The fracture of glassy polymers involves complex interactions between molecular-scale mechanisms and macroscopic deformation, yet identifying microscopic features that are consistently associated with failure under different loading conditions remains challenging. In this work, we investigate fracture in glassy polymers using a coupled particle-continuum simulation method, which resolves molecular processes at crack-tips under nonuniform deformation implemented by continuum mechanics. Across a wide range of temperatures, molecular architectures, geometric constraints, and bond breakage criteria, the deformation process organizes into three distinct stages: local yielding, necking, and fibrillar instability. Local yielding is characterized by atomic-scale dilatation and isolated nonaffine rearrangements. As deformation proceeds, these rearrangements coalesce into a spatially connected plastic zone, marking the transition to necking. In this regime, deformation is sustained by the stretching of entangled chain segments that carry the applied load. Continued loading destabilizes this state and leads to fibrillar instability, triggered by highly localized disentanglement events. Despite differences in macroscopic response and failure strain, the later stages of deformation exhibit remarkable consistency across systems. Once plastic activity becomes spatially connected, the macroscopic behavior is no longer governed by collective deformation but by a small number of extreme local events. Correspondingly, molecular-scale structural quantities, such as bond length, display progressively narrowing statistical distributions that converge immediately prior to fracture. This convergence is observed irrespective of molecular architecture, geometric constraints, or bond breakage criteria. Together, these results indicate that fracture in glassy polymers is associated with the emergence of a characteristic molecular-scale state reached after the onset of spatially connected plasticity, providing a unified perspective on fracture processes across different loading conditions.