Authority, Evidence and Durable Continuity: A Five-Cycle Rust Experiment in ESS-MAI

Research preprint accompanying ESS-MAI POC011 version 0.1.0, by Bledar Gjata. This four-page preprint describes a bounded five-cycle Rust experiment with scalar verification outcomes 0, 0, 0, 1, 1 and two committed local effects. It documents distinct LIGHT coordination, QUANTUM candidate elimination and SHADOW verification competences; D-BAB witnessing; a borrowed 0-NOI primitive; one-use FFI handle consumption; and context-linked durable memory. The proposed AI Authority paradigm is explained separately from the mechanisms demonstrated by this capsule. Recorded local evidence comprises 38 Rust tests passing in debug and release, 17 independent Python verifier self-tests, exact source provenance and SHA-256 manifests. Selected primitives are source extractions; the finite affine model and orchestration are documented adaptations. This is not independent peer review, security certification, a full native hierarchy proof or a completed VERITAS integration. Missing evidence remains distinct from a valid counterexample. RAM write/readback does not measure physical energy, hashes do not authenticate authority, and valid-prefix WAL rollback is an explicit limitation. Reproduction note, 30 October 2026: use Rust 1.98.0. Initial hosted compilation with Rust 1.99.1 stopped because fetch_update was deprecated and warnings are denied. Pinning the recorded toolchain allowed full Ubuntu and Windows reproduction in run 41077100247, without changing runtime code or tests. The deposited PDF and software archive remain unchanged; the repository compatibility note records the correction. AI tools assisted implementation and writing under the author’s architecture, direction and review. Independent replication, criticism and attributed technical contributions are welcome. Versioned software, evidence and checksum archive GitHub repository and review protocol Hugging Face byte-identical mirror Originating ESS-MAI project ESS-MAI archive Author ORCID Author LinkedIn OpenAIRE discovery route (indexing not yet confirmed) PREreview request (not a completed review or endorsement)