Nobunaga’s Ambition (NA1) — A Game-Design Crucible

Reverse-engineering NA1’s bytecode VM down to the actual formulas turned a 1980s strategy game into a ground-truth source for game-design theses: claims you can verify against the machine rather than infer from play. Several generalize well beyond NA1. This page is the index that ties the crucible together — the repo (the RE work), the vault theses it sparked, and the candidates still in the notes.

Links: Randomness as Termination (N≥3) (the broadest finding, promoted), The Dead-Verb Test (mechanism viability + the “does KOEI learn?” series question), The Multiplayer Coalition Problem, The Dominance-Frontier Lens (NA1’s per-fief combat maps are a worked instance), Game Annotation Series Repo: na1-decompiler (private) — extracted from this vault 2026-06-05; the decompiler, verified formulas, and emulator live there.

An index / crucible page (per Specimen & Thesis): NA1 is an external-repo specimen whose deep RE artifacts are repo-only, but it has fed in-vault theses that need a hub. This is that hub.

Why NA1 is a crucible, not just a project

Most game-design claims are inferred from play and stay arguable. NA1’s claims are decoded from the bytecode — the harvest formula, the combat terminator, the natural-death probability, the turn-order shuffle are read off the machine, byte-exact. That makes NA1 an unusually strong specimen: when a design principle holds here, it holds against ground truth, not against intuition. The generalizable ones are worth lifting out as portable theses; the NA1-specific mechanics stay in the repo.

Theses it has sparked

Promoted (in-vault):

• Randomness as the Termination Mechanism (N≥3) — the broadest finding: NA1’s six independent randomization sources are what let an N≥3 game terminate against the gang-up equilibrium. NA1 is that thesis’s canonical worked specimen.
• The Hollow Opponent — Where Game Depth Actually Lives — decoding the daimyo AI to ground truth (argmin-weakest + threshold, one-ply, no dominance frontier) exposed that NA1’s perceived depth lived in the player’s ignorance of the rule; once decoded, the depth migrated to the board (adjacency graph + turn-economy). The capstone of the AI epic, and the “you don’t meet your heroes” lesson generalized.
• The Dead-Verb Test — Mechanism Viability and Whether KOEI Learns — most of NA1’s command menu (Bribe/Ninja/Pact/Marry) is dead: one-shot verbs dominated by compounding develop and redundant with the don’t-be-a-target policy → no home state. Generalizes to a viability test (a verb needs a state where it’s the uniquely best response) and a falsifiable forward question for the series: does a later KOEI title give its verbs a home state? Decoded from the combat-attrition + war-commit-gate work.

Candidates still in the notes (repo + memory), worth promoting:

• Turn-economy — action slots are the binding constraint (4 decisions/year; the frontier is 4-D: terrain × count × quality × runway). A portable strategy-design thesis, currently memory-only.
• Binding-constraint determines analysis depth — a tight multi-axis constraint (NA1) yields per-position character; a loose single-axis one (Aerobiz) goes flat. A general game-analysis lens.
• Equilibrium & termination under deterministic play — the N=17 gang-up draw that motivates the randomization layer (the deterministic half of the n3 thesis).

These are flagged here so the crucible’s harvest is visible and the promotions don’t get lost in memory (the release-valve trap).

What stays in the repo (external-repo specimen)

The decompiler pipeline, the verified econ/combat formulas, the emulator, the per-fief tactical maps, and the label/data symbol tables are repo-only by design — they’re the RE artifacts, not portable theses. Per the two-altitude method, this page is their in-vault anchor; the deep work lives at na1-decompiler.

Tags

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