# Linked weight — motion lab `/lab/chain-links` has a centre-hung oval and a shallow photo arc attached to both chain runs. The arc is cut directly from the three supplied product photographs. The cropped image contains the metal and gem-set ends, without the photographed chain or cast shadow. **Pendant shape.** Photo arc is the initial view with 96 round rings per side, 3D tube rendering and the supplied rose gold cutout. The silver and gold controls select cutouts from their matching supplied photographs. Centre oval remains available; switching pendant style preserves the selected link count, ring shape and finish. The link-count control ranges from 12 to 96 per side, or 24 to 192 total. The controls link to the [Tiffany T Smile pendant](https://www.tiffany.com/jewelry/necklaces-pendants/tiffany-t-18k-yellow-gold-diamond-necklaces-pendants-1364184611.html) in a new tab. **Motion.** Two fixed top anchors suspend the chain. Each link is a mass-point distance constraint. The oval hangs from a shared hub. The bar connects distinct left and right chain endpoints to a heavier centre point with three fixed-length constraints, forming a rigid shallow triangle. Position-based Verlet integration advances gravity and damped velocity at 120 Hz. Bidirectional mass-weighted projection keeps chain links between 65% and 100% of their extended length, so they can loosen in flight. At high counts, the solver adds passes until relative stretch is below 2.5% or a 192-pass ceiling. Pinned anchors and a held point have zero inverse mass. The centre pendant point has inverse mass `1 / pendantWeight`. Inputs, velocity and step size are bounded. Drag a link, the oval or any part of the bar; release to swing. A short tap on the pendant cycles Silver → Gold → Rose gold. A swipe on empty canvas space up removes links; down adds them; left or right switches pendant style. The link count and chain redraw at each integer step while the finger moves, with the per-side count and change shown below the switches. A desktop wheel over empty space also changes link count immediately. Two compact buttons below the pendant switch pendant style and link shape; the settings tray keeps the 3D/2D view, finish, push, reset and physics controls. A drag that leaves the viewport releases after 900 ms without further pointer input, so a dropped pointer event cannot leave the pendant held. On a touch device, **Enable phone tilt** asks for device-orientation access from a button press. Screen orientation is accounted for, so tilting toward the right side drives the chain right; forward/back tilt alters the downward pull. The force is smoothed and bounded. If permission is denied or no valid sensor reading arrives, dragging, swiping, buttons and keyboard controls still work. No sensor is requested automatically. Nudge adds a bounded horizontal impulse. Weight defaults to 30×, gravity to 3.4× and air resistance to Medium; weight ranges 1×–46× and gravity reaches 6×. The page reports the largest relative chain extension. Rendering pauses when settled or hidden and resumes on input. Reduced motion weakens Nudge. **Appearance.** The 3D chain uses one instanced mesh for all closed tube rings. Round rings use torus geometry; Long links restore the elongated loop geometry. Alternating planes and depth testing show one ring passing through another. The arc is a source-photo cutout on the Canvas foreground in both view modes. Its photographed joins follow the two physical chain endpoints, and its low point follows the centre mass. It rotates and scales with that rigid bar, and it does not switch designs after leaving the screen. The oval is drawn with a metallic Canvas gradient. Soft moving shadows follow the rings and pendant. On phones, Controls opens a scrollable bottom tray capped at 28% of the small viewport height. The compact appearance buttons stay below the pendant while the tray is closed; opening the tray hides them to prevent overlapping hit areas. This is a 2D articulated motion study with 3D chain rendering and a 2D photo arc, not a full 3D jewelry simulator. There is no metal contact solver, friction between rings, axial twist or self-collision. Screen pixels are model coordinates, not calibrated metres. The three source files are `74957013_PDT_MAIN1X1.webp` (gold, SHA-256 `be145a500e8e7f3857df724940a33282d652dc55378aab6f69e3bdda3c681240`), `74957064_PDT_MAIN1X1.webp` (silver, `15652a2d690806b7a5c7fac64af14d28880c91816461cf0e86f72f7f829b8b87`) and `74956963_PDT_MAIN1X1.webp` (rose gold, `82b021b38f2957429612e38b15f91f181c88991b0f1f86850a6ebe5d288c3d71`). Each is 1256 × 1256. `scripts/physics-lab/crop-chain-bar.sh` verifies those hashes, takes the same 456 × 132 crop at source coordinate (400, 958), and applies an arc/end-cap alpha mask. The original photographs are not included in the lab package; only the three transparent arc cutouts are. Implementation: `physics-lab/motion/v1/chain-links.ts`, `physics-lab/experiments/chain-links.ts`, `physics-lab/chain-links/metal.ts`, `physics-lab/chain-links/render.ts`, `physics-lab/chain-links/main.ts`. Build and package to `.private/physics-lab` and `.private/physics-release`; the public lab Worker serves the page, assets and this document. Verify source hashes, numerical constraints, the 96-per-side setting, offscreen return without changing the arc, desktop/phone input, anonymous access and release separation. ## Performance sample `bun scripts/physics-lab/bench-chain-links.ts` measures requestAnimationFrame cadence while the photo arc is held, with round rings, Rose gold finish and shadow enabled. Weight was 30×, gravity 3.4× and air resistance Medium. Headless Chromium 156 on this Mac ran at 390 × 844 CSS pixels, device pixel ratio 2 (WebGL capped at 1.5), with two 2.2-second samples per view. Values below are rounded means. The 4× setting throttles CPU only; it does not simulate a phone GPU, memory pressure or thermals. Full run data: `evidence/physics-lab/chain-links-performance.json`. | Total rings | CPU setting | 3D tubes | 2D loops | | --- | --- | ---: | ---: | | 104 | normal | 50 fps | 60 fps | | 104 | 4× slower | 34 fps | 30 fps | | 192 | normal | 29 fps | 60 fps | | 192 | 4× slower | 17 fps | 19 fps |