Skip to main content
Modeling · Simulation · AI-Accelerated Analysis

The Right Model.
At the Right Fidelity.
For the Right Decision.

I've spent more than twenty years building models that support real decisions within national security. Now I'm applying that same discipline to commercial problems where the stakes are just as real. The approach is always the same: start with the question, match the model to the decision, and stop building when you have enough fidelity to act. AI agents compress months of analysis into days. The methodology keeps the answers honest.

20+
Years Matching
Fidelity to Decisions
MS
Applied Mathematics
Math & Physics undergrad
7
Years Running
My Own Firm
100+
Analyses Delivered
Defense & Commercial

The Analysis Drives the Model.
Never the Other Way Around.

Most modeling projects skip the hardest step: figuring out what to model, at what level of detail, for what decision. It doesn't matter if the model is a physics simulation, a statistical forecast, or a neural network. Let the question set the fidelity bar. Build to that bar. Validate what matters. Skip this and you end up with models that are technically impressive but don't help anyone decide.

"From the beginning, it's important to have a clear understanding of what goal or decision will be informed as this will drive subsequent choices throughout the process. [...] These decisions guide the specific questions for the activity as well as the degree of fidelity and level of analytic rigor needed from the results, findings, and conclusions."

Mission Engineering Guide 2.0, OUSD(R&E)
Step 01

Let the Question Set the Bar

Every question has a natural fidelity. “Is this system safe?” demands different precision than “which design is cheapest?” The question tells you what kind of model you need and how precise the answer has to be. That's the fidelity bar. Everything else follows from it.

Step 02

Build to the Bar — Not Past It

Start simple. Spiral fidelity upward only where sensitivity analysis shows it matters. A screening model answers basic questions; add detail selectively to the components that actually move your decision. Every layer beyond what the question requires is cost without value. The discipline is knowing where to stop.

Step 03

Validate What Matters

The model doesn't need to be perfect. It needs to be good enough for the decision it serves. Sensitivity analysis tells you which inputs actually move the answer. Uncertainty quantification tells you how much to trust it. More precise than the decision requires? You've overbuilt.

THE QUESTION What decision needs support? FIDELITY BAR How precise must the answer be? MODEL + METHOD Type, detail level, and computational approach DECISION-GRADE ANSWER Trusted, calibrated, defensible VALIDATE & ITERATE
Read the full ADM methodology →

AI Agents Do the Heavy Lifting.
ADM Keeps It Honest.

Most modelers I know treat AI as either a threat or a distraction. I think it's the best tool our field has gotten in decades. AI coding agents can run thousands of sensitivity sweeps, explore design spaces, stress-test scenarios, and validate across conditions — work that used to take a team of three several weeks. I bring the judgment. The agents bring the throughput. But without methodology they just generate noise. ADM keeps them pointed at the decision.
The Methodology

ADM Sets the Target

The question defines what to model and how precisely. That fidelity bar becomes the objective for every agent. No wasted computation, no gold-plating.

The Agents

Agents Do the Grinding

Coordinated AI agents run Monte Carlo sweeps, explore parameter spaces, build models, generate visualizations — in hours instead of weeks. One analyst orchestrating agents covers ground that used to take a team.

The Result

Deeper Analysis, Faster

Studies that used to take months, built in days. Multiple models, thousands of Monte Carlo runs, real data, real validation. Full analyses with actual findings — not slide decks.

Aiming This at What Matters

The methodology works on any domain. I'm proving it on the ones I care about most. Each study starts with a real decision, uses real data, and is honest about what works and what doesn't.

Energy

Virginia Grid — can PJM handle AI data centers?
Interactive vulnerability surface showing gas curtailment vs data center load heatmap with LOLE contour line and crosshair at Elliott replay position
PJM · Virginia 4 Questions · 200 Stochastic Draws · Validated

Can PJM Handle Virginia's Data Centers?

67 million people, 13 states, interconnected grid. Capacity market prices went from $29 to $333/MW-day. Four questions chained in sequence. The Monte Carlo changed the investment answer by $9B. Interruptibility is at least 5x cheaper than building gas plants. Every finding validated against PJM auction results and FERC/NERC reports.

Live — 4 investigations, 2 interactive tools, validated
Heatmap of Dominion hours unserved vs data center load and transmission capacity — shows local grid breaks at 1–2 GW while system-wide capacity is 19 GW
PJM · Dominion · Spatial 2-Zone Dispatch · MC Sensitivity

Beyond 19 GW: Where Data Centers Actually Matter

PJM can absorb 19 GW system-wide — but Northern Virginia breaks at 1-2 GW. Transmission, fleet composition, and demand concentration create a 10x local multiplier.

Live — 4 investigations, 2 interactive tools
Texas — grid resilience and infrastructure planning
Cascade iteration trace and phase diagram from the Texas energy grid study
ERCOT · Texas 8 Investigations · Validated

Can Texas Keep the Lights On?

27 million people, isolated grid, no backup from neighbors. Same question, radically different structure. The answer depends on whether you model the grid as a whole or where the load actually lands.

Live — 8 investigations, 2 interactive tools
EV charging network planner showing 88.5% coverage, 49 stations, and three-model cost comparison
Texas · EV Infrastructure 7 Investigations · Grid-Aware

Where Should Texas Build EV Chargers?

472,572 registered EVs, 3,976 existing stations, $408M in NEVI funding. Seven questions from where to build to what the grid can handle. Accounting for grid capacity saves $68M by right-sizing stations for what substations can actually deliver — because 30 of the top 49 locations can't get full power.

Live — 7 investigations, 4 interactive tools

Six domains, ten studies. Same methodology. More in development in the Lab.

Browse all studies →

Twenty Years of Getting the Model Right.

I've been building the kinds of models that defense organizations stake real decisions on for over twenty years. Missile defense, multi-domain operations, architecture-level trade studies. Every one built from physics, validated against data, tuned to what the decision actually required.

At MITRE, I led analysis teams working problems that cut across the entire Department of Defense. Before that, I ran my own defense consulting firm for seven years. Before any of that, I was a math teacher. That lesson from the classroom stuck: it doesn't matter how right your model is if the person across the table can't act on it.

The fidelity of the model determines the quality of the decision. I spent twenty years learning that in defense. Now I'm applying it to the problems that matter most to me.

Most modelers I know resist AI. I've leaned into it — not as a replacement for judgment, but as a way to do more analysis than one person should be able to. The methodology keeps it honest. The agents make it fast. I haven't met many people doing both.

Long-term, I want to point these methods at the biggest problems I can find — energy systems, environmental contamination, human health, education. Frame the problem, build the right model, show what's possible. Read the full vision →

MS Applied Mathematics
Fluid Dynamics, Probability & Statistics
BA Physics · BS Mathematics
Mathematical Physics & Operations Research
Chief Analyst, MITRE CUIC
Cross-cutting national security analysis
7 Years as Business Owner
Independent defense consulting firm
Active Security Clearance
Scientific Computing
Python, MATLAB, simulation & optimization tools
Former Math Teacher
Middle school, community college, full-time tutoring
The Arc
Math Teacher Teledyne Brown Booz Allen Hamilton Own Firm (7 yrs) MITRE Aerospace Prime Right Fidelity AI
Read the full background →

Got a Problem Worth Modeling?

If you have a hard problem and you think Analysis Driven Modeling can help — or you’re not sure yet — I’d like to hear about it.

michael@rightfidelity.ai  ·  Washington, D.C. Metro