fix(examples): redesign shader-node-group as five dipped-glaze stoneware mugs

[TMHSDigital]

What and why

The gallery review judged the shader-node-group hero programmer art: two plain glossy spheres, teal and magenta. The render did not show a shared group doing any work.

The redesign turns "one group, many tints" into a small glaze library:

• The group is DippedGlaze, a designed stoneware shader. It has bare buff clay at the foot, a wavy dip line (Generated Z against object-space noise), iron speckle (Voronoi) through clay and glaze alike, a glaze that breaks back toward clay over the rim, and a coat only on the glazed part. All of this lives inside the group. The interface still exposes only Tint, Roughness and Shader.
• The subject is five lathed stoneware mugs. Each has a foot ring, bellied wall, rolled lip and a swept oval handle, with subsurf at render. The tints run warm to cool: oxblood, amber, celadon, teal, cobalt.
• The staging is a staggered shop-display lineup: the odd mugs stand on a low walnut riser behind the even ones. All handles turn the same way, so colour is the only thing that changes along the row. The foot band, dip line, speckle and rim break are visibly identical on every mug.
• Render path only: the house wedge is added (the old stage had none), the key's spread is narrowed off the back wall, and the floor is darker. No floor labels.

Before/after: the contact sheet is committed at docs/gallery/contact-sheets/shader-node-group-contact-sheet.webp. The before is the two-sphere hero on main.

Check contract: at least as strict as before

Gate	Before	After
3 interface sockets	Tint/Roughness/Shader	unchanged
4 shared datablock	users == 2	users == len(mugs) (5)
5 instance tree identity	each of 2	each of 5
6 per-instance Tint	a != b	all 5 pairwise distinct

Gate numbers (5.2.1)

• Framing: fill x=0.841 (band 0.70–0.90), y=0.489. Margins: L 0.100, R 0.059, B 0.206, T 0.306 (min 0.02). framing_ok.
• Contact sheet (luma / stage / wedge warmth / saturation):
  • candidate: 0.2288 / 0.2043 / +0.0842 / 0.2188
  • armature-bend: 0.2787 / 0.2132 / -0.4279 / 0.1436
  • damped-track-aim: 0.1511 / 0.1448 / +0.2772 / 0.2305
  • bmesh-gear: 0.2358 / 0.1772 / +0.3027 / 0.2495
• Verdicts:
  • Stage darkness: PASS. It sits inside the calibration range, darker than armature-bend.
  • Wedge warmth: PASS. It is positive, but milder than damped-track-aim and bmesh-gear.
  • Subject fill: PASS.
  • Saturation: PASS, within range.
  • Thumbnail legibility: PASS. Five coloured mugs with pale clay feet read at card size.
• measure_hero_drift.py --only shader-node-group: mean_abs 0.00315, matches.
• Hero 26 KB, preview 23 KB, both written by render_hero.py.
• Asset-sheet gate: not run. This is a materials/shader example, not an asset-type (game prop/kit) entry.

Check runs

Binary (reported version)	check-only	--same-tint	--output render
.scratch/blender-5.2.1-windows-x64/blender.exe (Blender 5.2.1 LTS)	exit 0, users=5	exit 6	framing_ok, OK
.scratch/blender-5.1.2-windows-x64/blender.exe (Blender 5.1.2)	exit 0, users=5	exit 6	framing_ok, OK
.scratch/blender-4.5.11-windows-x64/blender.exe (Blender 4.5.11 LTS)	exit 0, users=5	exit 6	framing_ok, OK

Falsifiers (5.2.1)

• Gate 6: --same-tint gives exit 6: instance Tint values not pairwise distinct [(0.42, 0.035, 0.03, 1.0) x5].
• Gate 4: a probe adds a sixth material that instances the group, giving exit 4: group users 6 != 5 (not shared).
• Gate 5: a probe points Mug.Celadon's group node at tree.copy(), plus a stray user so the count stays 5. This gives exit 5: Mug.Celadon instance points at a different tree. Without the stray user, gate 4 fires first (users 4 != 5).

Proven by live run vs established by inspection

Proven by live run:

• Check-only and --same-tint on 5.2.1, 5.1.2 and 4.5.11.
• The render path, including the framing gate, on all three.
• The gate 4 and gate 5 probes on 5.2.1.
• render_hero.py, contact_sheet.py and measure_hero_drift.py on 5.2.1.
• build_gallery.py. The generated <img alt> in docs/gallery/index.html and docs/gallery/shader-node-group/index.html was read back, and so were the users == 5 witnesses.

By inspection only:

• The hero's visual verdicts, judged by eye at full size and on the contact sheet.
• CI smoke (pending on this PR).

🤖 Generated with Claude Code