Bracelet

The story

The inherited wrist logic broke the moment a wrist rotated, and a wrist rotates constantly. Instead of patching around it, this engine ran as research first: understand the problem, prove an approach, write down exactly where it will break, and only then build. That pass is complete. The bracelet is the one engine that is ready to build rather than built, and that status is deliberate.

The approach, proven

Wrist rotation yields to the same measurement that carries the ring engine: the palm-plane orientation signal, distance-proof by construction, read at the wrist. The proof of concept draws a bangle at the wrist with live orientation feedback, edge-on detection, a handedness debounce, and tuning sliders that update in real time. The placement approach is demonstrated, not hypothesized.

Five failure modes, written down

The research pass's real product is knowing exactly where the naive version breaks:

1. Placement bias. The bangle sits low and off to the side without a wrist-width model.
2. Fist-tilt dropout. Landmarks degrade when the fist tilts toward the camera.
3. Palm-side noise. Palm-up wrists read noisier than palm-down.
4. Handedness flips. Left-right confusion at certain tilt angles puts the bangle on the wrong side.
5. Camera lifecycle race. A play() interruption on re-entry, known and fixable.

Why hold

The other three engines set the bar: measured stability, live testing before anything ships, honest accounting of what remains. Shipping a bracelet below that bar just to claim four for four would be the wrong trade. The remaining work is a build, not a mystery. The approach, the asset pipeline, and the failure modes are all documented. What is left is a go decision.