fix(examples): redesign prop-origin-transform as a street utility pedestal with a thrown conduit elbow

[TMHSDigital]

What changed and why

The gallery review judged the prop-origin-transform hero to be programmer art: two teal boxes, each with a grey cylinder, cryptic TRAP / MPI KEEP floor text and unexplained green and pink tick marks. This PR redesigns both the asset and the composition.

Asset: a street utility pedestal, built as one mesh with 5 materials:

• precast concrete footing
• galvanised base plate with anchor nuts
• green cabinet on a black kick skirt, with a hipped lid
• door with hinges, a T-handle and a hazard plate
• louvred side vents, a conduit boss, and a sleeve in the pad

It is authored the way an import often arrives: origin at the geometric centre, with mesh coordinates divided by the non-uniform object scale (1.15, 0.92, 1.08). The data-API bake therefore lands on exactly the designed shape. The accessory is a flanged PVC conduit elbow (2 materials) with bolt heads, a strap clamp, and a coupling that seats in the sleeve.

Composition (no text):

• Left: the elbow is parented with matrix_parent_inverse and stays seated.
• Right: bare child.parent = parent applies the pedestal's transform a second time. The elbow is thrown 0.65 m past the slab edge and into the air, and a glowing cyan wireframe ghost marks the seat it left.
• A sidewalk slab raises both origins 0.20 m, so the doubled translation lifts the elbow instead of sliding it along the floor.
• Floor pivot rings centred on each origin, with red X and green Y crosshair notches, replace the buried axis stubs.

Wiring:

• The render path now also runs gallery_asset_quality (exit 11).
• scripts/asset_sheet.py gains a SELECT row for the asset (^(Pedestal|Conduit)\.Keep$).
• The alt text is updated in gallery.json and the README, and the gallery is regenerated.

Gates (Blender 5.2.1 LTS, E:\Blender-Developer-Tools\.scratch\blender-5.2.1-windows-x64\blender.exe)

• Framing: fill x=0.812 y=0.644 (band 0.70–0.90, ok). Margins: left 0.050, right 0.097, bottom 0.161, top 0.144 (≥0.02).

• Asset quality: parts=2, materials=7, dominant share 0.50, edge90=0.075, no default names. Result: aq_ok.

• Contact sheet: docs/gallery/contact-sheets/prop-origin-transform-contact-sheet.webp

  tile	luma	stage	warmth	sat
  prop-origin-transform	0.2498	0.2027	+0.0615	0.2304
  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. 0.203 is inside the calibration span (0.145–0.213). The first cut was 0.228 and was darkened.
  • Wedge warmth: pass. It is positive, but weaker than gear and damped-track.
  • Subject fill: pass.
  • Saturation: pass. 0.230 is in range.
  • Thumbnail legibility: pass. At card size the orange elbow hanging off to the right with its cyan ghost reads at once.
  • Mean luminance: 0.250 against calibration 0.151–0.279.

• Asset sheet: docs/gallery/asset-sheets/prop-origin-transform.webp, against collision-hull-proxy, socket-attach-points, vertex-color-ao, wheelbarrow and apothecary-shelf. Verdict: pass. The pedestal carries the most distinct parts and materials per silhouette of any piece in the lineup. It sits level with the hydrant and is not identifiable as the least-designed object.

• Drift: measure_hero_drift.py --only prop-origin-transform gives mean_abs 0.00367 and a verdict of matches.

• The generated <img alt> in docs/gallery/index.html and docs/gallery/prop-origin-transform/index.html was read back; the text is correct (193 chars).

Check runs

binary (reported)	check	--skip-mpi
5.2.1 LTS	exit 0, bare_jump 0.298469, mpi_err 0	exit 8 (err 2.985e-01)
5.1.2	exit 0, identical numbers	exit 8 (err 2.985e-01)
4.5.11 LTS	exit 0, identical numbers	exit 8 (err 2.985e-01)

tests/smoke/run_example.py on 5.2.1 gives [PASS]. The render path on 4.5.11 LTS also exits 0 with identical framing numbers.

Falsifiers

Check logic and thresholds are unchanged. Because the mesh changed, every mesh-dependent check was re-falsified on 5.2.1 with a probe that monkeypatches the helper out:

• origin bake removed: exit 6, local min.z=-0.580000 != 0
• scale apply removed: exit 5, scale (1.15, 0.92, 1.08)
• --skip-mpi: exit 8, err 0.2985 m

Proven by live run vs. established by inspection

• Proven by live run: every number above. That covers the checks and falsifiers on 5.2.1, 5.1.2 and 4.5.11, framing and asset quality on 5.2.1 and 4.5.11, the contact-sheet metrics, the asset-sheet render, the drift measurement, and the gallery build.
• Established by inspection only: the thumbnail-legibility and asset-sheet "not least-designed" verdicts, which are judgments made by eye on the committed composites.

🤖 Generated with Claude Code