An Axiom‑Driven Meta‑Paradigm for Long‑Term Human‑Large‑Model Collaborative Interaction Contracts

Large language models exhibit well‑known shortcomings within continuous conversational workflows: constraint decay, unauthorized intrusion of implicit model‑derived priors, and the lack of structured architectural mechanisms to contain inherent inference uncertainties. As dialogue context accumulates, high‑priority constraints defined by human users tend to degrade in effectiveness. Implicit assumptions embedded within model weights frequently intrude reasoning processes without explicit human authorization. Existing prompt‑engineering practices largely consist of ad‑hoc empirical templates and fragmented heuristics, and they lack a theoretically grounded, hierarchically constrained design foundation. This paper introduces an architecture‑centered meta‑paradigm for sustained human‑model collaboration, termed the Calm‑Ocean Contract Paradigm. The proposed paradigm establishes a multi‑layer constraint system composed of a supreme kernel for judgment, a task‑mapping carrier layer, an inference‑anomaly containment layer, and a pluggable business‑logic adaptation layer. Critical ontological specifications for the top‑level kernel, refined layer‑wise boundary criteria, and conflict‑arbitration logic are preserved as private intellectual assets of the author and are not released in this manuscript. This paper further formulates the Carrier Constraint Theorem, which explains why conventional conversational contexts cannot sustain stable enforcement of high‑priority contractual constraints, and identifies necessary architectural prerequisites for reliable long‑run human‑model collaboration. Defensive Note: This manuscript publishes only the conceptual architectural skeleton. A fully functional instantiation of the Calm‑Ocean Contract Paradigm depends on undisclosed private ontological rules. Any third‑party open‑source implementations are derivative works inspired by the public architecture, and shall not be regarded as complete reproductions of the original Calm‑Ocean system. Empirical performance metrics, failure‑mode analysis, full case studies, and core ontological texts are not included in this public preprint and are available via correspondence. Large‑scale quantitative benchmark evaluation is reserved for future public research outputs.