Adversarial Participation Without a Judge: Fail-Closed Containment for No-Meta, Observable-Only Agents

This record provides a machine-auditable specification for observable-only, no-meta autonomy: decision-making and coordination protocols that do not rely on privileged evaluators, hidden scoring, or unverifiable authority. The core design goal is fail-closed safety with replayable evidence. All eligibility, caps, closure, and action authorization are derived from deterministic, bounded, and publicly checkable rules over a canonicalized event log. Key contributions Deterministic canonicalization (encoding + normalization + domain-separated hashing) so independent parties can reproduce identical bytes and hashes. Authority-bearing commitments with non-circular roots (rejection noise and test channels are…