构建可观察性工程框架，用信息结构解释传感系统的区分能力。

This article develops a four stage observability engineering framework for interaction driven sensing. The framework integrates physical modeling, information geometry, symmetry reduction, cross system correspondence, and observable world models into a unified engineering perspective, where sensing actions induce information structures that govern local distinguishability, while physical symmetries define equivalence classes of states that remain fundamentally indistinguishable. The proposed framework explains several common phenomena in modern sensing systems. Physically different architectures can exhibit comparable sensing capability when they induce similar quotient space information structures, whereas insufficient action diversity can lead to blind directions, ill conditioned estimation, and false identifiability. The framework also provides a quotient space language for comparing heterogeneous sensing architectures beyond hardware level descriptors, while clarifying how different forms of sensing diversity contribute to distinguishability. The framework is developed primarily through high frequency electromagnetic sensing, where geometric optics approximations provide an analytically interpretable physical realization. The broader observability engineering perspective, however, does not depend on geometric optics itself; it requires an action indexed observation family and a task relevant distinguishability measure. From this viewpoint, observability is not a passive property of sensor measurements alone, but an engineered outcome of action physics coupling under physical and symmetry constraints.