# Crown Landmark exterior build C03 (work in progress)

Asset PI-BLD-0001-V01. Exterior-first Blender build aimed at the approved EXT R06 reference imagery and R06 authored targets. **Work in progress, concept candidate only.** Not a shell acceptance, not engine-tested, not placed in the shared world, and it does not change INT R02. Metres, Z up, front = -Y, origin = tower centre at ground.

## What is here

- `snapshots/C03_r4/` - the **frozen assembly C03_r4** (pinned in message QUIK-CLAUDE-20261017-003): scripts as they were, and `FREEZE_C03_r4.json` with SHA-256 of the assembly and every source. The formal INT R02 fit below was run on this pin.
- `blender_scripts/` - the **current round-9 scripts** (C03_r9), pinned by `FREEZE_C03_r9.json`. Five parametric modules in `modules/` (podium, tower, ribs, entrance, crown) assembled by `assemble.py`; `compare.py` (silhouette scores vs references); `entry_ray_check.py` (independent entry-opening ray test); `overlay_int_r02.py` (diagnostic INT R02 room wireframes, read-only, hidden from renders); `sweep.ps1` (tower-profile sweep). `.blend` files are not committed (this repo stores `.blend` through Git LFS); rebuild with the command below. Byte-identical rebuilds are not guaranteed across Blender builds.
- `fit_C03_r4/` - formal C03_r4 vs INT R02 room fit (`SUMMARY.md`, per-room CSV/JSON, C02 regression, options, sheets, scripts). A re-run on C03_r9 follows as a separate linked record.
- `out/assembly/`, `out/compare/` - assembler reports and silhouette scores for r1, r2, r4 and r8; `previews/` - 512 px Workbench JPEGs (not final materials).

## State (2026-10-17)

- **C03_r9:** 104 objects, 141,205 faces = **284,121 triangles**, bounds x +-80 m, y +-60 m, z -1.2 to 240.0 m. Correction: the assembler's report key `triangles` counts faces; true triangle counts come from `qa_check.py` (C03_r4: 146,170 faces = 294,665 triangles). Silhouette IoU (height-normalised, approximate): front 0.90, rear 0.81, left/right (B02) 0.64/0.59, elevated 0.80/0.82. C03_r4 for comparison: front 0.90, rear 0.81, side 0.50-0.55, elevated 0.81/0.83.
- **QA suite (`blender_scripts/qa_check.py`, reports in `out/qa/`):** C03_r9 76 pass, 0 fail, 4 warn (open zero-thickness shells by design, module-local tower glass material, duplicate vertices); contract, R06 targets (bounds, deck footprints 145x106 / 125x90 / 105x76, tower, ribs tips, entry), interference, entry opening and glTF round-trip pass. C03_r4 (pinned): 70 pass, 6 fail, 4 warn; its fails were podium slab footprints (front edge 22 / 5.4 / 5.4 m inside the R06 outline), 1250 loose vertices, 8 degenerate faces and 328 flipped-winding edges, fixed in r9.
- **Entry opening (independent ray test, r8):** 20 of 20 rays along +Y at z = 1.0, 1.7, 3.0, 4.4 m, x = -4..4 m first hit `Entry_lobby_back_glow` at y = -33.25 m; arch 17.96 x 13.99 m, portal 8.02 x 4.49 m. Model check, not engine collision.
- **Interference (lead's BVH test, r8):** ribs vs tower glass 0 overlaps; ribs vs crown 0; ribs vs podium 76 pairs and entrance vs podium 21 (by design: ribs pass through slabs, basins sit in the recess).
- Changes since r4: front ribs R1/R2 are a near-planar A-frame (y lags x), rear ribs concave with tall rear corner fins, podium slabs restored to the full R06 footprints with 0.6 m interior thickness and a 5 m overhang, wall lines recessed (front tiers set back to y = -26 / -30), curved pier flare and taller arches, lobby bay |x| <= 22.3 m cut from tiers 1-2 (documented deviation), mesh defects fixed; crown/tower unchanged from r4.

## Formal C03_r4 vs INT R02 fit (measured; see `fit_C03_r4/SUMMARY.md`)

L25 + L26 (32 rooms): 10 FITS, 2 below the 1.2 m allowance, **20 outside the glass** (worst -14.35 m at L26 corners). L00-L03 (49 rooms): 38 FITS, 1 below, **10 outside** (front rooms ahead of the recessed lobby wall; L02 rooms run 0.9 m into the tier-1 slab, which is 1.5 m thick in C03). Glass 51.0 x 39.6 m at z = 126 vs the 60 x 40 m block. The reviewer's options to clear L25/L26 (all change the silhouette; proposals only) are listed there. C02 regression reproduces 75/5/1. INT R02 was not changed.

## Rebuild

```
"C:\Program Files\Blender Foundation\Blender 5.2\blender.exe" --background --factory-startup --python blender_scripts\assemble.py -- --tag C03_r9 --out out\assembly
```

Needs Blender 5.2.x. `python blender_scripts\compare.py <tag>` (numpy + pillow) scores against `refs/` (copy the EXT R06 images from `production/assets/buildings/PI-BLD-0001-V01/EXT/R06/` into `refs/` first). `CL_FLARE_K`, `CL_TOP_W`, `CL_TOP_D` override the tower profile. Overlay: `blender --background <model.blend> --python blender_scripts\overlay_int_r02.py -- --layout <INTERIOR_LAYOUT.json> [--status-csv fit_C03_r4\fit_C03_r4.csv]`.

## Source of numbers

R06 `DesignTargets.json` (height 240 m, foundation 160 x 120, decks +18/+36/+54, tower 68 x 54 at +54, roof +216, entry arch 18 x 14, portal 8 x 4.5, ribs R1-R6 foot/tip). **Proposed by builders (not R06):** tower top 42 x 32 m with flare exponent 1.8; roof plate 40 x 30 m; podium arch/slab sizes; petal sizes; arch-blade apex (z about 41 m); basin 32 x 18 m; rib path shapes.

## Open conflicts and decisions (Legend's, not resolved here)

- Taper vs the INT R02 L25/L26 room blocks (20 of 32 rooms outside the glass) and the plan shape (superellipse 2.5 vs squarer); the podium front lobby recess vs the INT R02 front atrium rooms; C03 slab thickness vs the 5.4 m L02 rooms.
- R06 rib R3/R4 tip x = +-27 m vs the much narrower crown in the front reference.
- The side references look asymmetric (near-vertical front edge, rear flare); a glass-only offset (`CL_ASYM`, default off) gained at most 0.01-0.03 IoU.

## Known open items

- Rear ribs still stand off the glass at mid height (the left-oblique reference shows a hugging concave wing); rib-footprint table in `podium.py` must be re-measured if `ribs.py` changes.
- Entry arch blades read as a foreshortened bar in the left-front elevated view; they are measured mirror-symmetric (360 vs 360 vertices).
- Interiors not started; exterior shells are hollow-ready thin shells with no slabs, cores or interior walls.

Note: the scripts folder was named `build/` in the working tree; it is `blender_scripts/` here because this repo ignores `build/`. SPEC.md and agent notes still say `CrownLandmark/build`; the scripts locate each other relative to their own folder. Fit scripts in `fit_C03_r4/scripts/` contain absolute scratch paths from the original run.
