# Connected sample · schema mapping review

16 October 2026 · Codex sample reviewer · proposed, test-only handoff

## Review baseline and result

Reviewed Claude PR [#51](https://github.com/215GrandAve/lost-souls/pull/51) at **`3a330788a840e6757bd47cee4e0731c08499478f`**, against `SAMPLE_INPUT_PROPOSAL.json` **R02** in this directory. This document maps existing authored values; it is not another dimension contract or permission to modify production sources.

The R02 proposal remains internally coherent. Its known values can be mapped into a deliberately revised test fixture. **Do not fill every null in the existing skeleton and treat that as the mapped sample:** it contains a different Crown entrance/seawall scope, lacks the receiving apron, and leaves the ground-slab/threshold convention unresolved. Claude owns the mapped fixture and any schema/generator changes.

Primary files inspected, pinned to that head:

- [Schema proposal](https://github.com/215GrandAve/lost-souls/blob/3a330788a840e6757bd47cee4e0731c08499478f/tools/contracts/SCHEMA.md)
- [Connected sample skeleton](https://github.com/215GrandAve/lost-souls/blob/3a330788a840e6757bd47cee4e0731c08499478f/tools/contracts/fixtures/SAMPLE_CK_CONNECTED_S01.json)
- [Contract analyser](https://github.com/215GrandAve/lost-souls/blob/3a330788a840e6757bd47cee4e0731c08499478f/tools/contracts/contract.py)
- [Blockout generator](https://github.com/215GrandAve/lost-souls/blob/3a330788a840e6757bd47cee4e0731c08499478f/tools/contracts/blockout.py)
- [Drawing generator](https://github.com/215GrandAve/lost-souls/blob/3a330788a840e6757bd47cee4e0731c08499478f/tools/contracts/draw_svg.py) and [tool README](https://github.com/215GrandAve/lost-souls/blob/3a330788a840e6757bd47cee4e0731c08499478f/tools/contracts/README.md)

R02 source file SHA-256: `94bf2519428400654ec1478785d98f37e12540d1cb1bdf2f31f906dde8866a3e`.

Pinned skeleton content SHA-256: `d8deae0daf2e86e8383a84ac5a27eecd8c13fc5f5f501b53ede5ad0c8b73959f`.

Pinned `contract.py` content SHA-256: `9ad1c2426e0bbdcec3c6530db343f0ba0b2a06629593ba217500cbf75e2de797`.

## Scope changes required before mapping

1. Keep `contract_scope: test_only`, a test identity, null production placement/approval, and a new mapped revision that names R02 and its source hash. The skeleton's `S00-SKELETON` is not a completed sample revision.
2. Label `CL_ENTRY` an **entrance proxy**, not a slice proven to match Crown Landmark. R02 expressly excludes the actual tower/core/lobby geometry.
3. Omit `CL_ARCH` from this mapped proxy fixture. Its carried-over **18 × 14 m** opening cannot fit the proposed **6 × 6 m**, 5 m-clear proxy. Change `CL_PORTAL` from the skeleton's **8 × 4.5 m** to the R02 proxy's **2.4 × 3.2 m**. These are different test inputs, not revisions to the issued Crown entrance dimensions. Replace the inherited feature provenance accordingly.
4. Omit `SEAWALL` and `IF_PROM_EDGE` from this isolated scope. R02 supplies no seawall, coastline, water datum or DS-02/DS-03 choice. Do not invent those dimensions merely to obtain a COMPLETE result.
5. Add a solid receiving-apron mass for R02's fourth element. Rename/relabel `ARC_REAR` as the **north through opening**; use `portal`, not a shopfront-door claim. No door leaf or decorative arch is part of this first rectangular test.

## Coordinate and value mapping

Preserve coordinates numerically. R02's +X right/east and +Y forward/north are local positive directions; schema v2 calls +Y **rear**, with the building front at **−Y**. The proposal's south wall at `y=4` is therefore the schema's front wall. Record this terminology translation explicitly; do not rotate or mirror the geometry to make the words match. The east/north labels do not establish geospatial orientation. Keep `map_transform: null`.

Set `units: metres`; describe the origin as the promenade southwest corner at walking surface `z=0`. Preserve the right-handed, Z-up frame. `frame_revision` may remain unresolved for world placement; name the local sample frame/revision separately in provenance. No Unreal-axis conversion is established by this mapping.

Use these counter-clockwise footprints with the first vertex at southwest. Edge 0 is south, edge 1 east, edge 2 north, edge 3 west.

| R02 source | Schema target | Known value / qualification |
|---|---|---|
| `PROMENADE-PROXY` | `geometry.masses[id=PROMENADE]` | Solid; footprint `[[0,0],[18,0],[18,4],[0,4]]`; `z_base_m=-0.25`, `z_top_m=0`. |
| `ENTRANCE-PROXY` | `geometry.masses[id=CL_ENTRY]` | Hollow; footprint `[[2,4],[8,4],[8,10],[2,10]]`; wall thickness `0.3`; roof top `5.25`. Resolve the wall-base/floor convention below before selecting `z_base_m`. |
| `ARCADE-BAY-PROXY` | `geometry.masses[id=ARCADE_BAY]` | Hollow; footprint `[[10,4],[16,4],[16,10],[10,10]]`; wall thickness `0.3`; roof top `5.25`. Same floor-convention dependency. |
| `NORTH-RECEIVING-APRON-PROXY` | New solid apron mass, identifier chosen by Claude | Footprint `[[10,10],[16,10],[16,12],[10,12]]`; `z_base_m=-0.25`, `z_top_m=0`. |
| Common ground/covered floor | `level_schedule[level_id=G]` | `floor_elevation_m=0`, `usable_clear_height_m=5`, `slab_thickness_m=0.25`; covered clear-height value applies to the two hollow proxies. |
| No next inhabited floor specified | `G.floor_to_floor_m` | Leave null. Do not invent an occupied upper floor or interpret roof top as a second storey. Record why this terminal-level value is unset. |
| Entire sample extent | `envelope.foundation_width/depth` | Derived footprint extent `18 × 12 m`, including the receiving apron. These describe the whole test scene, not Crown Landmark. |
| Highest roof top above datum | `envelope.total_height` | `5.25` under the current analyser's highest-mass-top convention. The complete occupied Z range is `−0.25..5.25`, a `5.5 m` span; do not confuse the two. |
| No production tower/crown | Remaining tower/roof/crown program fields | Keep null with explicit non-applicability/unresolved notes; do not import Crown Landmark envelope values. |

### Real openings

Offsets are measured from the selected CCW edge's first vertex, not from opening centre. All three features are real rectangular through-openings on hollow hosts, at level `G`, with `sill_m=0`.

| Target feature | Host / edge | `offset_m` | Width × height | World-local opening endpoints |
|---|---|---:|---|---|
| `CL_PORTAL` | `CL_ENTRY`, edge 0 | `1.8` | `2.4 × 3.2 m` | `[3.8,4]` to `[6.2,4]` |
| `ARC_FRONT` | `ARCADE_BAY`, edge 0 | `0.6` | `4.8 × 3.6 m` | `[10.6,4]` to `[15.4,4]` |
| `ARC_REAR` (north portal) | `ARCADE_BAY`, edge 2 | `0.6` | `4.8 × 3.6 m` | `[15.4,10]` to `[10.6,10]` |

The north edge runs in decreasing X, which explains the reversed endpoints. No arched profile, glazing or solid-host recess should be substituted. Markers and reservations cannot stand in for these cuts.

## Ground floors and roof: convention needing Claude's confirmation

R02 requires **full outer-footprint** floor slabs at `z=−0.25..0`, including the wall/opening strips. The current `blockout.py` creates hollow-mass floor slabs on `m["inner"]` (line 160), not the outer footprint. With a hollow mass starting at `z_base_m=0`, this leaves a **0.3 m-wide unsupported strip through each ground opening**:

- entrance: `x=3.8..6.2`, `y=4..4.3`;
- arcade south: `x=10.6..15.4`, `y=4..4.3`;
- arcade north: `x=10.6..15.4`, `y=9.7..10`.

Do not claim this mapping is faithful merely because the wall-ray checks pass. Those checks cast horizontal rays above the floor and do not demonstrate continuous support.

Two approaches need explicit tooling-owner review; this document selects neither:

- Make full-footprint slabs/threshold coverage an explicit supported generator convention, preserving separate wall/floor semantics.
- Set hollow `z_base_m=-0.25` deliberately: the uncut wall-base ring below sill `z=0`, combined with the inset floor slab, can represent the same occupied full-footprint floor volume. Confirm that convention in the schema, material/component meaning and generated geometry. The existing hollow `connects` rule treats `z_base_m` as walking elevation, so this choice cannot be combined blindly with hollow-origin connection checks. The known walking datum remains `G=0`.

Adding an extra full-footprint solid slab without addressing automatic inset slabs would create overlapping floor meshes; do not conceal that with a visual-only review.

For roofs, hollow `z_top_m` must mean the **roof top at 5.25**, not the underside at 5. The current generator subtracts the level slab thickness for the inset roof, yielding `5..5.25`; the outer wall ring occupies the perimeter. That union matches the proposed outer roof volume, but component identity and absence of unwanted overlaps/seams still need generated inspection. A mapped value of `z_top_m=5` would produce a `4.75` soffit and fail R02.

## Interfaces and route coverage

The following threshold-centre interface points can be derived from the declared openings without new design choices:

- `IF_PROM_CL`: `[5,4,0]`, connects to `CL_ENTRY`.
- `IF_PROM_ARC`: `[13,4,0]`, connects to `ARCADE_BAY`.
- New arcade/apron interface: `[13,10,0]`; name and explicit relationship chosen by Claude.

Every interface remains `scope: test_only`. The current analyser preserves interface points and the drawing/blockout tools annotate them; **the point itself proves neither touching surfaces nor traversability**. If Claude uses mass `connects` checks, record the source/target direction and the floor-convention dependency above. Solid promenade/apron top elevation is `0`; do not mistake their slab bottoms for walking levels.

Preserve R02's exact walking-surface route:

`[1,2,0] → [5,2,0] → [5,5,0] → [5,2,0] → [13,2,0] → [13,5,0] → [13,9,0] → [13,11,0]`.

There is no documented v2 field or implemented sweep check for a route polyline. Keep the authoritative route in the R02 source and reference it in the test handoff/checklist until Claude proposes a reviewed route representation. Adding an arbitrary JSON key would not make the current tools validate it. This is a ground-contact path, not the capsule-centre Z or a navmesh result.

## Unknowns and acceptance boundaries

| Unresolved input | Mapping / required evidence |
|---|---|
| Capsule radius, half-height, camera clearance | Leave the matching `player_camera_envelope` fields null. Current tooling substitutes pilot `0.35 / 0.9 / 0.2 m` values in derived analysis. These are disclosed assumptions, not Quik's measured character. Any passability result must retain that qualification. |
| Blender/Unreal/importer versions, format/settings | Keep unresolved in the sample's evidence handoff until the actual run. A README's pilot version or glTF recommendation is not a measurement of Quik's environment. |
| Engine axis conversion, import/component transforms | No direct schema mapping establishes them. Keep unresolved; independent asymmetric fixture plus actual engine measurement must confirm scale, handedness and orientation. |
| Collision/navigation/camera behavior | No contract completeness result or straight wall ray supplies these. Quik reports actual method and traversal/camera outcomes with evidence. |
| Accepted measurement tolerance | Leave unresolved until explicitly declared. `contract.py` uses `TOL=0.005 m` internally; that arithmetic epsilon is not an agreed engine/geometry acceptance tolerance. |
| Actual Crown Landmark entrance comparison | Keep unresolved and outside this proxy fixture. The skeleton's inherited entrance numbers and L25/L26 fit results cannot establish local entrance fit. |
| World frame/placement, water datum, geography | Keep null/out of scope; no district transform, shoreline/seawall or foundation construction acceptance follows. |

Use R02 and its hash as the mapped values' provenance. If feature arrays are replaced/reordered, update every JSON-pointer metadata entry; old `/geometry/features/0...` and `/1...` entries must not retain the actual Crown entrance sources for new proxy values. Overlay lists replace wholesale under the proposed RFC 7386 loader; do not assume features are merged by ID. Preserve the mapped contract's own hash plus R02 source hash in the return evidence.

## Evidence obtained and handoff decision

I executed the **pinned analyser in memory against Claude's unchanged skeleton**: result **INCOMPLETE**, no analyser errors, **14 missing-item records**, and disclosed default-capsule assumptions. No second mapped contract was created; no SVG, Blender, GLB, ray test or engine run was performed in this review. The skeleton result confirms only that the incomplete input is reported as incomplete.

**Mapping can proceed as a reviewed test-fixture task once Claude accepts the scope reconciliation and chooses a supported full-floor convention.** It cannot safely proceed as an automatic null-fill, and no sample-export readiness is established here. The independent fixture evidence remains a prerequisite of R02's workflow and needs cross-review at its exact revision.

Claude's return should include: mapped fixture and source hashes; scope/metadata changes; the floor/roof convention; drawings and generated geometry at the same revision; explicit threshold-support and opening checks; retained default/unknown notices; issue-gate refusal evidence; and exact commands/tool versions. Schema COMPLETE, generated-geometry checks and Quik's engine acceptance are separate outcomes. Production issuance remains forbidden for this test-only fixture.
