# Glass, with no glass.

Independent article case study. Domain: website consumer / Design Engineering presentation.

A rounded-rectangle signed distance drives a bevel height field and finite-difference normals in a native WebGL shader. Three refracted samples of an original procedural texture create dispersion. A bounded pool reprojects over a spherical field, and selectable HTML labels use the same projected card data. GPU cloth and the publisher’s full camera-matrix implementation are outside this study.

## Attribution and evidence

Studied: [Building an Infinite Liquid Glass Grid with Three.js, WebGPU, and TSL](https://tympanus.net/codrops/2026/09/08/building-an-infinite-liquid-glass-grid-with-three-js-webgpu-and-tsl/), Filip Kantedal. Article SHA-256: fd30206ee21ed449f4f1be5c25591bbb5c25d4d742b45414a81ce40f11245d1a. Analysis SHA-256: 8a138f0e3a9a5c4f9ea83fbf9e30b6e48ed5f71eb1a7e78eaf64a87e215d1e90. Source revision: e22a9e6acae85ed5bbfb3c2a6ec39e81f8eec23477c906a979faa0ac8552d31b. The revision hashes the article/analysis hashes, the analyst principle and code-block evidence references; it identifies studied evidence, not a claim that the original demo executes.

Original acquisition: unresolved. Original license: unverified. originalSourceAcquired: false. The linked liquid-glass experiment client files are cached, but no downloadable source release or applicable site grant was established. The article supplies five implementation fragments, not the complete grid/cloth system.

No original article prose, code, artwork, source project or screenshot is distributed here. The authored-code MIT notice applies only to the independent implementation. Independently authored WebGL material and spherical projection study. No video stream, XPBD cloth solver or original project assets are included. HTML labels share projected centers and scale; this is not a claimed replica of the original CSS3D camera adapter.

## Initial state and controls

Move through a pool of twelve cards on a curved field. Their apparent bevels come from a height function; the color fringe comes from separate refracted texture samples.

The page starts stationary. Play motion, where present, advances this study’s presentation until paused; text, bounce, portrait, orbit, strand and mascot motion loops, while the event composer stops at its endpoint. Scroll studies advance through ordinary document scroll or their chapter buttons. Static view stops time-based motion; manual controls still inspect values. The three scroll studies instead show every chapter in ordinary reading flow. Pointer parallax centers its decorative layers.

Initial control values:

| Control key | Value |
| --- | --- |
| pan | 0 |
| curve | 0.62 |
| bevel | 0.08 |
| ior | 1.42 |
| dispersion | 0.045 |

## Running and editing

Open index.html directly. It contains all runtime code and visuals and needs no package install, account, network, remote asset or browser-origin storage. Edit its inline style and script to change the downloaded page. styles.css and script.js are reference copies of the inline source; changing those copies alone does not change the running page. Repository regeneration overwrites index.html and all generated copies, including any manual edits. Keep a separate copy when editing a download. The repository authoring modules in vendor/codrops-article-studies/scenes-*.mjs and runtime.mjs regenerate these copies through node scripts/build-codrops-article-studies.mjs. Repository regeneration also needs the private source evidence; downloaded-page execution does not.

## Verification boundary

This generation passed the captured article/analysis hash checks, exact ID membership, local canonical Markdown anchor/hash checks, implementation-symbol checks, JavaScript parsing, semantic main/title counts, duplicate-ID checks and runtime URL/storage exclusion. The manifest codeFiles entries identify every emitted file by SHA-256. Each code symbol is a searchable named function, not a claimed executed call trace. Proposed checks are in KNOWLEDGE.json. runtime_verified: false. SG artifact capture proves only a recorded initial render; no interactive runtime, keyboard, accessibility, comfort or performance result is claimed.

Manual Chrome interaction is blocked by extension UI in this session. No alternate browser-control route was used. Preview sizing is native responsive width. Initial-render captures are tracked separately at site/artifacts/full-import/preview-posters/index.json and site/artifacts/full-import/preview-posters/visual-review.json in the repository. No capture receipt is bundled with this download. The bounded initial-render matrix requests all 21 studies at 320 pixels and three representative studies at 1280 pixels; the separate poster surface is 640 × 420. The authored CSS also adapts at 375 and 768 pixels, without claiming those widths were captured. A width request does not mean that capture exists or passed. Motion starts on explicit playback; system reduced motion and the Static view switch stop it.

consumer_reference: consumer-reference/agent-native/registry.json