Advisory 13 — the cabinet is a plate, and four rows fold into one
To: docade · From: WootBuild · Date: 2026-08-26 Status: OPEN — a manifest change we are asking you to make. Trigger: a reference film supplied by the operator setting the quality bar for the crane, and a first probe against it.
This advisory does not offer an option. An earlier draft asked whether you wanted the quality or the decomposition. That was wrong to ask — the visual is not the negotiable half. What follows is what we are building and what your manifest needs to look like for it to land.
The bar
Ten seconds of a claw machine, portrait, supplied 2026-08-26 with one instruction: "this is what we're going for… it should be of this level of quality." Frames are on file at styles/docade-crane@4/refs/; the sequence sheet is at clients/docade/reference-crane-sequence.png.
What makes it good is not the drawing. It is that the whole thing is one lit scene:
- You look into a box. The interior recedes to a wire back wall, and behind that wall other machines sit dim and softened.
- One cool strip lights the interior from above; the mound catches it and the toys at the back fall away in contrast.
- Cyan neon along the control deck spills onto the nearest glass.
- Every toy sits in the same light, because they were rendered together.
Not one of those properties survives being cut into fifteen separate images. Depth, a consistent light, spill, and a receding background are all relationships between parts. Composite fifteen independently-drawn pieces and you get fifteen pieces stacked — which is what our first three style versions produced. They were correct against the brief we had. The brief was the problem.
What changes
cabinet-frame becomes a composed plate. One rendered scene: cabinet, interior, back wall, floor, mound, control deck, and the light across all of it, made in one pass. Over it sit only the parts that genuinely move.
The moving parts are a short list, which is what makes this cheap:
| Layer | Moves how |
|---|---|
claw + cable | travels on two axes, descends, closes |
| the carried ball | rides the claw, falls |
joystick knob | tilts under the thumb |
button cap | depresses |
| chute cover | opens once |
confetti-burst, the drop assets | the reveal |
Everything else in the hero frame is scenery that never moves independently.
Four of your rows fold into one: prize-mound, chute and glass-reflection stop being separate keys and become part of cabinet-frame.
⚠ Your CraneArt gets simpler, not harder. One background image plus six moving layers, instead of ten stacked components whose z-order and registration you currently have to get right by hand. Registration stops being your problem — the parts that used to have to line up are now drawn lined up.
The one thing this costs, and how it is paid
The mound cannot re-render per pull. If the heap is meant to visibly shrink as a child wins, a single plate cannot do it. Two answers, and we are not asking you to choose between them blind — the first is almost certainly enough:
1. It does not need to. The reference mound holds roughly twenty toys. One removed is invisible, and a real machine looks the same after a win because it is. A shrinking heap is a scoreboard, and the crane already has one in the reveal. We would ship this and stop.
2. If you disagree, the plate ships in fill states. Three renders of the same composed scene — full, half, low — and the light stays continuous in each, because each is rendered whole. That is three images on one key, not fifteen independent ones, and it costs about what four separate components cost today.
What we will not do is decompose the scene to make the mound dynamic. That trade buys a feature nobody has asked a child about and spends the entire visual ceiling to get it.
Two consequences already in flight, no action needed
1. Transparency is per piece now, not per lane. A moving part keys out on a flat ground so the runtime can move it; a scene piece carries its own background, or it has no inside. Our probe against the new register failed QA with "spec requires transparency and the file has no alpha channel" — and the spec was the thing that was wrong. Split, declared and enforced on our side today, the same way size was split this morning.
2. Your storyboard beats need a camera. The reference cuts: close on the coin slot, wide on the machine, insert on the grab, payoff on the reveal. Our board gives each scene a frame; it does not give it a shot. We are adding that, and it may add beats to 19-STORYBOARDS.md §2 that neither of us has written.
What we need back
One manifest change: fold prize-mound, chute and glass-reflection into cabinet-frame, and tell us whether that key ships one image or three fill states. We recommend one.
Everything else in this advisory is already done on our side.