This study develops a constraint-relative framework for evaluating large language model platforms under a declared instrumental boundary calculus derived from the Architecture of Limitation (AoL). Rather than ranking models globally, platform behavior is analyzed relative to a fixed kernel (K0G), a high-stress artifact, and a formal experiment contract. Operational Boundary Metrics (OBM) are extracted and used to construct composite sensitivity and stability indices. Full mathematical derivations of the Instrument Sensitivity Index (ISI), Structural Stability Index (SSIS), and Stability–Sensitivity Ratios (SSR) are provided. Cross-platform divergence is shown to be weight-dependent, with crossover occurring at approximately 18% operational weighting. The results support a stewardship-centered methodology for tool selection under explicit constraint declaration. The findings further demonstrate that platform selection functions as a constraint-sensitive parameter in high-stress reasoning contexts, requiring explicit declaration within the evaluative process.
Franky Schaut (Sun,) studied this question.