No-Meta Epistemic Irreversibility under Finite Memory
K Takahashi · Zenodo (CERN European Organization for Nuclear Research) · 2025
No-meta self-recognition for finite-memory autonomous systems: this preprint defines operational “meaning” as a versioned probe signature plus an explicit coding-convention ID, and builds a monitoring-and-commit instrument that can certify when meaning becomes non-reconstructable.It provides two complementary irreversibility certificates—an average-case conditional-entropy bound (mutual-information ceiling from a B-bit committed core) and a worst-case collision certificate (fractional occupancy / pigeonhole guarantees) under meaning-respecting and interface-closed commit contracts.It also connects epistemic irreversibility to physics via conditional erasure: resetting a meaning register while retaining side information yields a Landauer-consistent lower bound on minimal average work proportional to H(Z|S_core). To make this an auditable energetic ledger from monitoring, the instrument distinguishes certified lower bounds (for provable dissipation baselines) from warning-only upper-bound proxies. Operationally, the paper introduces “core meaning packages” Z=(signature, witness, identifiers), fail-closed admission-controlled commits with explicit format overhead, and an admitted-class decodability contract that prevents confusing “entropy feasibility” with “exact reconstructability.” Monitoring is structured into three states—certified-feasible, uncertain, and certified-impossible—with mitigation actions (coarsening, quarantine, budget increase, epoch roll-over) triggered by conservative certificates. Optional appendices outline anytime-valid two-sided entropy bounds under explicit sampling assumptions and bounded-memory sketching primitives for distinct-package growth, enabling practical self-monitoring under drift without relying on a privileged external evaluator.