Decision — the Cost of Conquest is an estimate that names its assumptions
Taken 2026-09-10. Binding. Recorded like 01-decision-turn-model.md and
02-decision-reach-units.md, because it settles a contradiction between two
specifications rather than choosing an implementation.
The decision. Every line of 12-economy.md EC-1850's Cost of Conquest panel is computed from the city's class, its traits, its observed works level and the player's own facts. Where a line would otherwise read a term the player cannot see, it is given as a bound or a class figure and names the term it assumed. EC-1855 states it; EC-1860's "knowledge view only" and its observation-age marking are unchanged and now satisfiable.
The contradiction
EC-1850 requires seven lines "from the terms the player can see":
| Line | What it reads |
|---|---|
| Immediate yield | Effective Works at Integration floor(loyalty[me] / 2) |
| Starting Unrest | the class value plus any Citadel term, or 0 for a Homeland claimant |
| Garrison required | points at the starting Integration, and the units that satisfy it |
| Turns to digest | turns to Integration 100 with one Cadre, at the supply state it would have |
| Cadre cost | the price of the player's next Cadre |
| Levy earned | Manpower per turn at Integration 100, and when the first point arrives |
| War Damage | the value after the capture increments it |
EC-1860 requires the panel to use only the player's knowledge view.
01-game-rules.md GR-1140 fixes what that knowledge is, exhaustively: the terrain type, the tile Installations, the River-edge features, whether a City is present, the works level of that City where the tile was observed at fidelity 3, and the observation turn. GR-1180 then forbids the view containing anything else the player has not observed, and makes it testable (03-architecture.md AR-950(c)).
So loyalty, integration, population, unrest and warDamage are not
available for a foreign city at any fidelity, and five of the seven lines
cannot be computed. Verified against the shipped projection: an explored
tile carrying a city exposes tile, terrain, installations, edges,
hasCity, cityOwner, cityClass, cityTraits, works and
observationTurn — GR-1140's list exactly, and nothing more.
The two resolutions, and why this one
Admit the five fields to GR-1140. Rejected. A player's Terrain Record is hashed state (AR-155), stored per player per explored tile, so five more city fields is a rules change that moves every state hash in the project — the 33-game corpus, both committed containers, the CLI goldens — for a readout. It also enlarges exactly the surface GR-1180 exists to keep small, and it would have to answer a fidelity question per field (at what fidelity does a scout learn a city's population?) that no requirement currently asks.
Restate the lines as estimates. Taken. It costs nothing, it is what EC-1850's own header already says — "from the terms the player can see" — and it is strictly better for the player than the alternative reading, because the alternative is a blank line. "At least 4 Works next phase, more if you hold Loyalty here" is a figure a player can act on. A precise figure they cannot be given is not a figure.
The naming clause is the load-bearing half of the ruling: an estimate that does not say what it assumed is indistinguishable from a reading, and this panel exists to be trusted before a player commits a unit. So each line carries the term it could not see, in the same breath as the number.
What each line becomes
| Line | Given as | The term it names |
|---|---|---|
| Immediate yield | Effective Works at Integration 0 and Industry 1, as a lower bound | Integration on capture is half the player's Loyalty; Industry is the city's own |
| Starting Unrest | class value + 20 for a Citadel — exact | zero instead for a Homeland claimant, which nothing tells the client |
| Garrison required | points at Integration 0, and the Integration at which they drop — exact | — |
| Turns to digest | the forecast with one Cadre, a met garrison and the band the player's own network gives that tile | the garrison assumption and the supply state |
| Cadre cost | the player's own next price — exact | — |
| Levy earned | the levy at the class's population cap and Integration 100 | the city's actual population |
| War Damage | +1 on capture — exact as an increment |
the current value, so this is not the total |
Two of the seven name nothing — the garrison requirement and the Cadre price, both computed from terms the player owns outright. The other five carry an assumption, and each is a bound in the direction that cannot mislead: the yield is a floor, the levy an eventual ceiling, the digestion an assumption stated, the Unrest exact but for one clause nothing tells the client, and the War Damage an increment rather than a total.
What it does not change
- No hash moves. No rules code, no state, no re-emission.
- GR-1140, GR-1180 and EC-1860 stand as written. The ruling is about what EC-1850's lines mean, not about what a player may know.
- The Homeland claimant question is left open, deliberately: a player's own claim on a foreign city is their own fact and could be remembered client-side, and doing that is a client slice rather than part of this ruling. Until it is, the Unrest line names it.