What it is. The interface is where the game reveals its state and takes your input — HUD, inventory, menus, tooltips. Not chrome bolted on after; a design surface you tune like any mechanic. Inventory is the special case everyone gets wrong: a container of items you sort, equip, and discard.

Player fantasy / why it’s fun. Two pulls. The loadout — “these are my tools, chosen by me, and I own the consequence.” And the glance — the screen tells me exactly what I need in the split-second I need it, and I trust it. When the UI disappears into the play, the player feels competent, not administrated.

When to use / when NOT

• Use an inventory when items are choices — carry this or that, equip A over B. The container is only interesting if it constrains.
• Use rich HUD when the player acts on hidden state under time pressure (health, cooldowns, threat).
• NOT when the “inventory” is just a bag with infinite room and no decision — that’s a database, and sorting it is a chore, not a game. Cut it or add a constraint.
• NOT more HUD to fix a readability failure. If the player can’t parse the play, fix the play’s silhouette first; see pattern-readability. Meters papering over an unreadable field is a smell.

Variants

Twist seams

• Inventory but limited slots force loadout choices (constraint) — the bag can’t hold everything, so what you leave behind is the decision. anchor-into-the-breach gives three mech slots, not thirty; every pick is a build. Pairs with system-build-diversity and pattern-risk-reward.
• UI but diegetic — the interface is in the world (theme) — no floating meters; health is the character’s posture, ammo is a hologram on the gun, the map is a paper the avatar unfolds. anchor-outer-wilds’ ship log, Dead Space’s spine bar. Trades instant legibility for immersion — measure the cost in pattern-readability.

Tuning levers

• Friction: intentional vs accidental. Intentional friction is a decision you want the player to feel (drop the loot or the medkit?). Accidental friction is a click tax with no choice inside it (open menu → scroll → confirm → close, ten times a run). Kill the second; sharpen the first.
• Information density. Show what the player acts on now; defer the rest to a pull-open panel. A HUD that shows everything shows nothing. See pattern-pacing-and-tension — the HUD should breathe with the play.
• Defaults. Auto-sort, auto-equip-best, auto-stack, quick-transfer. Good defaults erase 90% of the busywork while leaving the meaningful override one click away. The default is a design statement about what’s normal.
• Slot / grid / weight budget. The tightness of the constraint is the difficulty of the meta-game. Loosen it and choices evaporate; over-tighten and it becomes a sliding-block chore.
• Sort & filter affordances. One-tap sort, search, category tabs. Cheap to add, enormous friction relief on any list over ~12 items.
• Comparison surface. When equipping, show the delta (this vs. equipped) inline. The player should never do arithmetic in their head to make a choice.

How it wires to Hayao

• Menus and panels are DOM, not the sim. Inventory screens, settings, loadout — build them as screen chrome via the show-screen path, kept out of the world hash. The play field is the sim; the menu is the shell around it. (Invariant 4 in AGENTS.md.)
• In-world (diegetic) UI is view. A meter drawn inside the scene — a bar over a unit, a glowing gauge on the avatar — is a pure-view node marked cosmetic so it never enters world.hash(). State it reflects lives in the sim; the drawing is cosmetic.
• Item state is sim state. What’s in the bag, what’s equipped, counts and stacks — that’s authoritative game state and must be deterministic. Any shuffle or drop-order goes through the deterministic RNG, never Math.random.
• For the logic/view split that keeps interface out of the hash, examples/sokoban/ is the reference. For a single view mechanic in isolation (camera, particles), read the matching sandboxes/ lab rather than a whole game.

Fails when…

• Management outweighs play. If the player spends more attention arranging the bag than using its contents, the container has become the game — and a boring one. This is the inventory form of antipattern-decision-paralysis: too many trivial options, no clear default, every choice equally weightless.
• Currencies proliferate. Five overlapping resources, each with its own bar and shop, none clearly better — see antipattern-currency-spaghetti. One legible economy beats a wallet of confetti; cross-check system-economy.
• Fake loadout. Slots that look like a choice but where one option strictly dominates — that’s antipattern-fake-choice wearing an inventory costume. If auto-equip-best is always correct, you have a database, not a decision.
• Feature-soup HUD. Every subsystem earns a corner meter until the field is a cockpit; see antipattern-feature-soup. Restraint is a feature — pattern-restraint-and-negative-space.
• Diegetic taken too far. Immersion at the cost of the player not knowing they’re low on health. When the fiction hides load-bearing state, the input feels like a lie (antipattern-input-lie). Legibility wins ties.

Verify

Interface is a feel surface — prove it in [[design/JUICE.md]]. The bar: can a first-time player find and act on the state they need without a tutorial, and does opening/sorting the inventory cost near-zero friction? Judge the look from a headless render (Invariant 5), assert item/equip state via the sim’s probe/hash — never eyeball counts. If the HUD reads as debug scaffolding rather than shipped chrome, run the judge skill against design/JUDGE.md.