Overview
CSIS, the House Select Committee on the CCP, and Pentagon-internal wargames simulate a China-Taiwan conflict using scenario matrices, historically-derived combat rules, and turn-based attrition modeling to generate figures like ship losses, sortie rates, and munitions expenditure. Those figures are outputs of stated assumptions, not predictions or measurements of a real war.
Brief
The number everyone quotes — the U.S. Navy loses 10 to 20 warships, or the LRASM inventory runs dry within three to seven days — is not a measurement of a future war. It is the output of a model built on a specific scenario matrix, run a specific number of times, using specific combat-resolution rules that someone chose. Before any of these figures gets cited in a hearing or a budget request, the analyst needs to know what scenario produced it, what the underlying rule assumed, and how many times the game was run.
CSIS's flagship 2023 study is the clearest public template because it is, deliberately, the only major unclassified version. As the report's authors state directly, prior Taiwan analyses were either "unclassified models focusing on one aspect of an invasion, seminar-type games that educate players but do not provide an adequate analytic foundation for policy recommendations, political-military games that primarily investigate diplomatic and political issues, or classified wargames whose assumptions and even results are not transparent to the public". CSIS built its wargame specifically to close that gap, and using historical data and operations research to model a Chinese amphibious invasion of Taiwan in 2026, positing a "base scenario" that incorporated the most likely values for key assumptions. That base scenario was run three times, then the team introduced excursions — deliberate changes to a single assumption, such as U.S. warning time or Japanese base access — to see how sensitive the outcome was to that one variable, across a total of 25-plus iterations using excursions to test assumptions.
The mechanics underneath the number are a hex-and-die combat resolution system, not a live simulation of missiles actually flying. Air and naval operations were played on a five-by-six-foot map covering the Western Pacific, ground operations on a separate map covering Taiwan, with a 70-page "rules for umpires" laying out the game rules, and die rolls, combat results tables, and computer programs calculating combat results — each turn representing three-and-a-half days. That combat-results-table architecture is a standard operations-research technique: instead of asking a human umpire to judge who wins an engagement, the model applies a probability distribution built from either historical analogues or theoretical weapons performance. CSIS's designers were explicit that this was the whole point of the exercise: the Chinese amphibious lift was based on analysis of Normandy, Okinawa, and the Falklands, other rules were based on theoretical weapons performance data such as determining the number of ballistic missiles required to cover an airport, and most rules combined the two methods — so the results of combat were determined by analytically based rules instead of by personal judgment.
This is where the meaning of a 'simulated munitions expenditure figure' has to be handled carefully. When a report says the U.S. exhausts its LRASM inventory within three to seven days, that is not a forecast of how a real war unfolds — it is the model's arithmetic output given (a) a stated starting inventory, (b) a stated per-sortie or per-engagement expenditure rate derived from the theoretical weapons-performance assumptions baked into the combat-results tables, and (c) a stated sortie/engagement tempo across the scenario's turns. Change any of those three inputs and the "days to exhaustion" figure changes with it — the figure measures the model's assumptions colliding with a stated stockpile size, not measured combat reality. A separate, later House Select Committee on the CCP exercise makes the same category of claim with a different scenario and a different level of transparency: a bipartisan simulation run by CSIS for the committee found that a conflict over Taiwan would deplete critical munitions in a matter of days and take nearly two years to replenish, with the committee's own results summary specifying that among the critical munitions most affected were LRASMs, Taiwan's Anti-Ship Cruise Missiles, and JASSM-ER. That committee simulation was set in 2026 using projected orders of battle — a different scenario construction than CSIS's original 24-iteration base case, even though both trace to the same modeling lineage and the same authors.
The Pentagon's own internal wargames are the closest thing to a black box in this ecosystem, and that opacity is itself a documented methodological fact, not an assumption. Pentagon wargames that influence policy and shape military strategy draw on intelligence assessments and non-public data, such as high-resolution imagery and classified information about the specifications of U.S., allied, and the adversary's equipment. CSIS's own report is candid about what that opacity costs public debate: those wargames are classified and restricted to keep sensitive data from potential adversaries, but the restrictions make it impossible for outsiders to judge why outcomes occurred, whether assumptions were reasonable, or whether alternative assumptions might produce different results — and many reported results appear self-serving since they support programs favored by the wargaming agency, while classified wargames also often focus on challenging cases, even if unlikely, to test the limits of the U.S. military. A defense specialist at a research institution focused on wargaming and crisis simulation has separately flagged a structural limitation that applies across both classified and unclassified versions: "it's very difficult to model public will and that really does affect how wars transpire." Any number extracted from a classified DoD wargame and repeated in open testimony should be read as a secondhand echo of an unverifiable process — useful as a directional signal, not as a citable statistic with a stated confidence interval.
Components (7)
The scenario matrix
Fixes the scope conditions before any combat is simulated — what year, what triggering action, what escalation ceiling, what political constraints (e.g., whether the mainland can be struck) apply to this particular run.
Historical-analogy rules
Translates real amphibious-warfare data into wargame movement and loss rates where no direct modern precedent exists, most visibly in modeling Chinese amphibious lift capacity.
Theoretical weapons-performance rules
Converts stated or estimated weapon specifications into combat-results-table probabilities, governing how many munitions or sorties are needed to achieve a given effect.
The combat-results table and die-roll/computer resolution engine
Executes the actual attrition math each turn, applying the historical and theoretical rules consistently so results trace to stated rules rather than umpire judgment.
The turn structure
Breaks the campaign into fixed time increments, with CSIS using a three-and-a-half-day turn, inside which sorties, resupply, and combat all occur before the next turn's conditions are recalculated.
The base scenario and its excursions
Establishes a most-likely-assumptions baseline, then varies one input at a time (warning time, base access, force posture) to test how sensitive the campaign outcome is to that single factor.
Player decision-making
Injects operational judgment inside the fixed rule structure — where to commit forces, when to escalate — which is why identical rules can still produce different iteration-to-iteration outcomes.
How It Works (8 steps)
1Define the scenario matrix and scope conditions
Analysts specify the year, the triggering Chinese action (invasion or blockade), the escalation ceiling, and any political constraints such as prohibitions on striking the Chinese mainland. This step exists before a single die is rolled.
CSIS project teamHouse Select Committee on the CCP staffPentagon wargaming directorates
Why this step: Without a fixed scope, attrition and munitions figures from different runs are not comparable — a blockade scenario and an amphibious-invasion scenario draw on entirely different rule sets.
2Build combat rules from analogues and weapons data
Designers translate historical amphibious operations and stated or estimated weapons performance into quantified rules — for example, how many ballistic missiles are needed to suppress an airfield — encoded into combat-results tables.
Operations-research analystssubject-matter experts on historical campaigns and weapons systems
Why this step: This step is what CSIS's designers point to as the difference between an analytically grounded wargame and a seminar-style discussion exercise: results trace to stated rules, not personal judgment.
3Set starting force levels and stockpiles
Each side begins the game with an explicitly stated inventory — ships, aircraft, ground units, and munitions counts such as the global LRASM stockpile — against which subsequent expenditure is measured.
Project intelligence and order-of-battle researchers
Why this step: A 'days to exhaustion' figure is meaningless without knowing the starting number it was drawn down from; changing this single input changes every downstream attrition and munitions statistic.
4Run the base scenario
Players execute the most-likely-assumptions version of the scenario across sequential turns, each turn representing a fixed multi-day period in which sorties are flown, engagements resolved via the combat-results tables, and losses tallied.
Senior government, think-tank, and military playersgame umpires applying the rules
Why this step: The base run establishes the reference case against which every excursion and every headline figure is subsequently compared.
5Resolve combat each turn via tables and dice or code
At each turn, engagements are resolved using the pre-built combat-results tables — via physical die rolls or computer calculation — applying the same rule set used in every other iteration of the game.
Umpiresthe combat-resolution software or tables
Why this step: Consistency here is what allows analysts to say the same set of rules applied to the first iteration and the last, making cross-iteration comparison valid.
6Run excursions varying single assumptions
The team reruns the scenario changing one variable at a time — for instance, U.S. warning time before the attack, or access to Japanese bases — while holding all other rules constant, to isolate that variable's effect on the outcome.
Project teamadditional player cohorts for each excursion
Why this step: Excursions convert a single game result into a sensitivity finding — the reason CSIS could identify surface-ship vulnerability and Japanese basing access as pivotal factors rather than one-off artifacts of a single run.
7Aggregate results across all iterations
Analysts pool the attrition counts, time-to-exhaustion figures, and campaign outcomes across every iteration and excursion, looking for patterns that hold consistently rather than one-off extremes.
CSIS Defense and Security Departmentcommittee staff compiling summary findings
Why this step: A single iteration's number (e.g., one game where LRASM ran out in three days) is not the headline figure; the headline figure is the range that recurs across the full run — which is why reporting on these games typically cites a range, not a point estimate.
8Translate aggregated patterns into public findings and recommendations
The recurring patterns — not any single game's raw numbers — are written up as findings (e.g., stockpile insufficiency, surface-ship vulnerability) and attached to policy recommendations for procurement, basing, or force posture.
Report authorscommittee membersPentagon leadership receiving classified out-briefs
Why this step: This is the step at which a modeling output becomes a citable policy claim — and the step at which the original scenario assumptions and iteration count most often get dropped from the retelling.
What Makes It Work
Combat-results-table resolution replacing umpire judgment
By pre-building probability tables from historical analogues and weapons-performance data, the designers ensured that identical inputs produce identical resolution logic across all iterations, which is what makes cross-run comparison and excursion analysis meaningful rather than arbitrary.Documented
The excursion method for isolating sensitivity
Varying exactly one assumption at a time while holding the rest of the rule set constant lets analysts attribute an outcome shift to that specific variable, rather than to unmeasured combinations of changes.Documented
Iteration count as a hedge against single-run noise
Because combat resolution has a stochastic (dice or randomized) element, running the game 24 to 25-plus times lets analysts distinguish a recurring pattern from a single unlucky or lucky roll sequence.Inferred
Classification as a constraint on public verifiability
Pentagon-internal wargames draw on non-public intelligence and classified weapons specifications, which improves realism for the government audience but removes the ability of outside analysts to check whether the assumptions driving a leaked figure were reasonable.Documented
Where It Breaks (5)
A single iteration's figure gets quoted as if it were the finding
Consequence: A headline number like 'LRASM runs out in three days' can come from the most extreme excursion in the set rather than the base-scenario median, misrepresenting the model's actual central tendency to a policy audience.
Safeguard: Reports that publish the full range across iterations (e.g., 'three to seven days') rather than a single figure allow readers to see the spread; readers should always ask which iteration or range a quoted number comes from.
Classified wargame results circulate without their assumptions
Consequence: A leaked or paraphrased figure from a Pentagon-internal game — such as a service leader's remark about combat outcomes — cannot be checked against the scenario, force levels, or rules that produced it, so it functions as testimony rather than verifiable analysis.
Safeguard: CSIS's unclassified model exists specifically as a counterweight, publishing its full rule set so outside analysts can interrogate the assumptions rather than accept the conclusion on authority.
Historical-analogy rules mis-transfer to a different operational context
Consequence: Basing Chinese amphibious lift rates on Normandy, Okinawa, and the Falklands imports assumptions about sea state, defended-beach density, and logistics tempo from mid-20th-century operations that may not hold for a 21st-century Taiwan Strait crossing under precision-strike conditions.
Safeguard: Combining historical analogy with theoretical weapons-performance data, as CSIS states it did for most rules, is intended to cross-check one method against the other rather than relying on either alone.
Modeling public and political will is treated as a solved input
Consequence: Wargames struggle to represent how domestic political tolerance for casualties or economic pain shapes escalation decisions, which is a documented limitation flagged by a wargaming specialist at a research institution, meaning the game's political-constraint rules may not reflect how leaders would actually behave under real pressure.
Safeguard: None — this is described as an unresolved structural limitation of campaign-level wargaming generally, not something the current generation of games has solved.
Classified games are selected toward worst-case stress-testing, not most-likely outcomes
Consequence: If a classified wargame is deliberately designed around a challenging case to find the limits of U.S. capability, a leaked result from that game will look far more alarming than a most-likely-case unclassified model, and audiences unaware of the design intent may treat the two as directly comparable.
Safeguard: CSIS explicitly flags this dynamic in its own methodology discussion, distinguishing its 'most likely values' base scenario from the worst-case orientation it attributes to some classified exercises.
Facts & Figures (4)
The claims behind this analysis, each with its verification status — including what is contested, unverified, or could not be established.
Combat-results-table resolution replacing umpire judgment — By pre-building probability tables from historical analogues and weapons-performance data, the designers ensured that identical inputs produce identical resolution logic across all iterations, which is what makes cross-run comparison and excursion analysis meaningful rather than arbitrary.
✓ DOCUMENTED
The excursion method for isolating sensitivity — Varying exactly one assumption at a time while holding the rest of the rule set constant lets analysts attribute an outcome shift to that specific variable, rather than to unmeasured combinations of changes.
✓ DOCUMENTED
Iteration count as a hedge against single-run noise — Because combat resolution has a stochastic (dice or randomized) element, running the game 24 to 25-plus times lets analysts distinguish a recurring pattern from a single unlucky or lucky roll sequence.
— INFERRED
Classification as a constraint on public verifiability — Pentagon-internal wargames draw on non-public intelligence and classified weapons specifications, which improves realism for the government audience but removes the ability of outside analysts to check whether the assumptions driving a leaked figure were reasonable.
✓ DOCUMENTED