# Task D: collision readiness and mesh integrity, Crown Landmark SHELL C02

Read-only audit of `SHELL-C02/MODEL/PI-BLD-0001-V01_CrownLandmark_SHELL_C02.glb` (162 mesh nodes, 61,556 triangles, no node transforms). Source frame used throughout: x right, y back, z up, front = -y (GLB converted with (x, -z, y)). No Blender or Unreal was used. No package was installed (trimesh was not needed; numpy + shapely only, own ray caster). Labels: **measured**, **interpreted**, **not_testable**. All file text was treated as data.

Machine-readable: `findings.csv/json` (39 labelled findings), `part_integrity.csv/json` (162 rows), `group_integrity.json`, `walkable_rays.csv`, `walkable_summary.json`, `horizontal_surfaces.csv`, `door_rays.csv`, `arch_openings.csv`, `entry_results.json`, `interpenetration_pairs.csv/json`, `hull_and_misc.json`. Scripts: `load.py rc.py d2_integrity.py d3_walk.py d3b_horiz.py d4_doors.py d4b_entry.py d5_overlap.py make_findings.py`.

## 1. Naming / collision metadata (measured)

- No UCX_, UBX_, UCP_ or USP_ parts, in GLB or OBJ. No node named collision/walk. No glTF extensions at all.
- GLB extras: scene level `asset_id, revision, status, source_units, glTF_units, source_to_gltf, map_transform, saved_cameras_source_m`. Every mesh and node: `group` (10 values) and, on meshes, `review_status: CONCEPT`. Nothing marks walkable vs decorative.
- Materials: 9, all opaque (alpha 1, no alphaMode) and all `doubleSided: true`. Glass-like parts (podium_glass x3, blue x8, parapet x4, amber x1, water x6) have no transparency flag.
- README.md and MODEL_SPEC.json both say collision is unfinished and glazing is an opaque review skin.

## 2. Per-part integrity (measured; weld 1e-5; full per-part table in `part_integrity.csv`)

| Group | Parts | Closed | Open | Boundary edges | Note |
|---|---|---|---|---|---|
| Foundation | 1 | 1 | 0 | 0 | solid 1.18 m |
| Podium terraces | 12 | 3 | 9 | 932 | Decks closed (0.59 m slabs); 3 fascia, 3 parapet open; 3 coping non-manifold |
| Podium glazing | 3 | 0 | 3 | 40 | Podium_wall_1/2/3 = 16/16/8 triangles, zero-thickness rings |
| Podium arcades | 43 | 43 | 0 | 0 | solid frames, 2.2 m deep |
| Tower glazing | 2 | 1 | 1 | 152 | skin open (8,360 tris, open top and bottom); roof closed 0.39 m |
| Facade grid | 62 | 26 | 36 | 5,472 | 26 mullions closed; 36 storey bands are open 0.24 m tall sheets |
| Ribs and crown | 19 | 13 | 6 | 18 | Crown_glass = single triangles |
| Front atrium | 3 | 2 | 1 | 120 | amber skin is a single flat strip |
| Rear authority motif | 2 | 2 | 0 | 0 | |
| Entrance and basins | 15 | 14 | 1 | 44 | canopy open |
| **Total** | **162** | **105** | **57** | **6,778** | |

- Winding: 0 inconsistent edges in all 162 parts. Closed parts: all 105 have positive signed volume (outward); 0 inverted.
- 0 degenerate triangles, 0 slivers under 1e-4 m2, 0 unreferenced vertices, 1 connected component per part.
- Non-manifold: only Terrace_1/2/3_coping (2 edges each). Duplicate vertices: 9, same 3 parts.
- Thin-wall status: Podium_wall, Tower skin, bands, fascia, parapets, amber skin, canopy are zero-thickness single surfaces (no solid behind them). Wall/skin/band normals face outward; fascia normals face up; canopy normals face down.

## 3. Walkable surface inventory (measured)

Up-facing horizontal surfaces of 20 m2 or more exist only at: z=0.0 Foundation; 0.55/0.59 Basin rim/water; 18.0 Deck_1; 36.0 Deck_2; 54.0 Deck_3; 216.0 Occupied_roof. Ray down from each room centre and four corners +1.5 m inward, starting 1.5 m above the level:

| Level | z | Points | With surface at level z | First hit when none |
|---|---|---|---|---|
| L00 | 0 | 80 | 80 (Foundation, front-facing) | n/a |
| L01 | 6 | 55 | 0 | Foundation top z=0.0 |
| L02 | 12 | 55 | 0 | Foundation top z=0.0 |
| L03 | 18 | 55 | 55 (Deck_1_top_18m, front-facing) | n/a |
| L25 | 126 | 75 | 0 | Deck_3 top z=54.0 (72 m down) |
| L26 | 130.5 | 85 | 0 | Deck_3 top z=54.0 (76.5 m down) |

So only deck tops (18/36/54), the roof (216) and the foundation are floors. There is no slab at L01, L02, L25 or L26 (layout notes also say the L01/L02 atrium slabs are intentionally omitted, but the non-atrium rooms are unsupported too). Roof edge sits about 2.4 m inside the skin at y=0 (rays fall through to Deck_3).

## 4. Openings (measured)

- Interior walls do not exist in the GLB. Of 102 layout doors, rays at +1.0 and +1.7 m (+-6 m along the passage) hit geometry for 1 door only: L00-D001 (portal, 8 x 4.5 m), blocked 34/34 by Podium_wall_1. The other 101 doors have nothing to test.
- Main entry: `Main_entry_clear_18x14` is a closed solid arch frame (y -56.1..-53.9). The frame hole is real: ray grid 0.25 m passes x -8.75..8.75 (about 18 m) and at x=0 is free from z=0.05 to about 14 m.
- But `Podium_wall_1` has no hole: it is a continuous zero-thickness wall at y=-52.4 (x -39.2..27.2, z 0.2..17.35), 1.5 m behind the frame. 3,657 of 3,726 rays in the opening hit it; the other 69 are the z=0.05 row passing through a 0.2 m slit under the wall. Clear gap at 1.0 and 1.7 m height: 0 m wide, 0 m high. The entry is a marker/frame in front of a closed wall.
- All 43 arcade frames (42 T*_arch + Main entry): frame holes are open (10.5-18 m), but 0/43 have a through-path; every ray is stopped by Podium_wall_1/2/3 (34) or a rib/amber skin/gold tube (9).
- Walls hover: bottom 0.2 m above the surface below, top 0.05 m under the deck soffit.

## 5. Interpenetration (measured; 249 z slices; even-odd filled outlines; closed parts = solid, open parts = lines)

Top 15 solid-vs-solid overlaps by maximum slice area (45 pairs total, all involve a Rib or Basin; full list in `interpenetration_pairs.csv`):

| # | Pair | Max area m2 | at z | Approx. volume m3 |
|---|---|---|---|---|
| 1 | Deck_1_top_18m / Rib_R2 | 71.6 | 17.5 | 78.6 |
| 2 | Deck_1_top_18m / Rib_R1 | 70.7 | 17.5 | 77.6 |
| 3 | Deck_1_top_18m / Rib_R3 | 67.8 | 17.5 | 74.5 |
| 4 | Deck_1_top_18m / Rib_R4 | 67.6 | 17.5 | 74.3 |
| 5 | Deck_1_top_18m / Rib_R5 | 64.2 | 17.5 | 70.5 |
| 6 | Deck_1_top_18m / Rib_R6 | 64.2 | 17.5 | 70.5 |
| 7 | Deck_2_top_36m / Rib_R2 | 49.9 | 35.5 | 54.8 |
| 8 | Deck_2_top_36m / Rib_R1 | 49.3 | 35.5 | 54.1 |
| 9 | Deck_2_top_36m / Rib_R3 | 48.1 | 35.5 | 52.8 |
| 10 | Deck_2_top_36m / Rib_R4 | 48.0 | 35.5 | 52.7 |
| 11 | Deck_2_top_36m / Rib_R5 | 46.1 | 35.5 | 50.6 |
| 12 | Deck_2_top_36m / Rib_R6 | 46.0 | 35.5 | 50.5 |
| 13 | Rib_R5 / Basin_rim_R5 | 40.1 | 0.5 | 22.4 |
| 14 | Rib_R6 / Basin_rim_R6 | 40.1 | 0.5 | 22.4 |
| 15 | Rib_R3 / Basin_rim_R3 | 38.0 | 0.5 | 21.7 |

Also: Deck_3 vs Rib 30.7-31.9 m2; Rib vs Basin_water 33-36 m2; T1/T2 left/right/rear/front arcades vs Rib up to 11.9 m2 (T1_left_arch_2/Rib_R3). Shell-through-solid (20 pairs, wall line length inside solid): Podium_wall_1 / Rib_R3 and R4 14.0 m; Podium_wall_1 / Rib_R1, R2 9.2 m; Podium_wall_2 / Rib_R1 7.1 m; Tower skin / Rib_R1, R2 4.4-4.5 m over 118 slices; Podium_wall_1 / T1_front_arch_1 2.2 m. No overlap between walls and decks/fascias; no shell-shell crossings in scope. Limit: z-step 1 m, so thin slabs may be under-sampled.

## 6. Interpretation of a default engine import (INTERPRETED, UNTESTED)

Rests on: no collision meshes or flags (1); 57 open zero-thickness parts, outward normals, doubleSided materials (2); floors only at 0/18/36/54/216 (3); Podium_wall_1 unbroken at the entry (4); ribs buried in decks, walls and skin (5).

- Complex-as-simple (render triangles used as collision): 61,556 triangles from all 162 parts, including mullions, bands and gold tubes. Walkable at z=0, 18, 36, 54. A pawn in L01/L02 rooms falls to z=0; in L25/L26 falls 72-76.5 m to Deck_3. The entry (and portal door) is blocked by Podium_wall_1. Zero-thickness walls/skin block from the outside; from the inside it depends on the double-sided collision setting.
- Auto convex decomposition: decks and ribs behave as solids. Arcade/entry frames would be approximated by convex pieces; the convex hull of Main_entry_clear_18x14 is 605 m3 against 133 m3 of solid (arcades 3.8-4.7x), so the frame hole would likely be filled. Open walls and skin have no volume; hull volumes of the wall vertices are 157,025 / 99,641 / 47,334 m3 (vs 6,737 / 5,502 / 3,739 m2 of surface), so a hull-based result would be a degenerate or a solid block, depending on the algorithm.

## 7. Not testable without an engine

- Real collision result of either import mode (hull count, cook success, one-sided vs two-sided behaviour of the zero-thickness walls and skin).
- Capsule behaviour: step height, slopes, the 0.05 m and 0.2 m gaps against capsule radius, trace channels and presets.
- Importer response to the 3 coping parts with non-manifold edges and 9 duplicate vertices; self-intersection inside a single part (not measured).
- Navigation mesh, physics materials, LOD collision (no data in files).
