fix(examples): bmesh-gear turned-brass finish replaces moire banding

[TMHSDigital]

What changed and why

Gallery review: the bmesh-gear face was covered in coarse concentric banding that read as a moire/render artifact, not as machining. Two causes. First, a scale-16 RINGS wave drove roughness across a wide 0.14–0.36 range. Second, the wave was fed Generated coordinates, so the rings were centred on the bounding-box corner and crossed the face as off-axis arcs. The old still also clipped about 40k pixels, and the tooth flanks were close to black.

This is a material and lighting fix only. The mesh the check counts is unchanged:

• Lathe-faced caps. Rings run about the object Z axis (the gear axis) at scale 150, which is below a pixel. Roughness sits in a narrow 0.17–0.21 band with a 0.05 bump. The sub-pixel marks add up to the soft radial fan that a real turned face shows, with no stripes.
• Hobbed flanks. Flank roughness is 0.55. The object-space face normal separates flanks from caps, so the finish follows the geometry in any pose.
• Key highlight. A softbox sits on the camera ray's reflection about the front cap, offset to the upper left. Its soft reflection is the face's one clear highlight.
• Bounce card. A low card in front gives the downward tooth flanks something to reflect. The old hot Glint light was removed.
• Brass color. Base color is a little more saturated: (0.80, 0.47, 0.12).
• Framing gate added. The render path now runs the Layer 1 framing gate through gallery_framing.check_framing (exit 10), which it did not do before. The wedge is passed as staging.

Hub/bore: not added. A bore changes the closed-form counts (verts = 2 × 4 × teeth, faces = sides + 2 caps, edges = 3 × profile) that the check asserts, and faking a hole in the shader would be dishonest. The gear stays a solid blank, and the README now says so.

Calibration image changed

bmesh-gear belongs to the pinned contact-sheet calibration set (armature-bend, damped-track-aim, bmesh-gear). This PR changes that calibration image. Contact sheets that other examples committed earlier still show the old gear tile. They are historical evidence and were not regenerated. From now on, scripts/contact_sheet.py will composite the new hero.

Before / after

• Contact sheet (committed): docs/gallery/contact-sheets/bmesh-gear-contact-sheet.webp
• Before: the stripes shown in the previous docs/gallery/assets/bmesh-gear-hero.webp on main

Gate numbers (Blender 5.2.1 LTS)

Framing (silhouette): fill x=0.488 y=0.856 (band 0.70–0.90, ok). Margins: left=0.272 right=0.241 bottom=0.111 top=0.033 (min 0.020, ok).

Contact sheet (python scripts/contact_sheet.py bmesh-gear):

tile	luma	stage	warmth	sat
bmesh-gear (new)	0.2358	0.1772	+0.3027	0.2495
bmesh-gear (old hero)	0.302	0.236	+0.374	0.195
armature-bend	0.2787	0.2132	-0.4279	0.1436
damped-track-aim	0.1511	0.1448	+0.2772	0.2305

Verdicts:

• Stage darkness: pass. Stage is 0.177, within the calibration range (0.145–0.213). The old hero, at 0.236, was the brightest stage in the set.
• Wedge warmth: pass. +0.30, close to damped-track-aim (+0.28). The warm pool on the back wall is kept.
• Subject fill: pass. 0.856 in y.
• Saturation: pass. 0.250, the highest in the set, from the richer brass.
• Thumbnail legibility: pass. At 320×180 the image reads at once as a brass gear, with separate teeth, lit flanks and a clean face. It no longer looks like a striped defect.
• Mean luminance: 0.236 against 0.279 and 0.151 for the other two members, which is mid-set.
• Clipped pixels: 0 (any channel ≥ 254). The old hero had 40,050.

Hero drift (python scripts/measure_hero_drift.py --only bmesh-gear): mean_abs 0.00297, verdict matches.

Asset-sheet gate: not run. bmesh-gear is a mesh-API example, not a game prop or kit. Geometry is unchanged, so edge90 0.667 in VISUAL-STYLE's table still holds.

Check runs per binary

binary (reported version)	check-only	--no-extrude falsifier	--output render
E:\Blender-Developer-Tools\.scratch\blender-5.2.1-windows-x64\blender.exe (Blender 5.2.1 LTS)	exit 0, verts=112 edges=168 faces=58 watertight=True	exit 3, topology (56, 56, 1) != expected (112, 168, 58)	exit 0, framing_ok
E:\Blender-Developer-Tools\.scratch\blender-5.1.2-windows-x64\blender.exe (Blender 5.1.2)	exit 0	exit 3, same	exit 0, framing_ok, same numbers
E:\Blender-Developer-Tools\.scratch\blender-4.5.11-windows-x64\blender.exe (Blender 4.5.11 LTS)	exit 0	exit 3, same	exit 0, framing_ok, same numbers

Falsifiers

• Topology check (unchanged): --no-extrude gives exit 3 on all three binaries (see above).
• New framing call site: a temporary copy with cam_data.lens = 75.0 gave exit 10 on 5.2.1: framing_fill x=0.666 y=1.000 FAIL, margins bottom=0.000 top=0.000 FAIL. No PNG was written. The copy was then deleted.

Proven by live run vs established by inspection only

Proven by live run (headless):

• Check, falsifier and render exit codes on 5.2.1, 5.1.2 and 4.5.11.
• Framing numbers and the framing falsifier.
• Contact-sheet metrics and the clipped-pixel counts.
• Hero drift "matches".
• build_gallery.py regeneration. I read the generated <img alt> in docs/gallery/index.html and docs/gallery/bmesh-gear/index.html, and it is the new alt sentence in full.

Established by inspection only:

• The aesthetic verdicts: a turned-brass read, no moire at 1:1 or at thumbnail size, and flank legibility. I judged these by looking at the full-size render, a 3× nearest-neighbour crop of the face centre, and a 320×180 thumbnail.
• A very faint residual interference texture is visible only in the 3× crop and not at 1:1.
• The Live MCP was not used.

🤖 Generated with Claude Code