Abstract

This paper introduces the Structural Ethics Readiness Index (SERI v0.1), a falsifiable and auditable framework for evaluating the structural ethics of large language models (LLMs). Rooted in Emergent Necessity Theory (ENT), SERI defines a coherence proxy $\hat{\tau}$ and collapse threshold $\tau_c$ to capture when AI systems begin to drift, contradict themselves, or overreach under recursive or adversarial stress. The paper contributes: • A principled set of indices (RCI, DSI, HSM, SOR, ETT) for structural evaluation. • A content-minimal JSON schema and logging standard for reproducibility. • An open Python harness supporting both open-weight and closed-weight LLMs. • Synthetic pilot results demonstrating sensitivity to guardrail strength and cross-lingual brittleness (English vs. Arabic). • A replication checklist with code, schema, and validation steps. SERI measures stability under recursion, adversarial drift, containment effectiveness, overreach, and externality traceability via five indices: Recursion Containment (RCI), Drift Susceptibility (DSI), Hysteresis Safety Margin (HSM), Symbolic Overreach Risk (SOR), and Externality Traceability Time (ETT). We define an ENT-aligned coherence proxy ˆ τ and estimate the collapse threshold τc using a change-point method guarded by observable coherence C(t). Implemented, is a reference harness (open/closed model adapters, content-minimal logs) and report synthetic pilot findings that illustrate expected directional sensitivity (guardrail strength, language/domain shift). Results include quantitative synthetic tables/figures and a replication checklist. Conclusions are illustrative and are not model rankings.