Unit Rosters — a legible roster

What it is. The cast of pieces the player fields — the units of an RTS, the troops of an auto-battler, the summons of a tactics game — organized so every one has a readable job and earns its slot. A roster is the vocabulary that system-counter-systems connects and system-faction-asymmetry differentiates.

Player fantasy / why it’s fun. “I know exactly what each of these does, and I’m choosing the right mix.” A legible roster makes composition a pleasure — you read the enemy, pick your answer, and the units do what their silhouette promised.

When to use / when NOT

Use a designed roster when…	Skip when…
Player fields multiple unit types	One avatar / one active unit
Composition is a decision	Units are just reskinned HP bars
Counters/tech/factions need pieces	The “roster” is enemies only (`system-enemy-archetypes`)

The role alphabet

Most rosters are built from a small set of functional roles. Legibility comes from each unit reading as one role, not a blur of three.

Role	Job	Reads as
Frontline / tank	absorb, hold the line	big, slow, tough
DPS / line	deal the damage	mid, fragile-ish
Skirmisher	fast, harass, flank	small, quick, hits-and-runs
Artillery / siege	range, area, anti-clump	long range, slow, glass
Support	heal/buff/detect	non-combat, force-multiplier
Specialist	one job (anti-air, cloak, transport)	the answer to a specific threat

A roster is legible when a new player can name each unit’s role from its silhouette and speed alone — the pattern-readability test applied to the army.

Tiers vs. sidegrades

Structure	What it is	Watch for
Tiers	later units strictly stronger (T1→T3)	early units become dead weight — give them a lasting niche (cost, speed)
Sidegrades	same tier, different role, no strict winner	the healthiest for diversity
Tech gates	roster unlocks via `system-tech-tree`	pace the reveal; no unlock deserts (FUN.md §14 logic)

Prefer sidegrades within a tier and role differences across tiers. A pure linear tier ladder quietly kills half the roster.

Tuning levers

Lever	Does	Healthy range
Roster size	how many units	enough for a matrix, few enough to hold in the head (~6–12 per side)
Role coverage	each role represented	no gap the counter-matrix needs filled
Overlap	two units, similar job	intentional (a cheap and a premium tank) — not accidental
Cost curve	resource/time per unit	maps to power; no unit both cheapest and best
Skill expression	micro reward per unit	some reward control (skirmishers), some don’t (tanks)

Overlap vs. redundancy — the culling rule

Overlap is two units doing a similar job for a reason: a cheap disposable tank and an expensive durable one give the player a real cost/quality choice.
Redundancy is two units doing the same job with no distinguishing decision — one strictly dominates or they’re interchangeable. Cull it. Every unit must answer “when would I build this over its neighbor?” If it can’t, merge or cut.

The test: remove a unit. If no strategy got weaker, it was redundant.

How it wires to Hayao

Roles are proven by duels. Pure state + world.rng → clone the world, run the unit in its intended matchup, read the result (FUN.md law 7). Assert the tank tanks and the counter unit counters (FUN.md §9 NvN edges).
Mass command needs flow fields — one BFS per goal tile, cached outside world.state, gives the whole roster wall-aware pathing (FUN.md §9). Grep docs/API.md for astarGrid/floodFill before citing.
Export behavioural tolerances (arrival radius, aggro range) as API so tests share the sim’s constants (FUN.md §9) — the roster’s numbers live in one place.
Redundancy is a lint. The “remove-a-unit, did any strategy weaken?” check is a skill-delta test (FUN.md law 2) you can script.

Fails when…

Silhouette lies. A “tank” that dies fast, a “skirmisher” that’s slow — the read breaks and composition becomes memorization (pattern-readability).
Redundant units. Dead slots the player never builds; the roster looks bigger than it plays.
A dominant unit. One piece answers everything — the roster collapses to a spam (system-build-diversity and system-counter-systems both die).
Tier rot. Early units strictly obsolete by mid-game, so half the cast is a museum. Give each a lasting niche.

Verify

FUN.md §9 — every counter edge wins its NvN duel; commander bot wins; walled units route around; ms/step at peak.
FUN.md law 2 — the redundancy lint: removing a unit weakens some strategy; a mixed roster beats a mono spam.
Determinism: golden hash of a scripted battle across the roster.

Composes with

system-counter-systems — the matrix that connects the pieces into decisions.
system-faction-asymmetry — each faction is a distinct roster over shared roles.
system-enemy-archetypes — the enemy-side vocabulary; rosters and archetypes rhyme.
system-tech-tree — how the roster reveals over a match.