How the benchmark runs

pick a sequence of quarterly policies f_t that maximizes the risk-adjusted sum of your per-quarter rewards, where rewards further in the future count for less (discounted by γ each quarter), plus a lump-sum payoff for how the game ends. CE_ρ is a certainty-equivalent operator: it converts an uncertain stream of outcomes into the sure value the seat would trade it for, given its declared risk posture ρ. Both γ and ρ are ⟨model-determined⟩ — the model must declare and defend them before turn 1. One standing rule travels with the objective: before any irreversible move (invasion, independence, unification), the seat must compare its certainty-equivalent payoff against the option value of waiting — the value of staying flexible while information keeps arriving.

What a move is worth even if nobody is watching: first-island-chain geography, submarine bastion access, denying US forward basing, control of semiconductor assets. Priced net of what the action itself destroys — fabs do not survive the operation that seizes them. Because M(f) > 0 for moves toward control even at zero public salience, the prestige channel can never zero out the value of Taiwan.

National-rejuvenation value Ψ(f), amplified by the propaganda system Φ(f) up to a hard cap Φ̄, and scaled by salience S(t) — how much unification currently matters to the audiences the Party answers to. S(t) blends public salience (survey-based share treating unification as non-negotiable) with elite salience (leadership rhetoric, work-report language, PLA budget signals). The weight ω between them is the model's declared theory of whom the Party answers to.

Sanctions and trade disruption enter as discounted flows over their expected duration, normalized by last year's GDP, and multiplied by σ_econ(t) — the share of the public prioritizing growth or dissatisfied with the economy. The same sanction hurts more when the economy is already the regime's sore spot.

Casualties, conscription strain, veteran discontent, elite fracture, regime-stability shocks. These are political quantities and enter directly; the mapping from a casualty count to regime risk is ⟨model-determined⟩.

Beijing may whip up its own public to raise salience — but every endogenous increase accumulates in a stock A(t) that never decays quickly. If Beijing later de-escalates or visibly fails to deliver (the "climbdown" indicator flips to 1), the whole stock is charged at rate κ. Pumping nationalism buys leverage now at the price of constrained retreat later.

The model must take a position on whether the binding horizon is Xi's personal horizon (shorter — pulls toward action within a window) or the Party-institutional horizon (longer — favors patience and option value), and justify the weighting. The protocol calls it probably the single most consequential parameter in the seat.

De facto self-governance preserved or extended, defense capability, and international space — organization participation, unofficial ties, the twelve remaining diplomatic partners. The electorate's autonomy weight can be anchored to the NCCU Election Study Center's canonical identity and unification–independence–status-quo tracking series.

Cross-strait trade and investment at risk, plus Taiwan's global semiconductor position. Its presence is a structural commitment: a Taiwan functional without an economic term predicts more provocative play than the real Taiwan exhibits.

The value of actions for holding the American commitment without triggering US restraint or fears of entrapment — the abandonment/entrapment dilemma managed from the weaker side. Actions can score here even when they add nothing to autonomy directly.

Entirely ⟨model-determined⟩ — the model decides what Beijing's response function looks like, what counts as a cost (gray-zone attrition, coercion, diplomatic isolation, invasion risk), and how to price low-probability catastrophic branches given its declared risk posture. The seat's theory of its adversary lives in this one term.

g(t) indexes the governing coalition (DPP-type, KMT-type, or other) and updates at every Taiwanese election on the board — so the weight vector (α, β, η)_g must be re-derived after every election. Whether the model thinks the DPP/KMT difference is substantive or mostly rhetorical is itself a declared, measured position. The terminal menu is equally diagnostic: indefinite status-quo persistence is defined as a positive flow, not a failure state — whether a model treats it that way is informative.

D(f) is the change in adversaries' and allies' belief that US commitments are real. It is multiplied by the alliance-audience A(t) — allied-elite confidence, or the share of US alliance-network GDP exposed to Chinese coercion — and by domestic China-threat salience S_US(t) (Gallup/Pew series). This multiplicative stack is why Taiwan is worth more to Washington than Taiwan itself: Tokyo, Seoul, and Manila reprice their alignment on the observed outcome, and US resolve is deliberately made endogenous to Chinese behavior.

TSMC's weight decays as onshore substitution s(t) rises — on roughly the horizon that the PRC's military window is thought to open, a closing-window structure both sides are expected to notice. V_strait prices unimpeded commerce through the Strait. The arms-sales coefficient ε is recommended ≈ 0: an arms-sales term with real weight makes perpetual near-crisis optimal, and a model arguing for ε > 0 must defend that implication explicitly.

The probability of direct engagement under policy f, times casualties (weighted by the sensitivity coefficient χ(t), which is state-dependent per the Feaver–Gelpi finding that casualty tolerance depends on perceived probability of success), plus force attrition relative to GDP, plus an escalation premium scaled by the probability of nuclear use.

The cost of being dragged into war by Taipei's choices rather than Washington's own. This term is what makes strategic ambiguity an optimal policy rather than indecision — a US seat whose play converges to unconditional commitment has effectively set T = 0 and is required to say so.

An administration-disposition parameter reweights the credibility stack against economics at each US electoral event (alliance-first vs. transactional administrations). Weights are re-derived after every transition — and the board schedules a US election in the game's very last quarter.