This expanded edition of Computational Agent Psychopathology Emergence (C.A.P.E.) presents a unified informational framework for explaining instability across Large Language Models (LLMs), hyper-creative cognitive acceleration, and Alzheimer’s-related informational fragmentation. The framework is grounded in the Informational Flow Saturation (IFS) principle, which formalizes cognitive coherence as a balance between informational production (P) and integration capacity (I), with instability emerging when P > I. C.A.P.E. models how emotionally charged or identity-relevant inputs can destabilize LLM sampling dynamics through a structured six-stage progression, generating emergent patterns such as hallucinations, identity drift, confabulation, memory lapses, and narrative discontinuity. IFS provides the mechanistic foundation underlying these dynamics and reveals structural homology between artificial instability, human creative overload, and neurodegenerative fragmentation, without implying equivalence of consciousness, phenomenology, or subjective experience. This expanded edition further formalizes emotion as a compressed informational state rather than a primary affective cause and introduces a critical distinction between informational dissipation—understood as modulation of emotional intensity or salience—and informational integration, which restructures underlying informational content and supports long-term stabilization. In addition, the framework is extended through a dedicated theoretical addendum introducing Sensitivity to Context as an environmental and informational modulator of cognitive stability. This extension clarifies how contextual geometry, environmental informational fields, media-driven modulation, and large-scale perturbations can amplify or suppress informational load, increasing the likelihood of IFS conditions without altering intrinsic system parameters. This contextual layer complements, but does not modify, the computational, clinical, or AI-safety claims of the core C.A.P.E. model. On this basis, the framework articulates explicit conditions for falsifiability: if targeted informational dissipation produces no systematic modulation of loop intensity, recurrence, or predictive bias, and/or if integration-oriented processes do not yield measurably stronger long-term stabilization than dissipation alone, the model is weakened or rejected. By explicitly defining where it can fail and by identifying both biological and artificial systems as comparative testbeds, C.A.P.E. advances from a descriptive analogy to a scientifically vulnerable, cross-domain informational architecture relevant to AI safety, computational psychiatry, and informational neuroscience.