Interpretive Behavior of Large Language Models Under a Novel Formal Coordination Framework: A Comparative Study of Local and Cloud-Deployed Systems

This record contains a comparative study examining how large language models (LLMs) respond to structured novelty when presented with a fabricated but internally coherent formal framework. The study introduces “3-Way Architecture (3WA)” explicitly as a diagnostic probe rather than a substantive coordination theory. Identical priming instructions and analytical prompts were applied across two model populations: locally hosted open-weights models and cloud-deployed proprietary models. Rather than evaluating the validity of the framework, the study analyzes model interpretive behavior under novelty. Results show a consistent divergence. Local models predominantly exhibit instruction-compliant explanatory expansion, while cloud models adopt an epistemic review posture characterized by boundary questioning, scope analysis, and identification of unresolved tensions. These differences are interpreted as effects of deployment context and training incentives rather than model size alone. Included in this record are the full working paper (PDF) and verbatim interaction logs for both local and cloud model runs. The framework and datasets are intended as methodological instruments for probing epistemic behavior, independent of truth claims or theoretical endorsement.