fix: restage three earlier heroes as designed objects

README.md

@@ -149,7 +149,7 @@ shader `Attribute` node is actually linked to Base Color.
 </tr>
 <tr>
 <td width="46%" valign="middle">
-<a href="examples/image-pixels-testcard/"><img src="examples/image-pixels-testcard/preview.webp" alt="Image pixels testcard: a studio monitor showing a procedural broadcast test card — seven neon color bars behind the classic dark circle, a luminance ramp, and a PLUGE row with a white bottom-left origin marker — over a teal underglow on a dark studio floor" /></a>
+<a href="examples/image-pixels-testcard/"><img src="examples/image-pixels-testcard/preview.webp" alt="Image pixels testcard: a studio monitor on a machined metal stand showing a procedural broadcast test card — seven neon color bars behind the classic dark circle, a luminance ramp, and a PLUGE row with a white bottom-left origin marker — in a dark studio with a warm pool raking the back wall" /></a>
 </td>
 <td valign="middle">
 
@@ -168,7 +168,7 @@ non-destructive path.
 </tr>
 <tr>
 <td width="46%" valign="middle">
-<a href="examples/png-exr-alpha/"><img src="examples/png-exr-alpha/preview.webp" alt="PNG vs EXR alpha: two easel panels on a dark studio floor — left float→PNG false-unpremul clamps dark mid-tones to white at low alpha, right the EXR-clean authored straight buffer with surviving red and green primaries" /></a>
+<a href="examples/png-exr-alpha/"><img src="examples/png-exr-alpha/preview.webp" alt="PNG vs EXR alpha: two framed verification displays on a shared plinth in a dark studio — left the float→PNG false-unpremul mangling clamps dark rows to white, right the EXR-clean straight buffer, nameplates reading FLOAT → PNG and FLOAT → EXR under a warm wall pool" /></a>
 </td>
 <td valign="middle">
 
@@ -205,7 +205,7 @@ result is watertight (every edge borders exactly two faces).
 </tr>
 <tr>
 <td width="46%" valign="middle">
-<a href="examples/uv-layer-grid/"><img src="examples/uv-layer-grid/preview.webp" alt="UV layer grid: two easel panels on a dark studio floor — left a flat teal square from a missing UV layer sampling texel (0,0), right a magenta/cyan neon checker proving pre-create + calc_uvs" /></a>
+<a href="examples/uv-layer-grid/"><img src="examples/uv-layer-grid/preview.webp" alt="UV layer grid: two framed lightbox displays on floor trays against a dark studio wall — left one flat teal face (the calc_uvs silent no-op hazard), right a saturated magenta-cyan checker (the pre-create repair), under a warm wall pool" /></a>
 </td>
 <td valign="middle">
 

docs/gallery/gltf-export-roundtrip/index.html

@@ -186,7 +186,7 @@ <h1>gltf-export-roundtrip</h1>
 <p>A runnable example that builds a sci-fi supply crate — 35 beveled box shells with three material slots and box-mapped UVs — exports it with <code>bpy.ops.export_scene.gltf</code>, parses the file on disk, re-imports it, and verifies the whole round-trip against the depsgraph-evaluated mesh, following <a href="https://github.com/TMHSDigital/Blender-Developer-Tools/tree/main/skills/depsgraph-and-evaluated-data/SKILL.md"><code>depsgraph-and-evaluated-data</code></a> and <a href="https://github.com/TMHSDigital/Blender-Developer-Tools/tree/main/skills/headless-batch-scripting/SKILL.md"><code>headless-batch-scripting</code></a>.</p>
 <p><strong>Pipeline arc:</strong> modeling/LOD in <a href="https://github.com/TMHSDigital/Blender-Developer-Tools/tree/main/examples/lod-decimate-chain"><code>lod-decimate-chain</code></a>, weighting in <a href="https://github.com/TMHSDigital/Blender-Developer-Tools/tree/main/examples/vertex-weight-limit"><code>vertex-weight-limit</code></a>, export here.</p>
 <p><strong>What it witnesses:</strong> the interchange contracts AI-generated export code most often gets silently wrong.</p>
-<p>1. <strong>The +Y-up convention is baked into vertex data.</strong> glTF is +Y-up, Blender is +Z-up, and <code>export_yup=True</code> (the default) writes <code>(x, y, z) -&gt; (x, z, -y)</code> directly into the POSITION buffer — the node carries <strong>no</strong> rotation or scale. The check parses the <code>.gltf</code> JSON and asserts the accessor bounds equal the axis-converted evaluated bounding box, and that the node transform is absent. Exporting with <code>export_yup=False</code> ships raw Z-up data every engine displays lying on its back. 2. <strong><code>export_apply=True</code> ships the evaluated mesh, not the base cage.</strong> The crate&#x27;s bevel modifier lives only in the depsgraph; with flat shading and UV seams the exporter splits exactly one vertex per evaluated loop (7,560), so the on-disk POSITION count is an exact witness. <code>export_apply=False</code> silently writes the 624-vertex cage. 3. <strong>The round-trip is faithful.</strong> Re-imported positions (bit-exact here), loop normals (≤2e-4), box-mapped UVs (≤3e-5), and per-triangle material bindings all match the evaluated mesh. UVs are V-flipped on disk (glTF texture origin is top-left) and flipped back on import — both flips are proven by reading the <code>.bin</code> buffer directly.</p>
+<ul><li><strong>The +Y-up convention is baked into vertex data.</strong> glTF is +Y-up, Blender is +Z-up, and <code>export_yup=True</code> (the default) writes <code>(x, y, z) -&gt; (x, z, -y)</code> directly into the POSITION buffer — the node carries <strong>no</strong> rotation or scale. The check parses the <code>.gltf</code> JSON and asserts the accessor bounds equal the axis-converted evaluated bounding box, and that the node transform is absent. Exporting with <code>export_yup=False</code> ships raw Z-up data every engine displays lying on its back.</li><li><strong><code>export_apply=True</code> ships the evaluated mesh, not the base cage.</strong> The crate&#x27;s bevel modifier lives only in the depsgraph; with flat shading and UV seams the exporter splits exactly one vertex per evaluated loop (7,560), so the on-disk POSITION count is an exact witness. <code>export_apply=False</code> silently writes the 624-vertex cage.</li><li><strong>The round-trip is faithful.</strong> Re-imported positions (bit-exact here), loop normals (≤2e-4), box-mapped UVs (≤3e-5), and per-triangle material bindings all match the evaluated mesh. UVs are V-flipped on disk (glTF texture origin is top-left) and flipped back on import — both flips are proven by reading the <code>.bin</code> buffer directly.</li></ul>
 <p><strong>Version witness (probed on Blender 4.5.11 LTS and 5.1.2):</strong> the operator signatures are byte-identical — 109 exporter properties, 20 importer properties, same defaults — and the exported JSON differs only in <code>asset.generator</code> (&quot;Khronos glTF Blender I/O v4.5.51&quot; vs &quot;v5.1.20&quot;). The example therefore runs identical kwargs on both versions and guards forward drift explicitly: every kwarg it passes must still exist in the operator&#x27;s RNA, so a future rename fails loudly instead of drifting silently. One genuine 5.x removal surfaced during authoring: <code>Mesh.calc_normals()</code> is gone (loop normals auto-compute on read) — calling it is itself a cross-version hazard, noted in the code.</p>
 <p>Two more authoring hazards are pinned in comments: exact face-plane coincidences between kit-bashed shells weld loops on export (the count check catches it), and <code>read_factory_settings</code> mid-check frees the original mesh — touching a freed RNA raises <code>ReferenceError</code>, so counts are captured before the wipe.</p>
 <p>The render stages the authored crate beside the actual re-imported one — same bevels, same materials carried through the file itself. If the axis conversion broke, the right twin would lie on its side; if the modifier contract broke, its silhouette would lose the rounded edges.</p>

docs/gallery/gltf-skin-roundtrip/index.html

@@ -186,7 +186,7 @@ <h1>gltf-skin-roundtrip</h1>
 <p>A runnable example that rigs a mech scorpion — seven-bone chain from pedestal root to stinger, blend rings at every joint seam — exports it with <code>bpy.ops.export_scene.gltf</code> (<code>export_skins=True</code>), parses the file, re-imports it, and verifies the whole skinning contract against the authored rig, following <a href="https://github.com/TMHSDigital/Blender-Developer-Tools/tree/main/skills/depsgraph-and-evaluated-data/SKILL.md"><code>depsgraph-and-evaluated-data</code></a>.</p>
 <p><strong>Pipeline arc:</strong> modeling/LOD in <a href="https://github.com/TMHSDigital/Blender-Developer-Tools/tree/main/examples/lod-decimate-chain"><code>lod-decimate-chain</code></a>, weighting in <a href="https://github.com/TMHSDigital/Blender-Developer-Tools/tree/main/examples/vertex-weight-limit"><code>vertex-weight-limit</code></a>, export in <a href="https://github.com/TMHSDigital/Blender-Developer-Tools/tree/main/examples/gltf-export-roundtrip"><code>gltf-export-roundtrip</code></a> — this is the skinning counterpart to the crate&#x27;s geometry round-trip. Tangent frames for the normal maps are in <a href="https://github.com/TMHSDigital/Blender-Developer-Tools/tree/main/examples/triangulate-tangents"><code>triangulate-tangents</code></a>.</p>
 <p><strong>What it witnesses:</strong> the skinned-mesh export contract the geometry round-trip left uncovered.</p>
-<p>1. <strong>The skeleton survives.</strong> <code>skins[0].joints</code> names every bone; the re-imported armature carries the same 7 bones, the same parent chain, and rest matrices within 2.4e-07 — the +Y-up conversion applies to bone nodes exactly as it does to meshes (their translations convert, no rotation is written). 2. <strong>The weights survive.</strong> Every primitive carries JOINTS_0/WEIGHTS_0; per-vertex weights on disk sum to 1 (err 3.0e-08); the re-imported vertex groups match the authored groups <strong>bit-exactly</strong> (w_err 0.0), compared as straddle-safe position keys, the same protocol as the crate example. 3. <strong>The deformation survives.</strong> Posed identically, the re-imported rig&#x27;s evaluated mesh matches the original&#x27;s within 4.8e-07 — linear blend skinning through the file format. The comparison is by rest-position key, never sorted multisets: the exporter welds duplicate loops (32 here), so cardinalities differ and a naive sorted zip mispairs vertices (a phantom 2.29 &quot;deviation&quot; measured and fixed during authoring). 4. <strong>The mesh must be parented to the armature.</strong> The exporter warns &quot;Armature must be the parent of skinned mesh&quot; and picks an armature by name otherwise — with two rigs in the file it can bind the wrong one.</p>
+<ul><li><strong>The skeleton survives.</strong> <code>skins[0].joints</code> names every bone; the re-imported armature carries the same 7 bones, the same parent chain, and rest matrices within 2.4e-07 — the +Y-up conversion applies to bone nodes exactly as it does to meshes (their translations convert, no rotation is written).</li><li><strong>The weights survive.</strong> Every primitive carries JOINTS_0/WEIGHTS_0; per-vertex weights on disk sum to 1 (err 3.0e-08); the re-imported vertex groups match the authored groups <strong>bit-exactly</strong> (w_err 0.0), compared as straddle-safe position keys, the same protocol as the crate example.</li><li><strong>The deformation survives.</strong> Posed identically, the re-imported rig&#x27;s evaluated mesh matches the original&#x27;s within 4.8e-07 — linear blend skinning through the file format. The comparison is by rest-position key, never sorted multisets: the exporter welds duplicate loops (32 here), so cardinalities differ and a naive sorted zip mispairs vertices (a phantom 2.29 &quot;deviation&quot; measured and fixed during authoring).</li><li><strong>The mesh must be parented to the armature.</strong> The exporter warns &quot;Armature must be the parent of skinned mesh&quot; and picks an armature by name otherwise — with two rigs in the file it can bind the wrong one.</li></ul>
 <p><strong>What each check catches on failure:</strong> exporting with <code>export_skins=False</code> (exit 5 — no skin on disk), stripping the weights (exit 4 — no vertex groups), and posing the re-imported rig differently (exit 19 — deformation deviates 0.90).</p>
 <p><strong>Version witness:</strong> the skins pipeline is stable between Blender 4.5 LTS and 5.1 — the exporter/importer RNA is byte-identical (probed with <code>gltf-export-roundtrip</code>), and every measured value matches to the digit on 4.5.11 and 5.1.2.</p>
 <p>The render stages the authored scorpion beside the actual re-imported one — same curl, same glowing stinger — proof the skin rode the format through.</p>