The Dead-Verb Test — Mechanism Viability and Whether KOEI Learns

A game mechanism (a verb on the command menu) is viable only if it has a non-empty home state — a region of the game’s state-space where it is the uniquely best response. NA1 fails this for most of its menu: Bribe, Ninja, Pact, and Marry are dead verbs, dominated by compounding development and redundant with the one policy that already wins. This page states the test, the anti-pattern, and the falsifiable forward question for the Game Annotation Series: does a later KOEI title give its alternative verbs a home state?

Links: NA1 — A Game-Design Crucible (the specimen this was decoded from), Civilization Revolution — The City-Builder That Plays as a Rush (specimen #2, cross-franchise: an entire designed identity reduced to dead verbs), The Dominance-Frontier Lens (dead verbs are dominated edges), Capability Without Leverage (the player-side mirror: a capability with no leverage), The Hollow Opponent (the AI-tuning side) Grounding: the cost/effect formulas, the war-commit gate, the turn-economy, and the attrition accounting are byte-exact in the na1-decompiler repo (na1sim/FINDINGS-defensibility.md), not inferred from play.

The test

A verb earns its slot on the menu only if there exists a game state where playing it beats every other verb. Absent such a state, it is dead weight — it widens the option list (and the player’s cognitive load) while never being correct. The viability question is therefore not “is this verb ever affordable?” but “is this verb ever optimal?”

Why NA1’s alternative verbs are dead

The binding constraint prices every verb against compounding. NA1’s real scarcity is action slots (4 decisions/year — the turn-economy thesis), not gold. So a verb’s true cost is the opportunity cost of a develop slot, and develop compounds geometrically. Bribe, Ninja, Pact, Marry are all one-shot, non-compounding effects. Against a compounding baseline, a non-compounding action is dominated unless it (a) averts a catastrophic loss or (b) unlocks a compounding gain. None of NA1’s reliably clear that bar:

• Pact / Marry — buy safety from attack. But the dominant policy is already don’t be a target: the war-commit gate selects the weakest-provisioned border fief, so non-weakness means you are never selected and the battle never happens — zero cost, no slot spent. Diplomacy is redundant with economics.
• Bribe (gold-for-defection) — the one with a faint pulse (a maxed-gold pump fief with nothing compounding left to buy). Still capped by conquest: an assault garrisons you with the surviving army anyway, so Bribe pays gold for strictly less than force delivers.
• Ninja / sabotage — a temporary enemy debuff against an AI whose spring-boon rubber-band repairs it next turn. Negative EV by construction.

The anti-player drift makes it worse over time. NA1 hands the AI economic handicaps (linear develop vs the player’s √-curve; spring-boon loyalty/wealth). As the AI’s gold-denominated power outpaces the player’s, every fixed-gold verb (Bribe, Pact) gets relatively more expensive each year — the verbs decay toward uselessness exactly as the game lengthens.

The anti-pattern: a verb with no home state

Pact is the clean specimen. Its cost and its value are anti-correlated in time:

• when you’re strong you can afford it but won’t be attacked (no need),
• when you’re weak you’re attacked but can’t spare the gold/slot (can’t pay),
• and where the two would finally align, non-weakness has already pre-empted the threat.

This is insurance you can only buy when you don’t need it, layered on redundancy with the optimal strategy. Empty home state → zero marginal value. A mechanism is only as alive as the set of states where it’s the best response is large.

Legitimate handicap vs. design miss

Compensating for weak-era AI with resource handicaps is legitimate — it’s an honest patch for a real ability gap (the AI can’t search, so give it economic slack). The miss is letting that handicap price the player’s alternative toolkit out of viability, collapsing the game to a single dominant line (develop + don’t-be-target) while 80% of the command menu stays dead. A good design uses its surface; this one builds surface it then prices out. The fix is orthogonal to the handicap: tune the cost/probability and compounding of the alternative verbs so each owns a home state, independent of how much slack the AI gets.

The falsifiable series question — does KOEI learn?

Stated as a per-title scorecard for the annotation series: does the dominated fraction of the command menu shrink across the catalog? Three concrete repair moves to watch for, title by title:

1. Make them compound — a pact that buys uninterrupted development, a vassal that pays recurring tribute: converts one-shot into compounding.
2. Tie them to a non-redundant objective — diplomacy required for a victory condition, not just for a safety the economy already provides.
3. Rebalance cost/probability so the home state is affordable in the weak state where the verb is needed.

If KOEI learns, later titles show fewer dead verbs and more states with a unique best response. That’s a measurable prediction to carry into the next ROM, so we test a hypothesis instead of rationalizing after the read.

Tags

games, strategy, game-theory, game-ai, koei

See also: KOEI AI & combat evolution — the combat-resolution + AI-fairness arc across five decompilers; the empirical answer to “does KOEI learn?” on the combat axis (yes: one-shot compare → cheap round-sim; combat handicap → symmetric).