This paper presents an implementation-ready specification for no-meta autonomy: autonomous controllers that must operate without any privileged external overseer, using only their own observables under partial/late/missing records and potential adversarial contamination. Rather than hard-coding a single definition of “good,” the work defines auditable observable-only invariants that make harmful behavior harder to sustain and easier to detect, recover from, and constrain. The core specification is expressed as four enforceable floors—Visibility, Contraction, Contact, and Dissipation—computed from certified logs and bounded indicators. A concrete protocol, the Self-Recognition Kernel (SRK), enforces these floors via fail-closed gating, time-respecting action indexing, and commit-time rules. The design includes Hashable Numeric Encoding (HNE), taint-aware ingress normalization, deterministic hashing/canonicalization, witness-based certification policies, and anytime-valid epoch-wise bounds scoped to auditable randomness. To improve operational robustness and reduce reviewer ambiguity, the paper normalizes terms such as “suspicious,” “severe breach,” and “privilege” into testable checks with explicit inputs, observables, thresholds, and evidence traces. The appendices provide minimal schemas and object sketches (manifest/allowlist/quorum policies), operational check templates, and hard-containment triggers intended to support reproducible verification and practical implementation.
K Takahashi (Thu,) studied this question.