fix(examples): redesign sky-texture-sun-elevation hero as a sky-lit obelisk diptych

[TMHSDigital]

What changed and why

The gallery review found that the sky-texture-sun-elevation hero broke Layer 1 in three ways:

• The horizon band clipped to pure white under Standard.
• The frame was a hard left/right split with a black void below the horizon.
• The subject was a primitive cone.

The redesigned still is a sky study. A red-granite gnomon obelisk (stepped limestone plinth, bronze collar, bronze dedication plaque, matte-gilt pyramidion) stands on a brick-pattern limestone plaza with a bronze meridian scale inlaid along the shadow line. The ground runs to the horizon, so there is no void below it.

• The sky is the only light. The sun disc is the key and the dome is the fill; there are no studio lights. Both panels use one Background Strength (0.035), so sun_elevation is the only variable. If it failed, the panels would be identical.
• 8° panel: navy dusk, an orange glow on the left horizon, the obelisk's sun face raked orange, and a long shadow running out of frame.
• 55° panel: bright blue midday, sunlit paving, and a short shadow ending on the meridian scale.
• The split is deliberate: a 12 px dark rule between two 640×720 panels.
• Materials: 5 designed materials (granite and limestone use noise-mixed tones; paving uses a Brick texture) on 8 named parts. Stage deviation (world/sky contract) is documented in the README.

New render-path gates (the check-only path and CI smoke are unchanged):

• Framing: now wired through gallery_framing.check_framing for each panel (exit 10). The README's old "Framing deviation" is removed, and the helper's docstring no longer says this example skips it.
• Clipped highlights: new exit 12 fails if more than 0.05 % of the still is at 8-bit full scale under Standard. The committed still has zero clipped pixels.

Two root causes found along the way:

• sun_rotation is a compass bearing, clockwise from +Y. The first draft assumed the opposite and put the shadow on the wrong side; _sun_rotation_for_shadow now holds the measured mapping.
• A metallic pyramidion Fresnel-glinted the 55° sun to white at grazing angles regardless of roughness, so the gilt is a matte dielectric.

Before / after

Contact sheet (committed): docs/gallery/contact-sheets/sky-texture-sun-elevation-contact-sheet.webp

Gate numbers (5.2.1)

Gate	Result
Framing fill	x 0.566 / y 0.794 (band 0.70–0.90)
Framing margins	left 0.100, right 0.334, bottom 0.044, top 0.161 (min 0.02)
Clip fraction	0.000000 (gate ≤ 0.0005)
Panel mean luma	8° 0.112, 55° 0.294
Hero drift (measure_hero_drift.py)	mean_abs 0.00368, luma 0.2015/0.2011, matches
Hero / preview size	36.9 KB / 34.1 KB

Contact-sheet verdicts. This is a stage deviation, so luminance is information only:

• Mean luminance: 0.2015, between damped-track-aim (0.1511) and armature-bend (0.2787) / bmesh-gear (0.3024).
• Stage darkness: 0.1875 against 0.145–0.236. The sky replaces the stage by design.
• Wedge warmth: +0.20 (calibration −0.43 to +0.37). There is no wedge light; the warmth comes from the dusk glow and the sunlit plaza.
• Saturation: 0.54, higher than the calibration set (0.14–0.23), driven by the blue sky.
• Subject fill: in band.
• Thumbnail legibility: passes. At 320 px, dusk versus midday reads instantly.
• Overall: not sortable as a different gallery on materials, finish, view transform or framing.

Check runs (check-only path, tests/smoke/run_example.py --forbid-skip)

Binary	Reported version	Result
E:\Blender-Developer-Tools\.scratch\blender-5.2.1-windows-x64\blender.exe	5.2.1 LTS	PASS: MULTIPLE_SCATTERING, zenith rise 2.2607
E:\Blender-Developer-Tools\.scratch\blender-5.1.2-windows-x64\blender.exe	5.1.2	PASS: MULTIPLE_SCATTERING, zenith rise 2.2546
E:\Blender-Developer-Tools\.scratch\blender-4.5.11-windows-x64\blender.exe	4.5.11 LTS	PASS: NISHITA, zenith rise 1.4958

The --output render path also exits 0 on all three binaries with zero clipped pixels (4.5.11 panel luma 0.087 / 0.200).

Falsifiers (5.2.1)

• Existing: --unlink-sky gives exit 6: world links broken (sky→bg=False, bg→out=True).
• New clip gate: STILL_STRENGTH 0.035 → 0.05 in a temporary copy gives exit 12: clip fraction 0.020344 (8° panel 0.008019, 55° panel 0.033383), and no PNG is written. The copy was then removed.
• Newly wired framing gate: camera moved from y −4.2 to −3.2 in a temporary copy gives exit 10: fill y 0.925, left and bottom margins 0.000. The copy was then removed.

Proven by live run vs established by inspection only

Proven by live run:

• All three check-only runs above.
• The render path on all three binaries.
• The three falsifiers.
• The framing and clip numbers.
• render_hero.py, contact_sheet.py, measure_hero_drift.py and build_gallery.py.
• The generated <img alt> in docs/gallery/index.html and docs/gallery/sky-texture-sun-elevation/index.html was read back.

Established by inspection only:

• The aesthetic verdicts (the hero and a 320 px thumbnail were viewed).
• That the meridian-scale inlay and plaque read as intended at full size.
• CI (blender-smoke.yml) has not run yet; this PR relies on its run.

Not wired: gallery_asset_quality (exit 11). Per docs/VISUAL-STYLE.md it applies to asset-type examples, not world/sky subjects.

🤖 Generated with Claude Code