City / Colony Builder

What it is. A grid or map you fill with buildings whose value depends on what sits next to them. Placement is the whole verb: each tile scores from its neighbours, and a live “+N” under the cursor tells you exactly what a spot is worth before you commit. You optimize a growing layout against its own synergies.

Player fantasy / why it’s fun. “That corner is worth +7 — but it’d waste the river tile.” The pull is a legible optimization space where good placement is visibly rewarded and greed has a cost. Fun is the exposed score: the decision is on screen, quantified, and yours.

Pillars

The exposed score is the UI. One pure placementScore serves the sim, the bot, the tests, and the “+N” label under the cursor. The live number is not decoration — it’s the entire interface between intent and system.
Negative synergies create the only real decisions. If every placement is strictly additive, the greedy move is always right and there’s no game. Tradeoffs — this bonus costs that one — are what make a layout a puzzle.
The layout is a standing feedback loop. Each building reshapes the value of its neighbours, so the map’s optimum drifts as it grows. You’re steering a system, not filling a form.

The loop stack

Layer	The beat
Moment	Hover a tile → read the “+N” → place for the payoff (or the tradeoff).
Encounter	A wave/queue of buildings to place well before space or budget runs out.
Session	A map/island: raise the total score against its synergy constraints.
Meta	Unlocks, larger/varied maps, new building types with new adjacency rules.

Essential systems

System	Why this genre needs it
`system-economy`	Resources produced and consumed by buildings; the sinks that stop pure additive sprawl.
`system-resource-loops`	Extract→process→consume chains give buildings reasons to cluster and compete for space.
`pattern-feedback-loops`	Adjacency synergies are the loop; negative synergies are the damping that keeps it a decision.
`system-procgen-design`	Varied maps/queues as content; seeds give replay and a fairness sweep (queue always fits).
`system-onboarding`	Teach one adjacency rule at a time; the “+N” label is itself the tutorial.
`system-collectibles`	Milestones, cosmetics, layout goals — optional pull past hitting the target.

Content & difficulty model

One pure scorer, four consumers. placementScore must be the single source of truth for sim, bot, tests, and the on-screen “+N”. A score the label computes differently from the sim is a lie the audit must catch (score-honesty audit).
Prove the greedy bot faces tradeoffs. Assert greedy ≈ 2× random — skill separates from luck. If greedy never faces a negative synergy, neither does the player; the tradeoffs aren’t real.
Queue-always-fits sweep. Over N maps/seeds, assert the placement queue can always be legally placed — the builder’s connectivity/fairness proof (a solvability guarantee like the match-3 board’s).
Difficulty = tighter space / sharper tradeoffs. Harder maps give less room or stronger negative synergies, not just bigger targets. The optimization gets meaner, the score honest.
Content is building vocabulary. Each new building type should add an adjacency relationship (a new synergy or anti-synergy), not just a new sprite.

Signature-mechanic seeds

City builder but it’s a colony of people with needs and moods — placement serves a simulated populace and an AI director authors drama around it (mechanic-swap + structure toward anchor-rimworld; the score becomes lives).
City builder but the layout is a factory graph — buildings are machines, adjacency is throughput, the bottleneck is the puzzle (mechanic-swap toward anchor-factorio).
City builder but you place from a drafted hand — each turn deals building options; the bend fuses genre-deckbuilder draft-of-N onto placement (structure).
City builder but every tile is one screen and permanent — a hard spatial + no-undo constraint makes each “+N” a commitment (constraint).
City builder but it’s a garden, not a metropolis — tonal bend toward genre-farming-sim calm; adjacency is companion planting, growth is the score.

Common pitfalls

Two scores. A label that computes value differently from the sim breaks trust invisibly; enforce one pure placementScore.
All-additive synergies. With no tradeoffs the greedy move always wins and the game is a formality. Negative synergies are non-optional.
A queue that can’t fit. An unplaceable building is an unfair softlock; the queue-fits sweep exists to forbid it.
Score with no meaning. A number that doesn’t gate anything is wallpaper; tie it to progression or a target the player is chasing.
Building bloat. New types that don’t add a relationship dilute the read; every one should change an adjacency decision.

Anchors

anchor-factorio — adjacency-as-throughput, the layout as the toy, complexity scaling with the build; the factory-graph seed’s reference.
anchor-rimworld — the colony sim as a story generator, an AI director over the placement; the reference for the people-with-needs bend.

Verify

N-island queue-always-fits sweep; greedy ≈ 2× random skill delta; scoring-honesty audit; golden → design/FUN.md §17 · City/colony builder. Design the synergy space and the exposed score here; prove the delta and the audit there.

Composes with

system-economy — the sinks and resources that keep placement from pure sprawl.
system-resource-loops — the extract→process→consume chains that make buildings compete for space.
pattern-feedback-loops — the adjacency loop and the negative-synergy damping.