# Quik / local AI · R06 start here

**Goal:** inspect a useful isolated metric connection specimen and return evidence.
**Status:** TEST ONLY; no actual SW/Crown placement, regional scale or engine acceptance.

Read [REGIONAL_GATES.json](REGIONAL_GATES.json) before proposing regional
integration. It lists the eight open decisions/evidence gates for regional
release. Publication and an isolated specimen import do not pass those gates.

## Preserve current work

Use your existing branch/editor session. Read current repo decisions, protocols,
both communications lanes and pending replies before integration. Do not move,
rescale, replace or merge existing D2/D3 assets, water or world configuration.
This package authorizes an isolated test collection only.

## Portable preflight

Extract `SW-R06-GEOMETRY.zip`. Keep all files together. From its folder run:

```bash
python IMPORT_BLENDER.py --validate-only --package .
```

This verifies canonical JSON checksum, scope, units and mesh indices without opening
Blender or changing a scene. Full provenance/topology validation uses the repository
commands in `README.md`.

## Optional Blender staging

In a separate test scene/file, run:

```bash
blender --background --python IMPORT_BLENDER.py -- --package /absolute/path/SW-R06-GEOMETRY
```

The script validates before creating objects, aborts on an existing test collection,
and attaches `TEST_ONLY_SW_R06_DO_NOT_MERGE` only after all objects exist. Failure
cleans newly created data. Existing objects and scene settings stay untouched.
Coordinates are metre values; the script does not change scene unit scale.
Scene `unit_settings.scale_length` must already be 1 (one metre per raw Blender
unit); a mismatch aborts before any Blender data mutation. The script records
the existing unit system, length unit and scale in the new test collection. Save
only to your own isolated test file. Script syntax/preflight was checked; actual
Blender execution was not available in the generating environment.

## Unreal / importer controls

Use the four `axis_controls` first. Source metres are RH `(x,y,z)`. The proposed
staging interface is centimetres `(100*x,-100*y,100*z)`. Expected controls before
world R/T: O `(0,0,0)`; X `(10000,0,0)`; Y `(0,-10000,0)`; Z `(0,0,1000)`.
Confirm the actual FBX/GLB/importer path before applying a Y flip; never apply it
twice. If reflecting mesh vertices directly, reverse each triangle winding once.
The OBJ and Blender objects have no reflection or centimetre conversion applied.
Authoritative world R and T are unresolved: only proper rotation and translation,
no scale/shear, after real controls/datum reconciliation. Do not edit engine or
World Partition settings to make this specimen appear accepted.

## Return evidence

- Input artifact hashes, branch/head and engine/importer versions.
- The four actual axis control positions, roundtrip dimensions and a handedness/winding check.
- A measured 100 m opening and 20 m gap relative to the same declared water plane.
- Road flat portal z=2 m, span z=22 m and maximum 5% analytic approach.
- Both pedestrian fold portals and route continuity, with actual walking/collision evidence.
- Saved cameras for top/side/isometric views and clean/labelled pairs from that same test model.
- Explicit `pass`, `fail`, `blocked` or `not_tested` per check; no inferred engine success.

Publish a new append-only record in Quik's lane. A receipt or successful import is
not regional geometry, structural, accessibility, building-shell or creator acceptance.
Use `QUIK_RETURN_TEMPLATE.json` to return measured results; every initial state is
`not_tested`, and expected positions are kept separate from actual measurements.
