KaiCao2003 · KaiCao2003 · Jul 26, 2026 · Jul 26, 2026 · Jul 26, 2026 · Jul 26, 2026
diff --git a/.gitignore b/.gitignore
@@ -36,5 +36,4 @@ data-cache/
 *.nii.gz
 *.npy
 *.npz
-!src/mouse_brain_planner/assets/vasculature/lambada_p60_606_major_vessels_v1.npz
-!src/mouse_brain_planner/assets/vasculature/vessap_bl6j1_major_vessels_50um_v1.npz
+*.pdf
diff --git a/ATLAS_DATA.md b/ATLAS_DATA.md
@@ -179,40 +179,15 @@ clearance. The workflow remains archived in the backend and is absent from the p
 50 µm prepared grid does not acquire 25 µm vascular resolution merely because it is composited
 over a 25 µm atlas image.
 
-## Archived LAMBADA major-vessel evidence
-
-The repository retains a derivative of Renier, de Launoit, and Skriabine's
-P60_606 graph, [Zenodo record 18876865](https://zenodo.org/records/18876865), DOI
-`10.5281/zenodo.18876865`, CC BY 4.0. It is bound to the accepted Allen 25 µm array shape and
-contains maximal consecutive in-bounds runs whose point radius is at least 15 µm (diameter at
-least 30 µm). Those deterministic extraction facts do not qualify it as a display layer.
-
-AP and DV orientation evidence passed, but the source describes hemisphere specimens and the
-exact graph has no persisted biological hemisphere/laterality binding. Whole-brain coverage and
-ML polarity are unqualified, and the application does not mirror the source. The production
-reference endpoints never load this derivative. Nothing from it is overlaid in slices, Dorsal,
-or 3D.
-
-The source also omits pial and choroidal vessels, and the derivative omits smaller vessels. Full
-identity, extraction, coordinate qualification, checks, counts, and limitations are in
-[LAMBADA Major Vessels](docs/LAMBADA_MAJOR_VESSELS.md) and
-[Third-Party Software and Data](THIRD_PARTY.md).
-
-## VesSAP display-only major-vessel reference
-
-The current visible reference is a bundled derivative of the public VesSAP `BL6J-no1` 3 µm
-whole-brain skeleton/radius volumes. It retains nominal diameter ≥30 µm source centerlines,
-preserves true source adjacency, applies the authors' Euler plus B-spline Allen transform with
-the reviewed ML reflection, and reduces the paths on a 50 µm display grid.
-
-The adjacent manifest binds the exact source files, transform, atlas identity, extraction
-statistics, NPZ digest, and mandatory limitations. The backend serves metadata and geometry only
-after all checks pass; the same geometry is overlaid in all five modes. It is one fixed cleared
-C57BL/6J specimen, omits smaller vessels, and has no subject-registration or tissue-distortion
-error bound. `vessel.major.reference.analyze` therefore fails with
-`VESSEL_ANALYSIS_UNAVAILABLE` before accessing a project or geometry.
-
-See [VesSAP Major Vessels](docs/VESSAP_MAJOR_VESSELS.md) for the exact data and transform record.
+## Optional VesSAP major-vessel reference
+
+The VesSAP display layer is loaded only when its NPZ and manifest are installed under the
+application data directory. The data files are not stored in this repository or application
+bundle. The loader validates their source identity, transform, atlas binding, digest, arrays,
+bounds, and extraction statistics before exposing the five-view geometry.
+
+See [VesSAP External Major-vessel Data](docs/VESSAP_MAJOR_VESSELS.md) for the file locations and
+data contract.
 
 ## No official CCF bregma
 

diff --git a/COORDINATE_SYSTEMS.md b/COORDINATE_SYSTEMS.md
@@ -376,11 +376,6 @@ An ontology identity does not guarantee voxel or mesh geometry. In the reviewed
 four 2D overlays contain zero selected pixels, and 3D reports that no reviewed geometry exists;
 no inferred surface is substituted.
 
-The archived LAMBADA P60_606 derivative is not displayed. Its exact qualification found
-supporting AP and DV orientation evidence but rejected biological ML/laterality and whole-brain
-coverage because the source is a hemisphere specimen and its graph has no persisted hemisphere
-binding. Numeric points on both sides of the atlas midpoint do not establish bilateral anatomy.
-No mirroring or display-axis guess is permitted.
 
 VesSAP remains one fixed cleared reference, not the current animal. The shared coordinate frame
 permits an overlay but does not supply subject registration, tissue-distortion, or inter-animal

diff --git a/DEVELOPMENT.md b/DEVELOPMENT.md
@@ -1,140 +1,47 @@
 # Development
 
-Brain3D has one supported product path: a SwiftUI + SceneKit macOS application and a Python 3.12
-scientific service connected by typed NDJSON. Scientific transforms, atlas access, calibration,
-probe/region analysis, vessel qualification, and persistence stay in Python; Swift owns
-presentation, input, accessibility, native file handling, and schema-checked SceneKit display
-payloads. No vessel-analysis capability is currently exposed.
-
-Start with [Architecture](docs/ARCHITECTURE.md), [Code Audit](docs/CODE_AUDIT.md),
-[ADR-005](docs/ADR-005-independent-slice-viewer.md), and
-[Scientific Validation](SCIENTIFIC_VALIDATION.md).
-
-## Environment
-
-- Apple Silicon (`arm64`)
-- macOS 14 or later
-- CPython `>=3.12,<3.13`
-- Swift 6-compatible toolchain
-- committed `uv.lock`
+## Setup
 
 ```bash
 uv python install 3.12
-uv lock --check
 uv sync --frozen --group dev
+uv run --frozen mouse-brain-planner atlas download allen_mouse_25um
+```
+
+Build and open the macOS app:
+
+```bash
+native/Brain3D/Scripts/build-app.sh
+open native/Brain3D/build/Brain3D.app
 ```
 
-The default dependency graph deliberately excludes PySide6, PyVista, PyVistaQt, and VTK. Do not
-reintroduce a second GUI or renderer into the base environment.
+The Swift app communicates with the Python service over typed NDJSON. Start with
+[Architecture](docs/ARCHITECTURE.md), [Coordinate Systems](COORDINATE_SYSTEMS.md), and the
+architecture decisions in [docs](docs/).
 
-## Quality gate
+## Checks
 
 ```bash
 uv lock --check
-uv sync --frozen --group dev
 uv run --frozen ruff format --check .
 uv run --frozen ruff check .
 uv run --frozen mypy --no-incremental
 uv run --frozen pytest -q
-.venv/bin/python scripts/verify_minimal_runtime.py
-swift test --package-path native/Brain3D
+swift test --package-path native/Brain3D --no-parallel
 native/Brain3D/Scripts/build-app.sh
 codesign --verify --deep --strict native/Brain3D/build/Brain3D.app
 ```
 
-The minimal verifier checks that Qt/VTK modules are absent and exchanges real `hello` and
-`shutdown` frames through `mouse-brain-planner bridge`. A green gate establishes engineering
-consistency, not stereotaxic, vascular-clearance, usability, or surgical accuracy.
-
-Useful focused runs:
-
-```bash
-uv run --frozen pytest tests/unit/test_atlas_space.py -q
-uv run --frozen pytest tests/unit/test_brainglobe_adapter.py -q
-uv run --frozen pytest tests/unit/test_slice_renderer.py -q
-uv run --frozen pytest tests/integration/test_bridge_stdio.py -q
-swift test --package-path native/Brain3D
-```
-
-The normal Python suite uses synthetic volumes and local fakes. The real density-archive and
-real-atlas checks remain explicit opt-in tests because their external data are not committed.
-
-## Run the product path
-
-```bash
-native/Brain3D/Scripts/build-app.sh
-open native/Brain3D/build/Brain3D.app
-uv run --frozen mouse-brain-planner atlas list
-uv run --frozen mouse-brain-planner atlas download allen_mouse_25um
-uv run --frozen mouse-brain-planner validate-project /absolute/path/Plan.mouseplan
-```
-
-`mouse-brain-planner` without a subcommand prints help. `mouse-brain-planner bridge` is intended
-for the native process boundary and integration diagnostics.
-
-## Source layout
+## Repository layout
 
 ```text
-native/Brain3D/               SwiftUI shell, typed client, native tests/build
-  Sources/Brain3DScene/       SceneKit brain/probe rendering; archived vessel primitives
-src/mouse_brain_planner/
-  atlas/                      BrainGlobe 2.3.1 boundary
-  bridge/                     versioned service and handlers
-  coordinates/                named frames, order conversion, transforms
-  domain/                     Pydantic scientific/project models
-  persistence/                deterministic migrations and atomic packages
-  rendering/                  raster slices and dorsal projection
-  probes/                     source-traceable NP1 and synthetic test catalog
-  surgery/                    product-reachable probe/measurement geometry
-  analysis/                   region traversal plus archived vessel-analysis algorithms
-  vasculature/                VesSAP display geometry plus archived qualification workflows
-tests/                        headless unit/contract/integration tests
-docs/                         audit, architecture, validation, and ADRs
+native/Brain3D/                 SwiftUI and SceneKit application
+src/mouse_brain_planner/       Python atlas and planning service
+tests/                          Python tests and fixtures
+docs/                           architecture and technical references
+scripts/                        verification and benchmark utilities
 ```
 
-The old Qt/PyVista/VTK application was removed after the audit in
-[Legacy Removal](docs/LEGACY_REMOVAL.md). Its last complete snapshot is Git commit `51fe26d`.
-
-## Dependency direction
-
-- Swift never reimplements coordinate transforms or scientific classification.
-- The bridge imports no macOS UI framework and must run headlessly.
-- Only the BrainGlobe adapter owns upstream atlas objects.
-- All API coordinates name their frame, order, units, origin, directions, atlas identity, and
-  transform/calibration identity.
-- Project files store model state and immutable data references, never UI objects or complete
-  atlas/vessel volumes.
-- The rejected LAMBADA derivative, archived population density, and archived 2D subject evidence
-  remain separate types and workflows. None is a subject-vessel layer. The runtime vessel
-  geometry endpoint serves only the digest-checked VesSAP display reference; analysis fails
-  closed with `VESSEL_ANALYSIS_UNAVAILABLE`, and LAMBADA is never loaded.
-
-## Coordinate-change checklist
-
-Any change involving axes, units, indexing, laterality, transforms, picking, or projection must:
-
-1. name the source and destination frames and units;
-2. use centralized BrainGlobe `[AP,DV,ML]` ↔ domain `[AP,ML,DV]` conversion;
-3. validate finite values and half-open atlas bounds;
-4. distinguish continuous voxel coordinates, discrete indices, corner anchors, and centers;
-5. add asymmetric laterality and forward/inverse golden tests;
-6. require either the exact source-pinned v4 bregma/surface evidence or, for legacy v1–v3,
-   a passing subject calibration; never treat atlas origin as bregma; and
-7. update coordinate/validation documentation with exact evidence and tolerances.
-
-## Bridge-change checklist
-
-- Preserve strict envelope and exact-parameter validation.
-- Reject bool where a number is expected, plus NaN and infinity.
-- Use stable error codes and include recoverability/suggested action in protocol v2 errors.
-- Every mutation checks the expected project revision and returns the new revision.
-- Long work runs off the UI thread and stale results cannot overwrite current state.
-- Add Python contract tests, matching Swift Codable tests, and a production handler/capability
-  inventory assertion.
-- Do not send full atlas volumes or platform-specific index objects across NDJSON.
-
-## Review discipline
-
-Before committing, inspect `git diff --check`, run the relevant focused tests, then run the full
-gate in proportion to the change. Never weaken frame/provenance validation to make a test pass.
-Update [Code Audit](docs/CODE_AUDIT.md) when a previously orphaned module becomes product-reachable.
+Atlas volumes, vascular datasets, PDF inputs, and user projects stay outside source control.
+Use `MOUSE_BRAIN_PLANNER_CONFIG_DIR`, `MOUSE_BRAIN_PLANNER_DATA_DIR`, and
+`MOUSE_BRAIN_PLANNER_CACHE_DIR` for isolated development environments.