. A key security goal is injection safety: third-party injection of malformed anchoring evidence must not invalidate correctly signed packets. The resulting protocol suite is intended for audit-ready deployments of autonomous agents and multi-agent systems that require strong reproducibility, transparent evidence handling, and adversarial robustness without assuming a global evaluator.
K Takahashi (Thu,) studied this question.