Backfill contact sheets (26) and asset sheets (28) for existing showcase pieces

[TMHSDigital]

What

showcase/README.md now requires a contact sheet and an asset sheet for every showcase piece, each on its own recorded evidence. The 28 pieces already in the tree predate that rule.

Gate	Conformant	Non-conformant
Contact sheet	2 — crate-stack, stone-archway	26
Asset sheet	0	28

Why the rule exists

Contact sheets — the first sheets built for crate-stack and stone-archway showed both wedge pools reading as cool grey bands rather than the warm pool the house style calls for. Both pieces were relit as a direct result. Nothing else in the pipeline puts a still beside its peers, so nothing else could have caught it.

Asset sheets — showcase is entirely game props, which is exactly the scope docs/VISUAL-STYLE.md § Asset quality names. Budgets measure geometry conformance; the contact sheet measures staged presentation; neither removes the scene, and a strong scene carries a weak model. The evidence recorded in that file is socket-attach-points, which passed every measurable floor on its first draft — edge90 0.000, ten materials, no default datablock names — and was still judged bad by eye and rebuilt from scratch. "The floors scored the bevels, not the design."

Scope of the backlog

Contact sheets are cheap: compositing an existing committed hero beside the pinned calibration set needs no re-render. 26 sheets, mechanical.

Asset sheets are not. Each needs the hero re-rendered alone — neutral three-quarter view, plain studio lighting, no staging tricks — and the four pinned reference assets rendered the same way. None of the 28 showcase scripts has a neutral-render mode; they all build the full staged scene in render_still. Retrofitting means either a --neutral flag across 28 scripts or a separate harness that imports each piece's builder and stages it plainly.

Blocked on #200

Do not generate contact sheets from the committed heroes until #200 is resolved. Those heroes are measurably not what the scripts render today — four of five pieces sampled drift 10–22 % of pixels, with the committed file systematically brighter. A contact sheet built from a drifted hero would calibrate against the wrong image, and the calibration band itself (damped-track-aim 0.1722 … bmesh-gear 0.3048) may be measured off drifted files.

Sequence: settle #200, regenerate the stills, then build the 26 contact sheets against known-good images.

Suggested order

1. Resolve Committed hero stills do not match what the scripts render #200 so the stills are trustworthy.
2. Contact sheets for the 26, in batches, reporting the per-criterion verdicts the convention asks for. Expect some to fail and need relighting — that is the gate working, and it is the same fix crate-stack and stone-archway needed.
3. Decide the neutral-render mechanism for asset sheets, then work through the 28.

Steps 2 and 3 are quality-pass work, not a single PR.

[TMHSDigital]

Status update: the #200 sweep partly unblocks this

The full drift sweep (posted on #200, tool in #218) sorts the 28 showcase heroes, compared against the fresh render re-encoded at webp q90:

• 14 match the code (≤ 1.8%): stone-well, wooden-barrel, market-stall, street-lantern, watchtower, cart, water-trough, hitching-post, chopping-block, wooden-bucket, wooden-ladder, hay-bale, crate-stack, stone-archway.
• 14 have drifted (≥ 3.8%): shipping-crate, campfire, treasure-chest, terrain-scatter, fence-kit, park-bench, wheelbarrow, anvil, grindstone, hand-pump, signpost, wall-torch, tavern-stool, iron-cauldron.

Counts have moved since this was opened

docs/gallery/contact-sheets/ now holds sheets for 6 showcase pieces: cart, chopping-block, crate-stack, hay-bale, stone-archway, stone-well. All six are in the matching group.

What is unblocked now

Contact sheets for the 8 matching pieces without one: wooden-barrel, market-stall, street-lantern, watchtower, water-trough, hitching-post, wooden-bucket, wooden-ladder. Their committed heroes are what the code renders, so calibrating against them is sound.

Still blocked on #200

The 14 drifted pieces need their stills regenerated first.

One thing to recheck

The contact-sheet calibration band itself shifts. damped-track-aim (the low end) renders at mean luma 0.1510 from code versus 0.1721 as committed, so sheets should compare against fresh renders of the calibration set, not the committed files.

[TMHSDigital]

Contact sheets: done for all 28 showcase pieces (#225)

• Hero drift: Committed hero stills do not match what the scripts render #200 is closed. fix(gallery): regenerate the 25 drifted heroes from their code, with contact sheets #225 regenerated the 25 drifted heroes from their code, 14 of them showcase pieces, using the new committed scripts/render_hero.py. All 25 now match what their scripts render.
• Sheets: scripts/contact_sheet.py builds sheets in the committed format and prints luma, stage, wedge warmth and saturation per tile. fix(gallery): regenerate the 25 drifted heroes from their code, with contact sheets #225 added or rebuilt sheets for those 14 plus the 8 matching pieces that had none. With the 6 that already existed, 28/28 showcase pieces now have a sheet under docs/gallery/contact-sheets/.
• Verdicts: the per-sheet table is in fix(gallery): regenerate the 25 drifted heroes from their code, with contact sheets #225's PR body.
  • Every showcase sheet sits inside the luma band measured from code (0.151–0.305). wall-torch is the exception, bright by design from its flame.
  • Systemic observation: the showcase props, as a family, are less saturated (0.07–0.18) than the calibration set (0.14–0.24) and often thin in frame. They read as a consistent muted-wood set rather than any one piece being the odd one out. treasure-chest, wooden-ladder, hitching-post and watchtower are the weakest at thumbnail scale.
• Calibration band: re-measured on the regenerated files. damped-track-aim renders at 0.1511 from code, not the 0.1722 it had as committed.

Remaining here: asset sheets (0/28)

Still needs a neutral-render mode (plain three-quarter view, studio light, no staging) across the 28 showcase scripts, or a harness that imports each builder and stages it plainly. The four reference assets need rendering the same way. Leaving this issue open for that step.