v0.3.0 — Subject fidelity and crafted rendering

Subject fidelity and rendering selection

This revision fixes a real fusion gap: the initial six lessons demonstrated interaction but did not use the bundled rendering kits. The skill now chooses subject representation before projection, routes iso-figure to functional isometric devices and Anatomy to crafted rigid mechanisms, and uses domain geometry or illustration for organic anatomy and textiles. Teaching correctness and visual fidelity have separate acceptance gates.

• Human chest: licensed Human Reference Atlas meshes replace the cylinder and ellipsoids. Source alignment is preserved, WebGL uses smooth normals and depth testing, and missing-WebGL/context-loss paths retain text explanations. The diaphragm remains explicitly schematic. CC BY 4.0 attribution, source hashes and rebuild instructions are bundled.
• Rocket: actual Anatomy lathe, ring, rib and tube helpers generate a tank section, assembly separation and engine detail. Inspection remains separate from the unchanged flight model. The original generic structure does not claim real-vehicle performance.
• Ming robe: shaded fold surfaces, cloth overlap/thickness and an enlarged fold section replace flat strokes. Mobile captions use readable wrapping HTML. Reconstruction and museum evidence remain distinct.

Verification

21 browser cases and 4 model/spec checks pass. Atlas and rocket preparation scripts were rerun; standalone outputs work offline. An independent rendering/provenance review was completed and its mobile caption finding fixed. The portable archive passes skill validation and carries the required source licenses.

No new clinical validation, physical mobile GPU survey or exhaustive mesh-intersection audit is claimed. The prior full turning-dial audit covers that unchanged fixture.

Live demos: https://ywedai.github.io/teachable-figures/
Review: https://github.com/ywEdAi/teachable-figures/blob/main/reviews/rendering-fusion-audit.md

Download teachable-figures.skill and use the accompanying SHA-256 file to verify it.

Teachable figures v0.2.0 · repaired 3D and five new lessons

Geometry repair

Fixed the four turning-dial crowding defects through real socket/terminal geometry. The unchanged full audit passes with exit 0 and zero failures in every category. Rendering-kit code, audit thresholds and exclusions are unchanged. The raw audit log and exact builder digest are in the repository.

Five new offline lessons

• Cone sections with actual 3D rotation and area comparisons.
• Human chest 3D cutaway with cited anatomical relationships.
• Ideal rocket launch with fuel loss, liftoff, burnout and coast.
• Toy Transformer data flow with causal attention calculations.
• A museum-sourced Ming robe with inspectable overlap and construction.

The example directory links all six complete lessons including the original gear experiment. Each runs offline and includes native controls, reset, explanation and transfer prompts.

Verification

All 16 model/build checks and 19 browser cases pass. Independent review found and prompted fixes for a cone endpoint and stale feedback. Desktop/mobile and theme previews were inspected.

Download teachable-figures.skill and its SHA256 file below. The archive includes the repaired builder and all examples. The anatomy is schematic, the rocket is an ideal model and the garment is a documented teaching reconstruction. Learner and assistive-technology studies remain open.

Reviewed teachable-figures skill v0.1.0

Reviewed teachable-figures skill.

Includes independent UI/UX and pedagogical reviews, corrected lightweight SVG examples, a complete offline gear-ratio lesson, and 24 passing automated checks across models, builds, and browser interactions.

Download teachable-figures.skill for the portable skill package. Source, review reports, and reproducible checks are in the repository.

Known limitation: the upstream turning-dial quick orbit audit retains four crowding findings. Full 3D, screen-reader, and novice-learner validation remain open.