Artificial Intelligence Beyond Models: A Regime-Dependent Control Framework with an Executable Core

Contemporary artificial intelligence research is dominated by models, datasets, and performance metrics.Yet the most critical failures of intelligent systems — misalignment, runaway optimization, instability, and collapse — cannot be explained at the model level. This work introduces a regime-dependent control framework for artificial intelligence, reframing AI systems as dynamic control processes rather than static optimizers. The framework is grounded in the Singularity Seed Cycle Law (SSCL), which formalizes intelligence as a cyclic process:stability → amplification → drift → collapse → re-seeding. Instead of treating collapse as failure, this work shows that collapse is a structurally necessary phase for bounded intelligent systems. The record includes: a minimal executable demonstrator (Python) illustrating regime transitions without training data or specific architectures, a one-page control-regime lens suitable for rapid understanding and citation, and a foundational mathematical framework applicable across AI, AGI, and AI governance. This upload is intentionally model-agnostic, dataset-independent, and future-proof.It is designed for researchers, theorists, and system architects who seek structural understanding, not incremental tuning. If AI alignment, safety, or long-term stability is your concern, this framework should be executed — not merely read.