JACOB: Evidence-Native AI Systems — Provenance, Contradiction Control, Bounded Agency, and Reproducible Transitions (World Research Release v0.2)

Modern AI can generate convincing outputs and invoke tools faster than many systems can prove provenance, authorization, or state lineage. JACOB treats that gap as an infrastructure problem. JACOB is a project codename for an evidence-native architecture; it does not claim biological identity, consciousness, personhood, medical authority, or legal status. The architecture separates a probabilistic analytic lane from a deterministic authority lane: models may interpret and propose, while registered policy objects, capabilities, certificates, and transition receipts control state-changing actions. The release covers IER-G4 contradiction control, a resolution-certificate protocol, a machine-readable Model Birth Certificate, a Ten-Rail engineering contract, the RDM research interface, Möbius normalization (z=(s-1)/s, explicitly not a proof of the Riemann hypothesis), synthetic signal and visualization modules, a pairing-cryptography boundary, a security/threat model, evaluation metrics, and a visual provenance ledger. Speculative visuals, mathematical transforms, synthetic sensors, and software demonstrations are kept inside explicit scientific boundaries. World Research Release v0.2, September 2026. PDF rendered from the author's .docx (attached) with LibreOffice; text unchanged. The reference implementation (src/, ui/, schemas/) described in the release file map was not among the supplied files. Source: author's iCloud folder "Scientific research published final update" (deposited 2026-10-20).