destroyulator/LANES/LANE7-cubicle-hell.md

LANE 7 — CUBICLE HELL

The game stopped being a smash sandbox and became a game about a man having a bad day.

The pitch

You are at work. The work is designed to provoke you. A meter fills. When it fills you snap, and for half a minute you physically cannot go back to your desk until you've broken enough things. Then you sit back down and try to earn the rest of your shift.

Falling Down, but the payslip is the scoreboard.

The score IS your payslip

        NET  =  WAGES (tasks completed)  −  DAMAGES (everything you broke)

The load-bearing detail: rage drains a flat amount per object destroyed, but each object bills at its own value. During a meltdown the optimal play is therefore to smash a lot of cheap rubbish very fast — cartons, paperwork, monitors — rather than put the sledgehammer through a filing cabinet. You don't choose whether you lose your temper. You choose how expensive losing your temper is.

FURY (objects broken while actually melting down) is tracked separately and is deliberately not money, so "I ended the shift owing them $2,400" stays a brag.

---

BUILT (this lane, shipped)

Piece	File	What it does
Tilt meter	game/scripts/Rage.gd	WORKING → MELTDOWN → COOLDOWN. Provocations fill it, clean work drains it slightly, ambient office drains your patience passively. During a meltdown the clock stalls if you stop destroying — you MUST break things.
The blood vessel	game/scripts/RageOverlay.gd	Red vignette that closes in with rage and jumps inward on every heartbeat; procedural veins that creep further into frame the angrier you get; a synthesized lub-dub whose BPM goes 62 → 172. Zero assets: shader built in code, veins are draw_polyline, heartbeat is a generated WAV.
The job	game/scripts/Tasks.gd	Stations you walk to and are locked in place at (you're at your desk — that's the joke). Two tasks so far, below.
The shift	game/scripts/CubicleHell.gd	5-minute shift, wages vs damages, meltdown count, end-of-shift payslip with a verdict line.

The two tasks that exist

• SPREADSHEET — enter a figure in a named cell. Partway through, the selection silently drifts one cell — no sound, no animation. Commit without noticing and the figure lands in the wrong cell, you eat a big provoke, and you start again. This one task is the whole thesis of the game in about forty lines.
• PHOTOCOPIER — run a copy job. It jams. Clearing it needs the right tray, and the trays are labelled 1, 2, 2A, 3, 3B because that's what head office decided.

---

TO BUILD — more ways to be provoked

Each is a Tasks.Kind + a panel_state() case. The framework takes them without changes; the design constraint is that each must have one specific, identifiable way of wasting your time, not just "it's slow".

Task	The provocation
Staple removal	A stack with N staples. Click each. Too slow and the page tears — redo the sheet. The remover slips and takes a chunk out of the page.
Guillotine / paper slicer	Align the sheet to a line, pull the handle. Misalign and the cut is crooked → redo. The alignment guide is 2mm off and always has been.
The phone	Rings mid-task. You must answer it. It's hold music. Not answering keeps it ringing and ringing raises rage faster than answering does.
Printer driver	The printer is "offline". Toggling it takes four clicks in a dialog that moves its buttons.
Reply-all	An email thread with 40 recipients. You must reply to the sender only. The default is Reply All.
Toner	Replace the cartridge. It's the wrong model number by one digit.
Timesheet	Enter hours. It won't accept decimals. It also won't accept fractions.

Passive ambient provocations (no interaction, just attrition)

Someone microwaves fish. The air-con drips. A colleague's ringtone. The fluorescent tube above your desk flickers at a frequency chosen to be maximally irritating. Each is a small periodic rage.provoke() — cheap to add, and they're what makes the room feel hostile rather than neutral.

---

TO BUILD — modes

1. CUBICLE HELL (built — the core). Shift timer, work, snap, wreck, repeat, payslip.
2. TOTAL DESTRUCTION — no job, no pay. Destroy 100% of the level. Scored on time.
3. AGAINST THE CLOCK — fixed timer, maximise damage value. The inverse of Cubicle Hell: here expensive things are the goal.
4. THE GAUNTLET (designed below — the most interesting one).
5. VERSUS — two players, one office. Explore later; the interesting version is asymmetric (one working, one wrecking) rather than both wrecking.

THE GAUNTLET (design)

"you need to destroy every file in a specific cabinet but which one, heaps cabinets, real subtle hints, time limit, and you have to actually pull each drawer out, empty files out, destroy them individually or in small lots unless you have a specific tool"

• A room of many identical filing cabinets. Exactly one holds the target files.
• Subtle hints only: a memo on a desk names a client; a labelled folder on the copier; a sticky note with half a reference number. No waypoint, no glow.
• Destruction is granular, not one-hit: pull a drawer → drawer slides out as its own body → files inside become individual smashables → destroy singly or in small lots.
• The perfect tool. Each level has ONE tool that makes its gauntlet tractable — an industrial shredder for files, a crowbar for the drawers, a lighter for a wastepaper bin. It is somewhere in the level, and the game does not tell you what it is. Working out "what would actually make this fast" IS the puzzle. With the wrong tool the gauntlet is possible but you will not beat the clock.
• Needs: Drawer.gd (pull-out sub-body with its own contents), a Container concept on Smashable, and a Tool flag on Weapon marking it as a level's key tool.

---

TO BUILD — the other three workplaces

Reference floor plans provided: Parks & Rec (municipal, partitioned, outdoor patio), Silicon Valley hacker house (domestic-open-plan, kitchen + pool), The IT Crowd (basement, two rooms, glorious junk), plus the generic open-plan bullpen.

Each is a new script alongside Office.gd implementing the same interface:

build()                    # shell, fittings, lighting
configure_environment(env)
desk_spots() / chair_spots() / copier_spot()
spawn_point() / spawn_yaw()

They differ in materials and hazards, which is what stops them being reskins:

Level	Signature material	Signature provocation
Municipal (Parks & Rec)	laminate, pinboard, civic beige	the public counter; forms in triplicate
Hacker house (Silicon Valley)	glass, whiteboard, drywall, a pool	standup meetings; someone's monitor wall
Basement IT (IT Crowd)	server steel, cardboard mountains, CRTs	"have you tried turning it off and on again"

On procedurally generating floor plans

The repo you found — doratracyer/floor_plan — won't help. It's OpenSCAD that turns hand-drawn SVG wall paths into 3D-printable STLs. It doesn't generate layouts, doesn't parse plan images, and produces no room polygons or graph you could build a level from. It's also GPL-3.0, which we would not want to link against.

Better path, and we're most of the way there already: Office.gd is a parametric floor-plan builder that just happens to have its numbers hard-coded. Lift those into a data spec —

{ rooms:  [{name, x0, x1, z0, z1, walls, door_at}],
  fittings: [{kind, pos, yaw}],
  lights:  {grid_x, grid_z},
  spawn:   {pos, yaw} }

— and a generic Floorplan.gd builds any of them. Then each workplace is a data file, new levels are authorable in minutes, and a generator becomes possible later (rooms as a BSP split of the footprint, doors on shared walls, fittings by room type) because the consumer is already data-driven. Do the spec first, generation second.

---

Asset needs (MODELBEAST / tools/gen_office_props.py)

Already generated procedurally: office chair, vending machine, microwave, potted plant. Needed next, in rough priority order:

filing cabinet with pullable drawers (split mesh per drawer) · desk phone · guillotine / paper slicer · stapler + staple remover · wall clock · whiteboard · server rack · pinboard · water cooler bottle (separate from base) · wastepaper bin · hole punch · shredder (the gauntlet's perfect tool)

Rule of thumb established this lane: if it's a box, build it in code; if its silhouette can't be faked by a box, generate it.