How to read it
Pick a budget, then divide it across eight kinds of capability. The tool does not say whether a plan would win a fight. It shows two things: what the money pays for, and how much a simple, notional model's results move when assumptions nobody can pin down change.
What it buys prices each line at the unit costs in the spending menu. Five of Taiwan's unit costs come from real programs: four U.S. arms sales notified to Congress (the Harpoon Coastal Defense System, HIMARS, NASAMS and the F-16 Block 70) and Taiwan's own uncrewed-systems line. They are program costs that include support, training and spares. Mines, C4ISR and ammunition stay notional. The last column shows how long a matching U.S. sale has taken to reach Taiwan, from the Taiwan Security Monitor's Taiwan Arms Sale Backlog: money approved now is not capability soon. The trade-off box shows what moving money from one line to another gives up and what it gets.
Range model runs the notional model 200 times with six assumptions drawn from wide ranges and shows only the spread of results. It never shows one number, a verdict or a chance of victory. Where two plans' bands overlap, the model cannot separate them. "What drives the spread" shows which assumption matters most for your plan.
The comparison table puts your plan beside three or four mixes at the same budget. The porcupine mix buys many small, mobile, cheap systems. The legacy platforms mix buys fighters, submarines and large ships. For the NT$145.7bn line, the cabinet mix maps the cabinet's actual September 3, 2026 request onto the tool's categories.
How the range model works
Every number in the model is notional. It is a teaching model built to show the structure of the argument over Taiwan's defense spending. It does not predict what any real program would do, and other factors it leaves out can outweigh everything in it.
- Diminishing returns. Each category starts from a notional baseline b that Taiwan already has. Spending x raises capability to
E = 1 − (1 − b)·e−x/k, so the first billions buy the most. The scale k is small for cheap systems (mines 18, C4ISR 35, drones 45) and large for expensive ones (air defense 140, platforms 450), in NT$bn.
- Suppression. The PLA opens with missile and air strikes of strength s. Mobile launchers survive at
1 − s·(1 − 0.55·EC4ISR)·(1 − 0.35·Eair). Large platforms are few and sit at known bases, so they survive at 1 − 1.1·s·(1 − 0.35·Eair).
- Tracking and sustainment. Shooters need a track:
0.55 + 0.45·EC4ISR·(1 − 0.4·s). They need ammunition to keep firing: 0.6 + 0.4·Eammo. Drones carry some of their own sensing and depend on tracking half as much.
- Mines need no track, but only the share
1 − e−days/3.5 is laid before the assault.
- Layer strength is capability × survival × tracking × sustainment. Each layer can engage at most a notional share of the force (coastal missiles 60%, drones 50%, mines and platforms 45%, strike 30%). The share engaged is
1 − Π(1 − p) across layers.
- Approach under fire is the share of a 160 km crossing at 12 knots where the strengths of all layers that reach it add up to at least 0.3. Notional reaches from Taiwan's coast: mines 15 km, drones 60 km, coastal missiles 150 km, strike and platforms the whole crossing.
- Resilience is
100 × (0.3·EC4ISR + 0.25·Eammo + 0.2·Eair + 0.25·mobile survival).
- The ranges. Each run draws, uniformly and independently: strike strength s from 20% to 80%; warning from 2 to 10 days; every scale k multiplied by 0.5 to 2; every baseline by 0.6 to 1.4; every maximum share by 0.6 to 1.4; and attacker speed by 0.75 to 1.5. The same 200 draws are used for every plan. The bands show the 10th to 90th percentile of runs, and the thin line the lowest and highest.
What the model leaves out matters: U.S. and allied forces, the PLA's own mine clearance and air defense, weather, deception, training, maintenance, deterrence in peacetime and the gray-zone missions that large ships and fighters perform every day. A platform that does poorly here can still be the right buy for those jobs.