Constraint Alignment introduces a general framework for understanding stability in complex systems. It proposes that system failure is not primarily driven by external shocks or insufficient capability, but by the accumulation of misaligned and unbounded complexity relative to underlying constraints. The framework defines a lifecycle of complex systems—growth, drift, fragility, trigger, and collapse—where collapse is understood as a process of forced compression toward constraint-aligned structures. Within this model, stability is determined by a system’s capacity to continuously constrain, verify, and adapt its complexity. To operationalise this concept, the framework introduces the Constraint Alignment Score (CAS), a composite metric based on seven dimensions: bounded capability, verifiability, modularity, failure containment, human override, feedback fidelity, and drift detection. CAS enables comparative evaluation of system stability independent of capability or scale. The framework is designed to be domain-agnostic and is demonstrated through applications in artificial intelligence (CAS-AI) and education systems (CAS-Ed), showing consistent patterns of fragility, localized collapse, and selective persistence of constraint-aligned structures. Constraint Alignment provides both a diagnostic lens for identifying instability and a design principle for building systems that remain robust under real-world conditions. It predicts that future system development will shift from capability maximisation toward constraint-aligned architectures that prioritise verifiability, modularity, and controlled failure.