Where Frontier AI Confuses Change: A Recurrence–Structure Audit of Contemporary Language Models and Structural Intelligence Claims

Contemporary frontier language models are often described as adaptive, emergent, or self-improving systems. In many cases these descriptions conflate two distinct forms of change: variation in system states (Δx) and mutation of the governing rule class (ΔS). This paper introduces a recurrence–structure audit for evaluating structural claims in machine learning systems. Using the Structural Intelligence Theory (SIT) criterion, structure is defined as a rule object g = (O, C, U) composed of an objective class, constraint class, and update operator class under a declared invariance relation (~). Recurrence is defined as repeated activation of comparable problem families under declared invariances and logging conditions. The Rule–State Separation Assumption (RSSA) serves as a terminal gate: if state variation and rule mutation cannot be separated under the declared regime, structural classification resolves as Undefined. The audit evaluates three dominant claim clusters in frontier AI discourse: scaling and fine-tuning tool use and memory systems self-correction and recursive improvement Across typical deployment boundaries, most observed improvements classify as state variation within a fixed rule class (Ψ = 0). Structural transitions occur only when the rule object itself changes. The framework converts semantic debates about emergence and intelligence into structurally testable classifications under declared recurrence conditions. Intellectual Property & Licensing The KOGNETIK Research Series is released under the Creative Commons Attribution–NonCommercial 4.0 International License (CC BY-NC 4.0). All scientific works within the series may be cited, shared, and adapted for non-commercial research purposes with proper attribution. Commercial use—including consulting, advisory services, integration into commercial platforms, monetized training, certification, or system-level deployment—is not permitted under this license and requires a separate written agreement. Full license text:https://creativecommons.org/licenses/by-nc/4.0/ For licensing, partnerships, translations, or applied development inquiries:research@kognetik.dehttps://www.kognetik.de ORCID: https://orcid.org/0009-0000-8544-4847 Kognetik Series Information KOGNETIK — Minimal Operator Definition of Reflexivity (Ψ = ∂S/∂R) Reflexivity as structural rate-of-change:Ψ = ∂S/∂R measures structural drift under recurrence. Process, not state:Reflexivity specifies a transformation rule rather than a content or level. Domain-independent operator:Applicable across biological, cognitive, artificial, social, industrial, and geophysical systems. Non-ascriptive and empirically testable:Ψ enables comparative analysis of systems via observable structure and recurrence. Higher-order phenomena as specifications:Learning, adaptation, consciousness, governance, and identity are structured regimes of Ψ.