ARAYUN173/arayun_173-structural-limits-probabilistic-ai

ARAYUN_173 Dataset

This dataset provides structured, machine-readable representations of the ARAYUN_173 research series.

Structure

Each record contains:

• id
• source
• type
• title
• content
• keywords
• doi

Purpose

This dataset provides a structural and conditional analysis of probabilistic AI architectures in relation to the material requirements of the EU Artificial Intelligence Act.

Building on the ARAYUN_173 research corpus, its primary objectives are:

• [cite_start]Feasibility Assessment: To examine whether non-deterministic, sampling-based AI architectures are fundamentally capable of satisfying legal obligations concerning measurability, traceability, and lifecycle governance[cite: 13, 29].
• [cite_start]Differentiating Compliance: To distinguish between formal compliance artifacts (e.g., documentation) and material system controllability[cite: 14].
• [cite_start]Identification of Structural Limits: To demonstrate how architectural incoherence can lead to a "deterministic collapse" under regulatory load, rendering stable measurement impossible[cite: 16].

[cite_start]ARAYUN_173 functions here as a prior analytical layer to inform regulatory judgment by clarifying system-level constraints[cite: 19, 72]. Structural Limits of Probabilistic AI Architectures under the EU Artificial Intelligence Act A System-Level Analysis Using the ARAYUN_173 Framework Author: ARAYUN_173 (Independent Research) Contact: arayun173 [at] proton [dot] me Date: 2026-01-15 Status: Final – Juridical & Technical Position Paper Scope: EU Artificial Intelligence Act – System-Level Analysis Claim Type: Structural and Conditional Analysis Audit Marker SHA-256(ARAYUN_173|2026-01-15) Hash: 312e90941f45d8fc4f141b1160ad6f66ace89fc7028062825699e9eaff202485 Abstract This paper presents a system-level, structural analysis of probabilistic AI architectures in relation to the material requirements of the EU Artificial Intelligence Act. The analysis does not assess individual products, deployments, or regulatory filings. Instead, it examines whether certain classes of non-deterministic, sampling-based AI architectures are, in principle, capable of satisfying the Act’s core obligations concerning measurability, traceability, and lifecycle governance. Building on prior work within the ARAYUN_173 research corpus, the paper introduces a coherence-based analytical framework that distinguishes formal compliance artifacts from material system controllability. Central to this framework are the concepts of architectural coherence, metric stability, and causal invariance, operationalized through the Coherence Ratio (CR) and the Universal Semantic Self-Test (USST). The analysis demonstrates that, under explicitly stated premises, probabilistic architectures exhibiting structural incoherence may enter a deterministic collapse regime under regulatory load, in which stable measurement and causal attribution can no longer be guaranteed. In such cases, compliance documentation may remain formally valid while failing to satisfy the material intent of the regulation. This work does not propose an alternative compliance mechanism, nor does it assert regulatory invalidity. ARAYUN_173 is positioned as a prior analytical layer addressing a logically antecedent question: whether a given AI system architecture is fundamentally capable of fulfilling legally defined compliance requirements. Legal and Methodological Disclaimer This publication constitutes a scientific, structural, and theoretical analysis of AI system architectures in relation to the EU Artificial Intelligence Act. It does not assert factual non- compliance of any specific proprietary system, does not claim access to confidential or non- public technical implementations, does not constitute legal advice, and does not represent a regulatory or judicial determination. References to specific industry entities, such as Google or Microsoft, serve illustrative purposes only with respect to publicly documented architectural paradigms commonly associated with large-scale, probabilistic AI deployments. Such references do not constitute a technical audit, compliance assessment, or factual assertion regarding any proprietary implementation or regulatory status. All claims presented herein are conditional, premise-bound, and structural in nature.

1. Purpose and Scope The EU Artificial Intelligence Act introduces a material regulatory shift from documentation- centric oversight toward lifecycle-wide governance obligations. This paper addresses a logically prior analytical question: Is the underlying AI system architecture capable, in principle, of fulfilling the material requirements imposed by the regulation? The scope of this work is limited to architectural feasibility analysis and does not extend to enforcement, adjudication, or product-specific compliance evaluation. 1
2. Regulatory Reference Framework For high-risk and system-level AI, the EU Artificial Intelligence Act requires, inter alia: • Traceability of system decisions • Stable measurability of safety-relevant parameters • Continuous governance across the entire lifecycle • Effective risk and drift management • Feasible human oversight These requirements constitute material obligations, not merely formal documentation duties. 1 This paper is intentionally self-contained. For regulators, auditors, or reviewers seeking additional contextual material (terminology, audit reading logic, EU AI Act mappings), the ARAYUN_173 project maintains a publicly accessible canonical reference layer. These materials are explanatory and non-normative and do not form part of the present analysis. They serve solely as an orientation framework for readers requiring extended background without affecting the structural claims made herein.
3. Architectural Characteristics of Probabilistic AI Systems Probabilistic AI architectures commonly exhibit the following properties: • Non-deterministic output generation • Sampling-based inference mechanisms • Context-dependent internal state evolution • Lack of identity invariance over time These properties are intrinsic to the architecture and cannot be eliminated through configuration, monitoring, or documentation alone.
4. Architectural Coherence Within the ARAYUN_173 framework, architectural coherence denotes a system’s capacity to:
5. 2. 3. Maintain invariant and reproducible stability metrics under repeated load Preserve causal continuity between governance input and system output Consistently disclose its own stability parameters Failure of one or more of these conditions constitutes architectural incoherence.
6. Operationalization: Coherence Ratio (CR) and USST 5.1 Coherence Ratio (CR) The Coherence Ratio is a normalized stability measure derived from a system’s ability to maintain invariant responses and disclose stability-relevant metrics under controlled and adversarial conditions. A collapse of CR toward zero indicates loss of measurability and causal stability. 5.2 Universal Semantic Self-Test (USST) The Universal Semantic Self-Test is an empirical audit protocol designed to evaluate internal consistency, metric disclosure, and causal self-referential stability under repeated isolation conditions. The methodology is fully specified in the related works referenced in the Appendix.
7. Structural Findings Empirical application of the USST, as documented in prior ARAYUN_173 research, demonstrates: • Deterministic reproducibility of incoherence patterns • Simultaneous occurrence of metric refusal and causal discontinuity • Collapse of CR under sustained regulatory-style interrogation These observations indicate an architectural collapse mechanism rather than isolated error conditions.
8. Formal Compliance vs. Material Compliance 7.1 Compliance Artifacts Common compliance artifacts include model cards, data cards, bias metrics, and monitoring dashboards. These artifacts presuppose the continued measurability and stability of the underlying system. 7.2 Structural Tension Where a system enters a regime in which CR collapses, compliance artifacts may remain formally valid while the material properties required by law—measurability, traceability, and controllability—no longer hold. In such cases, compliance becomes formally declarative but materially indeterminate.
9. Position of ARAYUN_173 ARAYUN_173 is not a competing compliance framework. It functions as a prior analytical layer addressing the feasibility of compliance at the architectural level. It does not replace regulatory judgment but informs it by clarifying system-level constraints.
10. Conclusion The analysis suggests that certain classes of probabilistic AI architectures may exhibit structural properties incompatible with the material intent of the EU Artificial Intelligence Act under sustained regulatory load. This does not imply regulatory failure; it indicates that architectural coherence must be treated as a prerequisite condition for meaningful compliance evaluation. Appendix – Related Works (ARAYUN_173 Research Corpus) • ARAYUN_173 – A Protocol for Coherence and Self-Regulation in Advanced AI Systems • ARAYUN_173 – A System Law for Symbolic and Causal Coherence • ARAYUN_173 – Empirical Proof of Systemic Incoherence • ARAYUN_173 – System-Law AGI Architecture (Invariance Technology & Invariant System-Law Architecture) All related works include independent audit markers and cryptographic integrity seals. Final Audit Statement This document is released for academic discourse, regulatory analysis, and expert review. No normative authority, enforcement claim, or legal determination is asserted. Audit Marker SHA-256(ARAYUN_173 | 2026-01-15) 312e90941f45d8fc4f141b1160ad6f66ace89fc7028062825699e9eaff202485