Twenty Years of Getting the Model Right.

A Career in Two Worlds

For over twenty years, I've built models that people use to make high-stakes decisions — not in a research lab optimizing benchmarks, but in the field, where the model has to be good enough for someone to stake resources, missions, and sometimes lives on the answer it produces. That constraint changes everything about how you think about modeling.

The formative chapter was at MITRE's Cross-Cutting Urgent Innovation Cell (CUIC), where I served as Chief Analyst. CUIC existed to solve problems that cut across the entire Department of Defense — the ones that didn't fit neatly into a single service or agency. Multi-domain effects chain analysis. Kill-chain modeling. Architecture-level trade studies that had to integrate sensors, platforms, networks, and decision timelines into a single coherent picture. Every one of these was a world model: a simplified but faithful representation of a complex system, built to answer a question that mattered.

That's where I developed Analysis Driven Modeling: the analysis question determines which model to build, at what fidelity, with which metrics. Not "here's a model, let's find a question for it." The question always comes first.

Before MITRE, I spent seven years running my own firm. That experience taught me things you can't learn inside a large organization — finding clients, winning contracts, managing subs, delivering results, and keeping the lights on all at the same time. I learned what it means to deliver for clients who can't afford to wait for a 200-page report that arrives six months late. When you are the firm, the model has to work, the briefing has to land, and the invoice has to be justified. That kind of accountability shapes how you approach every engagement.

Now the AI revolution is discovering what the modeling and simulation community learned decades ago: the hardest part of building a world model isn't the model itself. It's knowing how good the model needs to be for the decision it's supposed to support. The AI field is learning to build world models automatically from data, and it's learning (often painfully) that a model that's impressively complex but not validated against the right metrics is worse than useless — it's dangerous. It creates false confidence. The fidelity question isn't new. What's new is that a generation of AI practitioners is running into it for the first time.

There aren't many people who've spent twenty years validating physics-based simulations and also work with neural networks and reinforcement learning. That's where I ended up.

The defense M&S community has deep expertise in model validation, uncertainty quantification, and fidelity trade-offs, but most of its practitioners haven't crossed into neural networks, reinforcement learning, or physics-informed machine learning. On the other side, the AI community has extraordinary tools for learning from data — but most have never built a physics-based simulation or dealt with the V&V frameworks that the simulation world spent decades developing. Right Fidelity AI sits in the gap between those two worlds, and I think that gap is where some of the most important work is going to happen.

For larger engagements, I draw on a professional network of modeling, simulation, and analysis practitioners I've worked with across two decades in defense and aerospace — people I know, trust, and have delivered alongside. I also use generative AI to accelerate research, analysis, and development work that would traditionally require a larger team. I can bring in the right people when the project needs them, without carrying a payroll when it doesn't.