fix(examples): redesign gn-modifier-inputs as three GN spiral staircases

[TMHSDigital]

What changed and why

The gallery review judged the gn-modifier-inputs hero programmer art: three plain cubes (teal/orange/pink), the largest one cropped and flat-lit. This is a redesign, not a touch-up.

The shared Geometry Nodes tree now builds a parametric spiral staircase from its one Float input (renamed Scale → Height; the identifier is still Socket_1):

• step count = round((Height − 0.3) / 0.1), and the riser is re-derived so the top tread lands exactly
• the carrier mesh (one bevelled oak tread with its brass baluster) is instanced up a helix at 22.5° per step (Mesh Line → Instance on Points → Realize)
• a teal enamel newel post, base plate and collar that span exactly [0, Height]
• a brass helical handrail (Spiral → Curve to Mesh, reversed so it follows the baluster tops)

Heights 1 / 2 / 3 m give 7 / 17 / 27 treads. The three staircases read left to right as one design at three settings of one input. All three share one tree and one set of materials, so the modifier input is the only thing that differs. A lost write drops a staircase back to the 7-step default, which the render would show.

Stage: the default dark studio. The key has a 36° spread and the rim sits high behind the stairs, so the foreground floor stays dark (the old low rim was washing it blue). There are no labels.

Before / after

• Contact sheet: docs/gallery/contact-sheets/gn-modifier-inputs-contact-sheet.webp
• Asset sheet: docs/gallery/asset-sheets/gn-modifier-inputs.webp
• Before: three flat-coloured cubes (the previous gn-modifier-inputs-hero.webp on main)

Gate numbers (5.2.1 LTS, EEVEE)

gate	value
framing fill	x 0.653, y 0.850 (band 0.70–0.90) ok
margins	left 0.191, right 0.156, bottom 0.111, top 0.039 (min 0.020) ok
contact sheet luma / stage / warmth / sat	candidate 0.221 / 0.209 / +0.084 / 0.173
calibration	armature-bend 0.279 / 0.213 / −0.428 / 0.144; damped-track-aim 0.151 / 0.145 / +0.277 / 0.231; bmesh-gear 0.236 / 0.177 / +0.303 / 0.250
asset floors (SpiralStair.H3)	parts 1, materials 2 on the carrier mesh (3 on the evaluated staircase), edge90 0.112
hero drift	mean_abs 0.00356, verdict matches
hero / preview size	36.6 KB / 33.7 KB

Contact-sheet verdicts:

• Stage darkness: pass. Stage 0.209 falls inside the calibration range (0.145–0.213).
• Wedge warmth: pass, on the cool side. The warm pool sits on the back wall. At +0.084 the candidate is between armature-bend (−0.43) and the warm pair (~+0.29).
• Subject fill: pass. 0.85 vertical.
• Saturation: pass. 0.173 is within 0.144–0.250. Teal, oak and brass are complementary.
• Thumbnail legibility: pass. At 427×240 the three helices and the step progression read immediately.

Mean luma is 0.221, against a calibration range of 0.151–0.279.

Asset sheet verdict: pass. The tall staircase is small in its panel because it is a tall, thin subject. With named parts, three materials, bevelled treads and a continuous rail, it is not the least-designed object beside the hydrant, drone, well, wheelbarrow and shelf. I added a SELECT row in scripts/asset_sheet.py for it.

Check runs (proven by live run)

Each binary reports its version via --version:

binary	reports	smoke (run_example.py)	--same-height	--same-scale	--api dict	--api rna
E:\Blender-Developer-Tools\.scratch\blender-5.2.1-windows-x64\blender.exe	Blender 5.2.1 LTS	PASS	exit 7	exit 7	exit 5 TypeError: … id properties not supported for this type	exit 0
…\blender-5.1.2-windows-x64\blender.exe	Blender 5.1.2	PASS	exit 7	exit 7	exit 0	exit 5 AttributeError: 'NodesModifier' object has no attribute 'properties'
…\blender-4.5.11-windows-x64\blender.exe	Blender 4.5.11 LTS	PASS	exit 7	exit 7	exit 0	exit 5 (same AttributeError)

--same-height fails with readback 1.0 != written 2.0 on SpiralStair.H2.

Falsifier for the geometry-bearing check (exit 8): I set RAIL_H = -0.5 via an import probe, which makes the treads climb 0.5 m past the newel post. It fails on both 5.2.1 and 4.5.11 with exit 8: evaluated Z-extent 1.934334 != height 1.0 on SpiralStair.H1. Nothing was changed in the file; the probe was a scratch script.

The check contract itself is unchanged: shared node_group (4), per-version write and read (5/6), readback (7), extent == written value (8), zmin == 0 (9), distinct extents (11).

Proven by live run vs established by inspection only

• Proven by live run: the check-only path on all three binaries; every falsifier row above; the framing gate; the render; the contact sheet and asset sheet; hero drift matches; build_gallery.py. I also read the generated <img alt> in docs/gallery/index.html and docs/gallery/gn-modifier-inputs/index.html and the witnesses callout.
• Established by inspection only: the 7 / 17 / 27 tread counts. They follow from the tree's math, are visible in the render, and match the evaluated vertex counts (1588 / 3668 / 5748 = 132 + 208·n). No check asserts the tread count directly. Rendering was only tested on 5.2.1; 4.5 and 5.1 ran check-only.

🤖 Generated with Claude Code