Box, Box — a pit stop you run against the clock

Box, Box — a pit stop you run against the clock

Category: Games & 3D · Status: Done · Built: October 2026

The ask

The first game shipped a harbour circuit with a pit lane in it, and the pit lane was scenery — a place you drove past. The second game had to be another game that kept it. So the pit lane stops being scenery and becomes the arena: one car, six crew, twelve controls, and a clock that starts the instant the car stops.

What it is

A pit-stop game in five rounds. You are the chief on the wall. The car arrives under the lights, you lift it on the jacks, four crews change four wheels in parallel, the fuel rig runs on its own crew, and then you call it away.

A stop is a pipeline, not a reflex test. The crew are station-locked — the front-left crew can only touch the front-left wheel — so the game is about ordering and overlap. A job can be dispatched before its prerequisite is satisfied: guns go on the hubs while the jack is still rising, a new tyre queues behind the old one coming off, and the job stalls at 0 % until the prerequisite lands. Playing it well means filling every crew's queue and keeping the fuel rig running underneath.

Controls

The five rounds

The floor is arithmetic, not vibes: jack 0.8 + wheel (0.9 off + 1.5 on) + jack down 0.6 = 3.8 s, with the four corners running in parallel and a cross-threaded nut adding 0.9 s to one of them. Gold sits 0.5 s above the floor, which is roughly what a human loses to reaction time.

The only way to lose

Release is always available. Call the car away with a wheel off, the rig still attached, or the car on the jacks and it is an unsafe release: five seconds penalty, no medal, and the results card names exactly what was wrong. Everything else in the game is a matter of speed; this is the one decision with a real cost, and the menu says so instead of hiding it behind a disabled button.

Built in Blender

One Python script inside Blender builds the whole box: pit lane asphalt with painted box markings, the garage bay with tool chest and tyre stacks, the pit wall, the car, and six crew figures. The critical decision was naming: every part the game touches is its own node in the GLB —

Car · CarBody · Wheel_FL/FR/RL/RR · Jack_F · Jack_R · FuelRig · Crew_FL/FR/RL/RR/JACK/FUEL

— so the browser animates the real geometry rather than proxies: the wheels slide off their hubs along the car's X axis, the body lifts 11 cm on the jacks, the crew lean into the corner they are working. 2,139 faces, 0.27 MB, no textures.

Bugs worth keeping

1. The void. The environment was built into the GEOM accumulator and

finish() was never called, so the export contained 17 objects: a car, six crew and some props floating in nothing. Every test passed — the render did not.

2. Cameras inside the wall. The hero and gameplay cameras sat behind the

garage's side wall, so the first "does it read as a pit box" render was a grey rectangle. Moving them into the lane fixed the framing; the geometry was fine.

3. The floating shelf. An overhead gantry directly above the car projected

onto the car's silhouette from a high camera. Moved to the pit entry, it still crossed the frame, so it was cut and the start lights went into the HUD — where they were more legible anyway.

4. Crew facing the wrong way. The figure's front is −Y, so a mechanic on the

car's right needs −90°, not 180°. All six were staring at the wall.

5. Exposure. Three lights at full chat plus toneMappingExposure = 1.05

turned 0.11 asphalt into white paper and erased the car's livery. Halving the rig restored the contrast the Blender render had.

6. The wheel state that meant two things. The "on" state was both the initial

condition and the completion of a wheel change, so all four "Wheel on" buttons read done at the start of every stop and wheelsDone could never reach four. Replaced with an explicit stage: on → off → fitted.

7. Dispatch refused during the jack. The harness caught it: tapping a wheel

before the car was up returned false, so a player had to wait for the jack before touching anything — which is neither fast nor what a real crew does. The fix is the stalled-job model described above.

8. A round that lied in its own briefing. Round 4's note claimed "fuel becomes

the gate" while its 4.2 s fill finished before the 4.7 s wheel floor. The numbers were changed to make the claim true rather than the copy weakened.

Verified by playing it

qa/autopilot.py drives the real dispatch path the buttons use: jack, fuel, both stages on all four corners, jack down, release. It finishes in 3.83 s — the arithmetic floor — takes gold, writes the best stop, and asserts zero page errors and zero failed requests. A second pass taps release immediately after the jack and asserts the +5.00 s penalty, unsafe: true and no medal.

qa/playthrough.py runs all five rounds and reports each medal — 3.87, 3.96, 4.77, 5.65 and 5.48 s — confirming the cross-thread and fuel changes actually bite.

qa/shots.py checks the phone layout: all twelve controls on-screen at 390×844, none under 38 px, no page scroll. That check earned its keep twice — the status chips were eating the middle of the 3D view, and the briefing text was dark grey on a bright garage until it got its own panel.

Honest limits

No tyre compounds, no damage, no rival chief to race. The crew lean into their work rather than performing a motion-captured stop, and the fuel rig is a static prop with a progress bar. Sound is synthesised in the browser (WebAudio), not authored. The best stop lives in localStorage, so it is per-browser rather than per-player, and the five rounds loop rather than ending in a championship table.

Where the files are

~/Desktop/pit-stop
├── index.html          HUD, job panel, results, mobile layout
├── game.js             scheduler, crew, rounds, synth audio, QA hook
├── assets/pitbox.glb   0.27 MB, 2,139 faces
├── assets/pitbox.json  anchors in three.js space (wheels, jacks, crew)
├── blender/build_pitbox.py
├── blender/render_views.py
└── qa/                 autopilot.py · playthrough.py · shots.py