fix(examples): redesign attribute-domain-shear as striped patio parasols

[TMHSDigital]

What changed and why

The gallery review judged the attribute-domain-shear hero weak. The "wrong" POINT umbrella read as a smoother, arguably prettier rainbow gradient, which undercut the lesson. The "CORNER" / "POINT — last write wins" floor labels could not be read at card size.

Redesign: two striped patio parasols.

• Subject. Each parasol is eight named parts with five materials: Canopy (attribute fabric, fully matte), Ribs (aluminium ribs and stretchers), Runner / Finial / TiltKnuckle (brass), PoleUpper / PoleLower (teak), and Base (cast iron, lathe-turned). The canopy tilts 24° toward the camera so the whole top reads.
• Witness. The canopy is the check's fan grown into fabric: eight gores around one shared apex vertex, with every gore vertex on a seam. It is painted by the same assign_corner / assign_point_naive functions the check asserts. CORNER (left) keeps crisp crimson and cream stripes. Naive POINT (right) smears the stripes pink along every seam. Gore 0, crimson by intent and turned to face the camera, renders cream, because gore 7 wrote its seam and the apex last. At a glance it reads as broken, not as a nicer gradient.
• Palette. The render uses a two-tone palette. The check keeps its eight distinct hues, because an off-by-two ordering bug would pass unnoticed under a period-2 palette. The README says so.
• Labels. There is one bold standing block label per parasol: CORNER and POINT. Both read at 427×240.
• Code. assign_corner / assign_point_naive gained a faces_per_wedge stride (wedge-major face order). On the check's fan it is 1, the write order is unchanged, and the check output is identical.
• Gates. The render path now also runs gallery_asset_quality (exit 11) on one parasol.

Evidence

• Hero: docs/gallery/assets/attribute-domain-shear-hero.webp
• Contact sheet (committed): docs/gallery/contact-sheets/attribute-domain-shear-contact-sheet.webp

Gate numbers (5.2.1 render path)

• Framing: fill x=0.800 y=0.639 (band 0.70–0.90), margins L 0.100 / R 0.100 / B 0.083 / T 0.194. framing_ok.
• Asset quality (left parasol): 8 parts, no default names, 5 materials, edge90 0.061. aq_ok.
• Drift: measure_hero_drift.py --only attribute-domain-shear gives mean_abs 0.00321 and matches.

Contact sheet verdicts

tile	luma	stage	warmth	sat
attribute-domain-shear (candidate)	0.2700	0.2060	+0.0725	0.1441
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

• Stage darkness: pass. The stage (0.206) sits between bmesh-gear and armature-bend.
• Wedge warmth: pass, weak. The warm pool is visible behind the canopies, but its warmth (+0.07) is lower than damped-track-aim and bmesh-gear. The cream canopy dilutes the tile measurement.
• Subject fill: pass. Fill is 0.80 in x.
• Saturation: pass. It matches armature-bend. The saturated crimson carries it.
• Thumbnail legibility: pass. Crisp stripes against smeared stripes with a missing red panel read in under a second at 427×240, and both labels are legible.
• Mean luminance: 0.270, against 0.151–0.279 for the calibration set.

Check-only runs (tests/smoke/run_example.py)

binary	reported version	result
E:\Blender-Developer-Tools\.scratch\blender-5.2.1-windows-x64\blender.exe	Blender 5.2.1 LTS	PASS, shear 0.751031 = closed form
E:\Blender-Developer-Tools\.scratch\blender-5.1.2-windows-x64\blender.exe	Blender 5.1.2	PASS, identical
E:\Blender-Developer-Tools\.scratch\blender-4.5.11-windows-x64\blender.exe	Blender 4.5.11 LTS	PASS, identical

The render path also ran on 4.5.11 (framing_ok, aq_ok, still written).

Falsifiers (re-proven on 5.2.1 after the authoring-function change)

probe	exit	measured
CORNER: face 3 painted wedge 2's color	4	corner_hub_max_err 3.895e-01
POINT: wedges written in reverse order	5	hub reads palette[0], not palette[7]
POINT: ring vert 2 overwritten after the loop	6	point_ring_max_err 7.790e-01
constant palette	7	shear 0.000000
--no-overwrite	5	hub is not the last write

Proven by live run vs established by inspection

Proven by live run:

• The check passes on all three binaries with identical numbers.
• All five falsifiers fire.
• The render path passes framing and asset-quality on 5.2.1 and 4.5.11.
• The hero was regenerated via render_hero.py on 5.2.1, and the drift check reports matches.
• The contact sheet was built by contact_sheet.py.
• build_gallery.py was re-run after rebasing onto main, and the generated <img alt> was read in docs/gallery/index.html and docs/gallery/attribute-domain-shear/index.html.

Established by inspection only:

• Which gores smear, and why gore 0 turns cream. The render path does not assert these; the check asserts the same last-write rule on the fan.
• The visual verdicts above.

Asset sheet: not committed. This is an attributes teaching example, not an asset-type prop, so the gate does not formally apply. I still ran scripts/asset_sheet.py as a sanity check. The parasol is clean and plausible, but it is the simplest object in that lineup of hydrant, drone, well, wheelbarrow and shelf.

🤖 Generated with Claude Code