# TASK A - doors + horizontal circulation (INT R02 vs shell C02)

Read-only audit. Re-run: `C:/q2/v/Scripts/python.exe C:/q2/out/A_doors/audit_doors.py`.
Inputs: `C:/q2/layout.json`; repo commit 4ce28d8 (`DOOR_SCHEDULE.csv`, `ROOM_SCHEDULE.csv`, access diagram SVG text, MODELING_BRIEF, DesignTargets); SHELL-C02 GLB sliced with the helper from `C:/q2/refit.py` (executed, not edited).
Labels in findings.csv `label`: measured = computed from file values; interpreted = my reading of names/intent; proposed = value the source itself calls proposed. Capsule rows are parameter sweeps (arithmetic on schedule widths), not engine results.

Files: findings.csv / findings.json (267 findings), door_table.csv (per-door metrics), capsule_sweep.csv, graph_edges.csv, shell_checks.csv, audit_doors.py.

## Method / tolerance
- Door "on edge / inside room" tolerance 0.01 m. Observed max deviation of any door centre from its rooms or shared edge: 0.000000 m (data are exact).
- Axis semantics (inferred, consistent for all 102): `axis` = direction the door's wall runs (width spans along it).
- Door span (centre +/- clear_width/2) must lie within the shared edge segment.

## Door validation (measured)
- 102/102 pass: centre on shared edge of from/to rooms, inside both extents, axis matches edge, span within edge, same level, door height < both room clear heights. No room rectangles overlap on any level. No duplicate IDs, no coincident positions.
- Not "room to room" (2): L00-D001 -> token EXTERIOR (portal); L00-D010 -> token PODIUM_BACK_OF_HOUSE. No room geometry exists behind D010 (opens to unmodelled space).
- Same-pair multiples: D086/D087/D088 on every level (3 lift doors RM080 -> RM092, spans do not overlap) - interpreted as the three passenger lift entrances.
- Numbering gap: D089 absent on all 6 levels (intent unknown).
- Tightest end margin on shared edge: L00-D009 0.5 m; all else >= 1.0 m. Tightest lintel headroom: L00-D002/D003/D004 0.4 m (5.0 door in 5.4 room).

## Graph (measured)
- Door edges (90 distinct) == `connections` edges (90 distinct): no one-way listings, none missing either way. DOOR_SCHEDULE.csv == JSON doors in all 10 fields; ROOM_SCHEDULE.csv == JSON rooms (geometry, name, access, connections).
- No orphans. Each level is one connected component (door graph and connections graph).
- Public entrance (interpreted, not ambiguous): L00-D001 portal 8 x 4.5 m into L00-RM001 Arrival atrium; cameras 1-4 and EntranceIn/Out views are there. Second external token = staff/service door L00-D010 (into RM006).
- From the entrance all 16 L00 rooms are reachable. Reachable without passing through a service-access room: 11/16; RM005, RM006, RM007, RM093, RM094 require service rooms (this is the staff route in the brief).
- One-door rooms: every Stair A/B (RM090/RM091) and service lift (RM093) on all 6 levels; L00 RM002, RM004; L01/L02 RM001, RM002 (galleries); L03 RM001; L25 RM002, RM006; L26 RM004, RM005. RM092 (lift shafts) has 3 doors to one neighbour on every level. Whether these are intended terminal rooms is not stated.
- Vertical links are NOT in the data (no cross-level door, no stair/lift flight/cab/landing geometry). Building-wide reachability (81/81) holds only under my interpretation that RM090-093 stack on the six levels (MODELING_BRIEF); L04-L24 have no layout, so L03 -> L25 continuity is unverifiable.

## Capsule / height sweep (not engine results)
- Widths present: 1.2 x5, 1.5 x50, 1.8 x7, 2.4 x5, 3 x2, 4 x29, 8 x4. Narrowest 1.2 m: L00-D006, L25-D006, L25-D007, L25-D008, L26-D004.
- Diameter 0.6 / 0.7 / 0.8 / 1.0 m: 0 doors have clear width below the diameter. Fail only for diameter > 1.2 m. Slack per side at 1.0 m on the five 1.2 m doors: 0.10 m.
- Door heights 2.4 (lift doors, 18) to 5.0 (L00-D002..D004); portal 4.5. Headroom over a 1.8 m / 2.0 m character at the lowest door: 0.6 / 0.4 m. Tallest room 14 m (L00-RM001).
- UNKNOWN: swing direction/hinge side/leaf, thresholds, lift cab inner clearance, stair headroom, true character capsule radius/height - none are in the files, so these were not tested.

## Shell C02 vs exterior doors (measured, slicing of GLB converted to source frame)
- L00-D001 (0,-52), 8 x 4.5 m: Podium_wall_1 (extras group "Podium glazing") front surface is the plane y=-52.4, x -39.2..27.2, z 0.2..17.35, zero thickness, no aperture. The door is 0.40 m behind that plane, not on it. The wall surface covers 95.6% of the door opening rectangle (uncovered only the 0..0.2 m strip below the wall). So C02 has no cut through the wall at the portal.
- Main_entry_clear_18x14 frame: y -56.1..-53.9 (mid-plane -55.0), x +/-10.5, z 0..15.5; 1.9 m in front of the door; portal x +/-4 is inside its +/-9 opening. Entry_glass_canopy underside z=5.0 (0.5 m above portal top), rear edge y=-52.0 equals the door y. Front_amber_atrium_skin lower edge z=4.5, y=-52.6, x +/-8.
- L00-D010 (-38,0): nearest shell segment is Podium_wall_1 at 16.5 m (x=-54.5); the door is not on a shell wall line. D010 destination is unmodelled space.
- No other door touches the exterior. Upper-level (L25/L26) shell fit belongs to another task.
