Decision-making, under fire.

Choose a simulation. Then take your seat.

Each runs as a live, multi-screen session. Pick one to see its seats and open the view for yours.

Or a crisis built for your program.

Each simulation in the library started as a program's requirement: a region, a seat list, a calendar, and the specific failure modes a program needed its people to walk into. The engine carries no scenario knowledge of its own, so a new crisis is a new scenario file — the same live console, seats, projectors and scoring, pointed at a different problem. The three above run today. Everything below is a concept, and the one your syllabus needs may not be listed.

Tell me the region, the seats, the length and what you need your people to fail at, and I will build the scenario around it. Start with an email →

The crisis answers back.

A scripted exercise runs the same events in the same order, whatever the room does. Here the prepared injects are a floor. What happens next is drafted from what the room did — and nothing reaches the participants until the facilitator has read it, rewritten it, or thrown it out.

A session leaves evidence behind.

A page of the report. Illustrative figures — open it full size →

First the models. Then the crisis.

This is a course with a live crisis in the middle, and the sequence is the method: participants study the decision-making models first, then take their seats and apply them with the clock running. An AI adversary-analyst works beside the room the whole way; the author works the front of it. The reasoning is what gets graded.

Each course ends in a live crisis.

The teaching and the exercise are one product. A course establishes how decisions are actually made; the crisis at the end is where that gets tested under a clock, against an adversary cell, with the AI advisor in the room. Built and taught as graduate courses, delivered on site, compressed to the days your calendar allows.

• Decision-Making in Public Policy: National Security — the fifteen models, applied. Ends in Storm Across the Strait or Iron Corridor.
• Emerging Technologies and National Security — AI and autonomy, cyber, disinformation, space, quantum. Ends in Cyber Clash with China.
• Escalation Management and Deterrence — thresholds, signaling, the move that cannot be walked back. Ends in Iron Corridor or Storm Across the Strait.
• Alliance and Coalition Decision-Making — consultation as a constraint, consensus under a clock, what a partner will and will not do. Ends in Iron Corridor.
• Adversary Reasoning and Red Teaming — how the other side sees the same board, and what it wrongly believes about you. Ends in any crisis in the library, run from the Red Cell.

A course on a subject not listed is built to order on the same structure. So is a crisis: name the region, the seats and the authorities those seats hold.

Who spoke to the room.

Each of these practitioners guest-taught a session of the national security decision-making course, bringing careers spent in NATO command, the Senate, and the intelligence and counterterrorism agencies. Titles are the posts they held. What any of them has said about the simulation is their own view, published on their own account, and speaks for no institution they have served.

• Gen. John R. Allen, USMC (Ret.)Former Commander, NATO ISAF and U.S. Forces Afghanistan
• Amb. Douglas E. LuteLt. Gen. USA (Ret.) · former U.S. Ambassador to NATO · former Deputy National Security Advisor for Iraq and Afghanistan
• Lt. Gen. Karen Gibson, USA (Ret.)Former 41st Sergeant at Arms of the U.S. Senate · former Deputy Director of National Intelligence
• Lt. Gen. Michael K. Nagata (Ret.)Former Director of Strategic Operational Planning, National Counterterrorism Center
• Lt. Gen. Leonard J. Kosinski, USAF (Ret.)Former Director for Logistics, Joint Staff
• Katrina MulliganFormer Principal Deputy Assistant Secretary of Defense (SO/LIC)
• Sameer PunyaniSenior national security and technology executive
• Ali WyneAuthor, America's Great-Power Opportunity

If your network blocks linkedin.com, each card falls back to a direct link.

What the exercise is built on.

Two things: fifteen years of building and running these exercises, and the published work the assessment rests on.

• Fifteen years designing, running and reviewing crisis exercises — at the Pentagon, at the Joint Special Operations University, and in the U.S. Navy. Design work, facilitation, and the less comfortable job of reviewing other people’s exercises after the fact.
• Every mechanic here came out of that — how injects are timed, how much latitude the adversary cell gets, what the debrief measures, why reasoning is graded and outcomes are not. Each one is a correction to something that did not work in a real room.

Fifteen models, taught before anyone takes a seat. The exercise then asks which of them explains what the room actually did.

• Rational Actor — Allison 1971; Waltz 1979
• Organizational Process — Allison 1971; Halperin 1974
• Bureaucratic Politics — Allison & Zelikow 1999; George 1972
• Dual-Process — Kahneman 2011; Stanovich & West 2000
• Prospect Theory — Kahneman & Tversky 1979; McDermott 1998
• Bounded Rationality — Simon 1957
• Cognitive Bias — Jervis 1976; Heuer 1999
• Heuristics — Kahneman & Tversky 1974; Khong 1992
• Incrementalism — Lindblom 1959
• Poliheuristic Theory — Mintz & Geva 1997
• Cybernetic Theory — Steinbruner 1974
• Naturalistic Decision Making — Klein 2008
• Groupthink — Janis 1972
• Polythink — Mintz & Wayne 2016
• Applied Decision Analysis — Mintz 2005

• Decision quality — Eugene Bardach & Eric M. Patashnik, A Practical Guide for Policy Analysis: The Eightfold Path.
• Cognitive offloading and surrender — Steven D. Shaw & Gideon Nave, Thinking—Fast, Slow, and Artificial: How AI is Reshaping Human Reasoning and the Rise of Cognitive Surrender, Wharton School working paper, January 2026 — three preregistered experiments, 1,372 participants.
• Scenario figures — public estimates only (CRS, DoD posture reporting, published campaign analysis). Teaching models, not live intelligence.

The construct comes from a working paper. The findings the assessment relies on are published and refereed, and the last two are here because they cut against the easy version of the argument.

• Human–AI teams, measured — Michelle Vaccaro, Abdullah Almaatouq & Thomas Malone, Nature Human Behaviour, 2024. A meta-analysis of 106 experiments and 370 effect sizes: human–AI teams beat people working alone, and fell short of whichever of the two performed better alone. On decision tasks the combination was worse than its best member.
• Assistance and later performance — Hamsa Bastani, Osbert Bastani, Alp Sungu, Haosen Ge, Özge Kabakcı & Rei Mariman, PNAS, 2025. A class-randomized field experiment with nearly 1,000 students: unrestricted AI raised practice scores 48 percent, and the same students scored 17 percent below the control group on a later exam taken without it. A version built to withhold answers and require an attempt did no such damage.
• Automation and the loop — Onnasch, L., Wickens, C. D., Li, H. & Manzey, D., Human Factors, 2014; Endsley & Kiris, Human Factors, 1995. Higher automation improves routine performance while degrading situation awareness and the ability to resume control when the machine fails. The sharpest break comes when automation stops analyzing information and starts selecting the action.
• Automation bias — Parasuraman & Manzey, Human Factors, 2010; Goddard, Roudsari & Wyatt, JAMIA, 2012, a systematic review of 74 studies. Workload, task complexity, time pressure and misplaced trust raise the rate at which people follow a wrong recommendation.
• What reduces it — Buçinca, Z., Malaya, M. B. & Gajos, K. Z., Proceedings of the ACM on Human-Computer Interaction, 2021. Requiring a person to commit to their own judgment before the machine's answer appears cut overreliance where explanations alone did not. Participants rated the added friction less favorably.
• What the evidence does not establish — Pedersen, T. A., and colleagues, Endoscopy, 2026. Thirteen clinicians across 5,013 procedures: after the AI was withdrawn, unaided performance was statistically unchanged from baseline. Durable skill loss from AI use is not a settled finding, and nothing here claims it is.

What an exercise like this can establish, and what it cannot.

Every research method has a boundary, and a buyer is entitled to see this one drawn before the invoice. The framing below follows the standard RAND set out for its own wargaming in August 2026, and this exercise is held to it.

Standard applied: Zachary Burdette, Barry Wilson, David R. Frelinger, Hiwot Demelash & Julia Arnold, Exploring AI's Operational Implications in a Future U.S.-China Conflict: Initial Observations from the Camp(ai)gn Wargame, RAND Corporation, August 2026 — specifically its statement of what gaming cannot do.