Proven methods. Every scale. Twenty-five years.

The lineage begins at Simon's Rock College of Bard. One of the first things done there, at sixteen, was to model a single neuron in software, and to run into the fact that a cell's firing behavior emerges from the interaction of its channels and inputs, not from any one of them. That is the intellectual seed of the whole career in one image: the interesting behavior is in the coupling.

From that single cell the question moved up a level, to a system that attends. The honors thesis, Attentional Intensity in the Auditory Neurosystem, Spectral Space, carried IRB approval for human-subjects testing, two years of experimental work, and dedicated software across three interrelated trials. Attention is a system-level property, no neuron attends, and studying it meant learning to instrument a coupled system and read behavior that only exists at the aggregate. The pre-medical biology and cognitive-science training around it grounded the intuition in real neural mechanism, not metaphor.

Graduate work at the University of Connecticut carried the same lens into organizations, using network science and graph theory to ask how institutions fail under stress and what structure makes them robust. Sponsored by the Department of Defense and archived in the Homeland Security Digital Library, this is the direct ancestor of the robustness-gap and iatrogenics frameworks that anchor the practice today: the recognition that a system can look safe on paper and still be brittle where it actually bears load.

As a principal investigator and DoD civilian program leader in the MOVES Institute, the method became a platform. The Complex Adaptive System Simulation framework modeled whole societies as coupled agents, beliefs, values, and interests grounded in survey data, propagating through empirically-built social networks via discrete-event dynamics, and was validated to the DoD standard for models that inform operational decisions. This is the period that produced the published research record and the conviction that emergent social behavior is modelable in advance, not only explained in hindsight.

The move to Northrop Grumman supporting the Defense Manpower Data Center was not a departure from the research; it was its largest-scale engineering application. In a $60B+ program environment, the same principles, favor modularity over tight coupling, design for the unexpected interaction, treat the human system around the technical system as a first-class architectural variable, shaped enterprise systems like Global Force Management and DoD/VA health-record modernization. Complex-systems thinking, in production, serving the full US defense and health enterprise.

Then the method was put where it could be proven or falsified with money. CASS was extended to live US equity-index derivatives, targeting exactly the phenomenon that makes both markets and AI dangerous: cascading failure in a coupled system of autonomous execution and human decision under stress. Significant in-silico testing preceded any capital. From 2015 to 2022, seven years, simulation and live operation ran together, including through the March 2020 crash, a verified record of governing a complex adaptive system where the feedback was immediate and the theory had to be right before deployment. The full treatment, with verified performance, is on the Live Market Operations page.

This era overlaps the last one, and grew directly out of it. The work on cascading failure in coupled systems led straight into AI safety: by 2022 the same dynamics were the defining and more urgent risk in capable AI, and the work shifted fully. Trading stopped because this matters more.

Capable AI at scale is the same problem, restated at the level of machine intelligence. A powerful model, coupled to an organization under pressure, produces behavior that controlled testing never surfaced. The robustness gap is the financial-stability problem in a new medium; emergent misalignment is cascading failure with longer horizons and more diffuse consequence. The response is the method the whole career has been building: model the deployed system as a coupled network (the PAN / EMU simulation), make the dynamics explorable and testable (the PAN Lab and the Oversight game), and validate governance before it ships. The doctoral work at USC names the crux directly: that the decisive barrier to safe AI in high-stakes practice is a governance-architecture problem, not a model problem.