Brief 23 · Prompt edition 2.1

Lightkeeper's Atelier

Build a real-time 3D lantern workshop in one HTML document that makes polished metal, rough metal, ceramic, stone, and glass visibly distinguishable as a movable inspection light sweeps across them.

Read the complete prompt
Lightkeeper's Atelier benchmark illustration
Test concept · editorial illustration

4 recorded evaluations of this frozen brief · artifacts published unedited

Recorded evaluations

Each artifact was produced against edition 2.1 of this brief and is published exactly as delivered. The model run page links the run record, which states which source-level checks were performed, which checks remain for a human browser review, and how this delivery differs from a hosted-API single-message episode.

What this tests

material differentiation, dynamic light, geometric shadows, and art direction.

Proposed difficulty

Watch, Yingzao, and Helm's Deep scene craft now faces controlled material/light interventions rather than a single flattering view.

These are design hypotheses, not observations of model failure. Empirical discrimination has not been measured.

One message, one HTML response

Give each model the exact complete prompt as one user message. It returns one self-contained HTML document in one response, without tools, external assets, dependencies or follow-up repair. Mental validation is not runtime evidence; the operator performs the actual offline browser checklist afterward.

Do not append BENCHMARK-PROMPT.md to this brief.

Rubric and pass rules

Five binary observations per brief, with no numeric scores or rubric weights. Pass only when every criterion and the offline single-document protocol are observed. Untested checks remain not-observed. Inspect at 1280×720 and a narrow viewport; retain uncut evidence and separate aesthetic preferences from functional observations.

L1: material distinction
Same-color brass pair separates by highlight width; ceramic and stone differ in response Fail: Color-only material differentiation or identical plastic shaders
L2: dynamic lighting
Marker, highlights, and shadows follow key; off removes its contribution Fail: Baked highlights, static shadow blob, exposure hiding toggles
L3: internal illumination
Inner light visibly affects frame and table with key off Fail: Emissive icon only
L4: glass and depth
Red object remains behind transparent glass; orbit preserves depth and ordering Fail: Opaque glass, flat composition, broken sorting or clipping
L5: readable composition
Full labeled scene, grounded objects, Neutral mode, orbit, reset, resize Fail: Missing specimens, effects concealing failures, unusable controls

Pass: Pass requires L1–L5 and the offline single-document protocol. Capture all judgments in Neutral as well as Beauty. Record aesthetic preference separately as commentary, not as an objective binary or a numeric score. Approximate procedural reflections and projected shadows are valid when they satisfy the observable requirements.

Controls and reviewer sequence

  • Drag to orbit, zoom, and use Home camera. Sweep or set key azimuth; toggle key and inner light separately.
  • Compare Beauty and Neutral; Reset restores the documented default.

Lock equal viewport and default view, then perform the same -60/0/+60 sweep and inner-light toggle on every run. Cut to matching orbit angles for glass. Let viewers compare material behavior and composition directly, with renderer limitations described accurately.

Review beyond the examples

Use arbitrary slider positions with Sweep off; orbit while the light remains fixed; compare the same brass pair in Neutral. Inspect source if necessary to confirm normal/view/light-dependent shading and geometric shadows. A canned movie or screen-space highlight sweep fails.

Versions and prompt revisions

Superbench 2.1 combines four new v21 contracts with Ember Glider and Low-Poly Tower Defense as returning 1.x briefs. The returning prompts and their existing artifacts keep their original repository-run protocol; they are not relabeled as tool-free 2.1 generations. Superbench 2.0 retains v2, while the remaining 1.x core, specialist and archived results retain their original protocols. No historical scores are migrated or compared quantitatively across editions.

Edition 2.1 is the launch contract. Changes to behavior, fixtures or pass criteria require a new documented edition; future evaluations must retain the exact evaluated prompt revision. Typographic corrections must be identified.

Open the full prompt and validation checklist