// METHODOLOGY

Structural Assessment Methodology.

SORT-AI treats advanced AI systems as coupled execution fabrics rather than isolated model instances. In such systems, model execution, runtime orchestration, control policies, memory paths, serving layers, and agentic workflows interact as a composed system. The most relevant effects often arise between components, across layers, and along coupling surfaces. SORT-AI therefore provides a Level-0 Structural Assessment Framework for reading these composed systems before implementation-specific telemetry, scoring, weighting, thresholds, intervention logic, or runtime execution engines are introduced.

Publication Sequence

The methodology is organized across three linked research artefacts. The SORT-AI Domain Paper defines the canonical domain architecture: Domain, Cluster, Application, Structural Dimensions V1–V4, Scenario Class, Metric Set, and Regime Classification. The V1–V4 Diagnostic Protocol formalizes how an observed AI-fabric condition becomes a structurally assessable case. The Kernel-Damping Evidence Protocol then defines how selected declared risk-transition cases can be reproduced mathematically as analysis-layer evidence.

Why Methodology Matters

Modern AI systems may show healthy local metrics while degrading as composed systems. A model may pass benchmarks while the deployment fabric develops latency variance, retry amplification, cost escalation, control incoherence, or reduced reproducibility. The methodology layer exists to make these conditions structurally readable. It separates the act of identifying a structural problem form from the act of making that problem form assessable under declared scenarios, metrics, regimes, and evidence interfaces.

Public Analysis Layer

The Public Analysis Layer is the bridge between architectural classification and reproducible evidence. It does not replace observability systems, benchmarks, runtime logs, or production telemetry. Instead, it provides the ordering logic that determines how observations from those systems can be read as structural conditions.

• S_D — structured system state in domain D
• V1–V4 — diagnostic dimensions
• A_j — application identity
• C_jℓ — scenario class
• M_jℓ — metric set
• ρ_jℓ — regime classification
• E_jℓ — evidence interface

The first segment, from system observation to V1–V4, is diagnostic. It identifies what is visible, what structural relation makes it intelligible, what effect space it enters, and what decision or evidence surface becomes relevant. The second segment, from Application identity to Evidence Interface, is assessment-oriented. It anchors the diagnosis to a reusable application, a typed scenario class, a declared metric set, a regime classification, and a compatible evidence interface.

The Public Analysis Layer explains how a structural observation becomes an assessment case before it is connected to a reproducible evidence protocol or future execution layer.

Public Boundary

The methodology is published as a public analysis layer. The boundary below separates what is disclosed from what remains implementation-specific.

Methodology Pages