Re-plan the Blender tracks around the local skills and a 3D model review

AGENTS.md

@@ -28,7 +28,8 @@ The project should be developed in small, reviewable steps. Agents must keep pla
 
 `TODO.md` is the priority source. The active order is:
 
-1. unversioned development tracks (currently character development and local map development);
+1. unversioned development tracks (currently 3D model re-evaluation, character development and
+   local map development);
 2. versioned release milestones, in version order;
 3. future/unversioned candidates and deferred work.
 
@@ -36,6 +37,65 @@ Do not assign a version number to an unversioned track until its scope is accept
 release. When a version is assigned, it must be the next semantic version that matches the
 scope (major/minor/patch), not a placeholder.
 
+## Local Blender skills (development aid)
+
+The MIT `blender-skills` pack is installed locally at `.agents/skills/blender-skills/`
+(94 skills plus a shared `references/` folder). It is **gitignored on purpose**: it is not
+game content, not part of the build, tests, CI or release packages, and every contributor
+may or may not have it. Pi gates project `.agents/skills` behind project trust, so the
+trust prompt has to be accepted before the skills load; `/skill:blender-director` forces one.
+
+Use them when a task touches `tools/blender_assets.py`, `tools/npc_pipeline.py`,
+`tools/bake_ao.py`, `assets/models/blender/**`, or Blender-authored venue/prop geometry.
+
+### Routing
+
+Start with `blender-director` for anything multi-discipline, then load only the
+specialists the task needs:
+
+| Work | Skills |
+| --- | --- |
+| Plan/route a Blender task, reference photo match | `blender-director` |
+| Character mesh, face, hands, proportions | `character-artist`, `character-archetypes`, `retopology` |
+| Skeleton, weights, pose/contract rigging | `rigging` |
+| UV layout, texel density, UDIM | `uv-workflow` |
+| Bakery, atlas, normal/AO/ORM maps | `texture-workflow` |
+| PBR materials and shading | `materials` |
+| Screenshot-driven material/light iteration | `lookdev` |
+| Triangle/vertex budgets, naming, scene hygiene | `asset-optimization` |
+| LOD tiers and impostors | `lod-pipeline` |
+| Collision shapes and proxies | `collision-proxy` |
+| GLB export for Godot | `godot-export`, `export-pipeline` |
+| Gear, props, street furniture | `prop-artist`, `hard-surface` |
+| Trees, hedges, foliage cards | `vegetation-artist` |
+| Facades, venues, modular kits, dressing | `environment-artist`, `archviz`, `set-dressing`, `geometry-nodes` |
+| Lighting, framing, final render/composite | `lighting`, `camera-cinematography`, `rendering`, `compositing` |
+| Style direction (stylized / low-poly / realistic) | `stylized-style`, `lowpoly-style`, `realistic-style` |
+| Ship/no-ship review | `qa-review` |
+
+The same skills cover the local map track (procedural geometry, facade kits, OSM massing,
+splat reference renders) and the NPC roster work.
+
+### Project rules that override a skill
+
+1. **The project contract wins.** `npc_base` mesh node, the 13 contract bones at their rest
+   positions, five tint slots, front on Blender −Y / Godot +Z, grounded feet, 8,000
+   triangles / 8,000 glTF vertices per character, 1,500 per gear, 4,000 per prop. A skill
+   may recommend another topology, naming or budget scheme; do not adopt it unless the
+   change is explicitly planned and the contract, tests and docs are updated together.
+2. **There is no Blender MCP here.** The skills assume MCP tools; this project runs Blender
+   as headless flatpak scripts (`tools/blender_assets.py`, `tools/npc_pipeline.py`). Translate
+   MCP steps into script changes or CLI runs. Never add an MCP dependency to the build, CI or
+   runtime.
+3. **CI stays offline and skill-free.** Skills are local guidance only; every gate must still
+   work from a fresh checkout with committed `.glb`/`.png`/`.tres` files and Godot.
+4. **No new asset source without the licence gate.** Skills may suggest reference packs or
+   generated meshes; ADR-0015 (CC0, or CC-BY with attribution in `CREDITS.md`) and the
+   provenance rules in `docs/npc-model-sourcing.md` still apply.
+5. **Evidence, not vibes.** After a skill-informed change, run `blender_assets.py inspect`,
+   `audit` and `fit` plus the relevant capture/smoke gates, and say in the handoff which
+   skills informed the change and what the measurements were.
+
 ## Required Workflow
 
 ### 1. Start by Inspecting the Repository
@@ -268,6 +328,8 @@ Before handing off work, an agent must run every check below that applies to the
 - Run automated tests.
 - Run formatting, linting, or type checks.
 - Start the game or application and perform a basic smoke test.
+- For 3D/Blender changes: `blender_assets.py inspect`, `audit` and `fit`, the relevant capture
+  gate, and a note of which local skills informed the change.
 - Check that documentation and task status reflect the actual state.
 - Review the final diff for accidental changes or unfinished markers.
 

MEMORY.md

@@ -4,6 +4,14 @@ Session memory for durable decisions, state, and gotchas. Keep concise, remove o
 
 ## Decisions
 
+- **3D model re-evaluation first (2026-11-03):** `TODO.md` is re-planned around a new unversioned
+  `3D model re-evaluation` track, ahead of character and local-map work, because the crowd-variant
+  roster would otherwise multiply whatever the existing models get wrong. It scores all 8
+  characters, 18 gear, 20+6 props and the three venues in `docs/3d-model-review.md`, fixes ranked
+  P0/P1 only, and must not change the `npc_base` contract, renderer or 0.10.0 candidate. The MIT
+  Blender skill pack is now routed through `AGENTS.md` §Local Blender skills; skill advice loses
+  to the project contract, and there is no Blender MCP here (headless flatpak scripts instead).
+  MEMORY is ~960 lines with dated historical entries; trimming it is a tracked maintenance task.
 - **GLB intake inspection (`blender_assets.py inspect`, 2026-11-03):** runs inside Blender, clears
   parents and applies object transforms, then reports tris/budget, raw and seam-split verts,
   Godot-space bounds (height/feet for characters), bones, materials, UVs, skinning and
@@ -356,11 +364,11 @@ Session memory for durable decisions, state, and gotchas. Keep concise, remove o
   **Mockup copies:** 21 design references live in `assets/images/ui/`; `docs/screens/designs/`
   remains the reference-only copy. `tools/screen_designs.py` reads/writes the runtime root.
 
-## Handoff (read this first)
+## Handoff — historical snapshot (2026-11-01; superseded by Current State / Plan vs code)
 
-- **Status (2026-11-01, after the venue-geometry pass):** v0.7.0 is released. Post-v0.6.0 the
+- **Status (2026-11-01, after the venue-geometry pass):** v0.7.0 was released at that point. Post-v0.6.0 the
   realism pass, the visual-alignment pass and the NPC face/hand pass all landed. Headless
-  **229/229**, Python tooling **87/87**, `tools/check_screenshots.sh` **11 good / 0 bad**,
+  **272/272**, Python tooling **108/108**, `tools/check_screenshots.sh` **21 good / 0 bad**,
   `screen_designs.py compare` 11 good / 0 bad, `scene_concepts.py audit` all GOOD,
   `blender_assets.py audit` exit 0, `smoke_npc_budget` OK (16 avatars, 6.94 ms of 50 ms).
   `PLAN.md` is compact; README/architecture track the current controls, asset pipeline, CI and
@@ -914,13 +922,14 @@ Session memory for durable decisions, state, and gotchas. Keep concise, remove o
 - Shift length: `GameManager` uses the scenario's `shift_duration_s` (900s ≈ 10-15 min);
   tests and the screenshot tool override elapsed time directly.
 - Headless (dummy renderer) logs "Parameter m is null" from `mesh_get_surface_count`; it does not occur with a real GL driver and does not affect logic.
-- Steps are visual-only (no collider) so NPCs can path to the entrance; the guard spawns on the landing so it never needs to climb.
+- Steps have per-tread colliders (`_build_steps()` adds a `StaticBody3D` with one `BoxShape3D`
+  per tread), so the guard and NPCs walk up the entrance; the guard spawns on the landing.
 - Persistence tests write into the real `user://shift_results.json` (harmless local pollution of the demo history).
 - macOS export required `rendering/textures/vram_compression/import_etc2_astc=true` in `project.godot` (universal build demands ETC2 ASTC enabled).
 - Venue `door_staff_lines` are logged through `GameManager` (inspect / wand / refuse). They are not a separate dialogue UI.
 - Access loadout is implemented; do not re-fix `override_access` as if it were still unread.
 - `Save.SETTINGS_FILE` is declared and unused; settings persistence goes through `Save.load_settings()` / `save_settings()`.
-- Headless suite is **272/272**; Python tooling is **101/101**; screenshot compare is **21 good / 0 bad** and the automatic judge is **21/21 match, 0 diff**. Remaining visual follow-ups are in `TODO.md`.
+- Headless suite is **272/272**; Python tooling is **108/108**; screenshot compare is **21 good / 0 bad** and the automatic judge is **21/21 match, 0 diff**. Remaining visual follow-ups are in `TODO.md`.
 - Typed spawn signal extra args: handlers must accept the 4th `event: Dictionary` or Godot will not call them (`test_shift_coordinator_runs_a_shift` failed until `_on_spawn` took four args).
 - GDScript inference trap that broke the main scene: a `var` typed as a base class
   (`radial: CanvasLayer`) exposes script members as `Variant`, so `var x := radial.is_open` is a

PLAN.md

@@ -170,7 +170,7 @@ The action wheel prioritizes de-escalation: Stop, Look, Point, No, Ask, Inspect,
 The visual direction is readable stylized realism:
 
 - **Character direction:** 8 skinned NPC variants, 18 gear pieces and 20 street/venue props, with four crowd-variety variants prototyped. The generator-first expansion is the active character track ([ADR-0015](docs/adr/0015-npc-model-variety-sourcing.md)); Quaternius is a reference and MakeHuman is deferred to a local pipeline.
-- **Realism pass:** AI-assisted per-phase skyline panoramas, v2 surface tiles with normal maps, wet asphalt at night, a warm day haze, phase-specific sun/exposure/fog, a visible flashlight beam, and denser foliage.
+- **Realism pass:** AI-assisted per-phase skyline panoramas, v2 surface tiles with normal maps, wet asphalt at night, a warm day haze, phase-specific sun/exposure/fog, a tight flashlight spot (no beam geometry), and denser foliage.
 - **Local map direction:** the venues are currently stylized reconstructions. The local-map track evaluates licensed OpenStreetMap/3D building data, licensed reference photography, AI-assisted composition and Gaussian splatting as a development tool; online data and splat rendering are not runtime dependencies.
 - **References and QA:** runtime UI art lives in `assets/images/ui/`; reference/art-direction sets live in `docs/screens/designs/` and `docs/screens/concepts/`; captures live in `docs/screenshots/`; realism runs live in `docs/screens/realism/`. `screen_designs.py compare/judge` and `tools/check_screenshots.sh` guard drift.
 - **Storage convention:** see [ADR-0014](docs/adr/0014-runtime-image-root.md). Never overwrite an existing image; new outputs get a new path or `-vN` suffix.

TODO.md

@@ -3,16 +3,119 @@
 Work is grouped by theme. Each item should be small enough to complete and verify on its own;
 see `AGENTS.md` for the workflow and `MEMORY.md` for the decisions behind these plans.
 
-**Priority order:** Character development → Local map development → versioned release milestones →
-future/deferred work. See `AGENTS.md` §Priority and Versioning.
+**Priority order:** 3D model re-evaluation → character development → local map development →
+versioned release milestones (0.10.0) → future/deferred work. See `AGENTS.md` §Priority and
+Versioning and §Local Blender skills.
+
+The MIT Blender skill pack lives locally at `.agents/skills/blender-skills/` (gitignored). Every
+3D workstream below names the skills to load; they are guidance only, and the project contract,
+`blender_assets.py audit`/`fit`/`inspect` and the capture gates stay authoritative.
+
+## 3D model re-evaluation
+
+Goal: re-judge every shipped 3D asset against the newly available Blender skill guidance, rank
+the real gaps, and fix only what the rankings justify. This is an unversioned development track:
+it must not change the `npc_base` contract, the renderer or the 0.10.0 release candidate.
+
+Scope: 8 library characters (`npc_casual`, `npc_jacket`, `npc_coat`, `npc_vest`, `npc_dress`,
+`npc_hoodie`, `npc_firefighter`, `npc_guard`), the frozen `npc_base` fallback, 18 gear pieces,
+20 library props, 6 generated props, and the three venue families.
+
+**Exit:** `docs/3d-model-review.md` scores every catalogue asset; every P0/P1 is fixed or
+explicitly accepted with a measurement; `inspect`/`audit`/`fit`/proportions, the 24-NPC budget
+and the capture gates are green; no runtime contract, dependency or release artifact changed
+except the rebuilt assets themselves.
+
+### A. Baseline, scorecard and routing
+
+- [ ] Load `blender-director` plus the per-family specialists before touching geometry, and
+      record the skill stack and the advice that changed a decision in `docs/3d-model-review.md`.
+- [ ] Re-run the neutral NPC baseline (48 frames), the final-context review (96 frames), the
+      prop gallery and the 24-NPC budget on the current build; keep raw frames in `tmp/` and do
+      not replace committed sheets until a fix lands.
+- [ ] Create `docs/3d-model-review.md` with a per-asset scorecard: silhouette, anatomy/proportion,
+      topology/edge flow, UV/texel density, textures/materials, rig/weights, LOD/impostor,
+      collision/grounding, export contract, provenance, skill route, verdict (P0/P1/P2/accept).
+- [ ] Collect the baseline measurements in one table: `blender_assets.py inspect`, `audit`, `fit`,
+      `npc_proportions.py`, texel density, triangles/meshes, 24-NPC frame time.
+- [ ] Review the scorecard with the project owner, rank the P0/P1 gaps and decide fix vs accept
+      per item; record the decision and the reason in the review document.
+
+### B. Character re-evaluation (8 variants + fallback)
+
+*Skill route: `character-artist`, `retopology`, `rigging`, `uv-workflow`, `texture-workflow`,
+`materials`, `hair-groom`, `asset-optimization`, `lookdev`, `qa-review`.*
+
+- [ ] Face/head: review eye/mouth loops, cheek/brow/chin transitions and faceting at the 2 m
+      portrait distance against `character-artist` and `retopology`; write the concrete edits.
+- [ ] Hands/grip: re-check palm, finger segments, thumb and gear contact; decide between a
+      finger-curl rig or the current open hand plus `HAND_ITEM_FORWARD`, and record the limit.
+- [ ] Weights/pose: capture idle, walk, sit and wave; inspect shoulder, hip and neck collapse;
+      fix the worst offenders using `rigging` practice without changing the 13-bone contract.
+- [ ] UV/texel density: measure the 256² slot maps; decide whether atlas, UDIM or a resolution
+      change is justified. Do not change `CHAR_MAX_TRIS`/`CHAR_MAX_VERTS` without re-running the
+      budget smoke and the Web/Compatibility path.
+- [ ] Cloth and hair: re-check coat/dress hem clipping and the hair/beard/headgear meshes against
+      `hair-groom` guidance; decide explicitly whether cloth simulation stays out of scope.
+- [ ] Fix only the ranked P0/P1s, one asset family per commit; refresh the comparison sheet and
+      re-run `inspect`, `audit`, `fit` and `npc_proportions.py` after each fix.
+- [ ] `qa-review` sign-off: a ship/no-ship verdict per variant with accepted limits recorded.
+
+### C. Gear and prop re-evaluation
+
+*Skill route: `prop-artist`, `hard-surface`, `vegetation-artist`, `materials`, `collision-proxy`,
+`lod-pipeline`, `qa-review`.*
+
+- [ ] Gear: review the 18 pieces for silhouette, straps/buckles, helmet and lanyard readability;
+      verify attach space and body clearance with `fit`; rebuild only what fails the scorecard.
+- [ ] Props: review the 20 library props plus the 6 generated props for bevels, chamfers,
+      silhouette, ground contact and shared material language (`prop-artist`, `hard-surface`);
+      list the rebuilds that are actually worth the triangle budget.
+- [ ] Foliage: replace the flat-shaded `_shrub()`/`_tree()` clumps with `vegetation-artist`-style
+      carded clusters; measure triangles and keep the 60k `LIBRARY_TRIANGLE_BUDGET`.
+- [ ] Materials and emissives: `lookdev` pass on bulbs, signs, lanterns, the marquee and the
+      sedan at day and late; confirm the flashlight stays a tight spot with no beam geometry.
+- [ ] Collision: verify every prop collider and navigation obstacle matches its silhouette and
+      add the missing `collision-proxy` shapes where walking is blocked.
+- [ ] Exit: a prop/gear capture sheet, budgets and collision checks pass with no unreviewed asset.
+
+### D. Venue and camera re-evaluation
+
+*Skill route: `environment-artist`, `archviz`, `set-dressing`, `lighting`, `camera-cinematography`,
+`lookdev`, `rendering`, `qa-review`.*
+
+- [ ] Facade/venue review: rank the top five shape-language fixes for Hollywood, the Joke Shop and
+      the TV Studio (`environment-artist`, `archviz`) and check them against the map-budget tests.
+- [ ] Lighting: `lighting`/`lookdev` pass on sconces, canopy, marquee and street lamps with
+      day/late captures; verify local attenuation stays at 1.0 and no constant-falloff light returns.
+- [ ] Set dressing: review entrance and sidewalk prop clusters (`set-dressing`) for occlusion,
+      navigation obstacles and readable storytelling clusters.
+- [ ] Camera height: decide 1.7 m gameplay framing vs the ~1.55 m concept frames using
+      `camera-cinematography`, and record the decision instead of leaving the mismatch open.
+- [ ] Hooded/winter portrait: close the deferred close-up by staging a types 5–7 capture
+      (`--portrait-type`) or adding a scenario spawn that provides one.
+
+### E. Gates and documentation
+
+- [ ] Add the measurable checks the review needs (texel density, weight/pose sanity, collider
+      coverage, LOD/impostor check) as stdlib or GDScript gates where possible; CI stays offline.
+- [ ] Add a coverage test so every catalogue asset has a scorecard row and verdict.
+- [ ] Update `docs/blender-assets.md`, `docs/npc-visual-review.md`, `docs/3d-model-review.md`,
+      `docs/architecture.md` and `AGENTS.md` with the skill routes and the new gates.
+- [ ] Record accepted limits and deferred rebuilds in `docs/known-limitations.md` and `MEMORY.md`.
 
 ## Character development
 
 Goal: break the eight-character clone effect with four licence-cleared crowd-variety looks while
 keeping the frozen `npc_base` bone, tint-slot and gear-attach contract. The procedural generator is
 the primary path; Quaternius is a reviewed reference and MakeHuman is deferred to a separate local
-development pipeline. This is the current development priority; keep it out of the frozen
-0.10.0 release candidate and assign a release version only when its scope is accepted.
+development pipeline.
+
+The roster build is deliberately sequenced **after** the 3D model re-evaluation: the new variants
+must inherit the reviewed face/hand/gear baseline instead of multiplying whatever it finds.
+
+*Skill route: `character-artist`, `retopology`, `rigging`, `uv-workflow`, `materials`,
+`asset-optimization`, `lowpoly-style`, `godot-export`, `qa-review`.*
 
 ### Source review and decision
 
@@ -41,13 +144,13 @@ development pipeline. This is the current development priority; keep it out of t
       (`blender_assets.py inspect --file ...` applies parent/object transforms, converts to
       Godot Y-up, ignores the importer's bone-shape helper and prints budgets, bones, slots,
       UVs, skinning and animations with issues/warnings; `--json` and `--category` supported.
-      Seven synthetic contract tests added (Python 108/108); all 46 library GLBs inspect clean,
-      and a wrapped 2x fixture correctly reported a 0.975 m local/world delta. Blender-side only;
-      `audit` remains the final gate.)
+      Seven synthetic contract tests added; all 46 library GLBs inspect clean. `audit` remains
+      the final gate.)
 - [ ] Add an `adapt` path for future external assets: scale/ground, weld seams, decimate, map the
       five tint slots, and either rebuild the contract armature or record an explicit second-rig
-      contract. The Quaternius prototype showed that a contract rebuild needs a proportion warp,
-      so this is not on the character-development critical path.
+      contract. Use `retopology`/`rigging`/`uv-workflow` guidance for the manual steps, but keep
+      the Quaternius no-go: a contract rebuild needs a proportion warp, so this is not on the
+      character critical path.
 - [ ] Keep `tools/blender_assets.py audit` as the single final gate; add a regression test for the
       intake report/adaptation using a deterministic synthetic or licence-cleared fixture.
 - [ ] Document the adapter commands and the go/no-go checks in `docs/asset-tooling.md`.
@@ -71,13 +174,16 @@ development pipeline. This is the current development priority; keep it out of t
       accessory; no clothing that labels the NPC's behaviour). (`npc_crowd_a` bun/slim/sunglasses,
       `npc_crowd_b` bald/broad/beard/shoulder bag, `npc_crowd_c` coat/umbrella, `npc_crowd_d`
       hood/cup; prototype sheet committed.)
+- [ ] Re-derive the four approved specs against the re-evaluated character baseline (face, hands,
+      proportions, gear offsets) so the new variants do not reintroduce the gaps the review found.
 - [ ] Add the four approved specs to `CHARACTERS`, build and commit their GLBs/AO maps, and
-      update the manifest with roles and tint data. (Prototype assets are local under `tmp/`;
-      do not add them to the current release candidate.)
+      update the manifest with roles and tint data. Build only the new assets, never a no-op full
+      rebuild, and remember that `export_filter="all_resources"` packs committed GLBs into the
+      next release build: keep this off `main` until after the 0.10.0 tag or accept the size.
 - [ ] Replace the fixed `character_model_for()` mapping with deterministic per-role pools and add
-      tests for repeatability and duplicate avoidance.
+      tests for repeatability and duplicate avoidance (`godot-export` contract check included).
 - [ ] Rebuild and verify gear fit for every new variant (`blender_assets.py fit` plus the runtime
-      gear checks).
+      gear checks) and re-run `inspect` on the four new GLBs.
 
 ### QA and release
 
@@ -104,6 +210,10 @@ Goal: build a local, data-backed map pipeline that can replace or supplement the
 geometry with licensed OpenStreetMap footprints, 3D building massing, reference-derived photoreal
 facades and AI-assisted composition. This is an unversioned development track, not a release gate.
 
+*Skill route: `environment-artist`, `archviz`, `hard-surface`, `geometry-nodes`,
+`procedural-modeling`, `vegetation-artist`, `set-dressing`, `lighting`, `materials`,
+`texture-workflow`, `lookdev`, `scene-assembly`, `qa-review`.*
+
 ### Data, licensing and scope
 
 - [ ] Define the first target block and whether the result replaces, supplements or only informs
@@ -128,16 +238,19 @@ facades and AI-assisted composition. This is an unversioned development track, n
 - [ ] Separate development-only references from shipped textures; commit only images with clear
       redistribution rights and keep the rest under `tmp/` or the ignored reference folders.
 - [ ] Define the photoreal target and a material-derivation workflow (albedo/normal/roughness,
-      AO, weathering) that matches the existing Compatibility renderer.
+      AO, weathering) that matches the existing Compatibility renderer (`texture-workflow`,
+      `materials`, `lookdev`).
 
 ### Geometry and AI-assisted composing
 
 - [ ] Prototype OSM footprint import into Blender at game scale and orientation, then compare it
       against the current venue footprint/collider layout.
 - [ ] Import 3D building massing for the target block, clean the topology and assign LOD bands;
-      do not import unbounded city-scale meshes into the runtime.
+      do not import unbounded city-scale meshes into the runtime (`scene-assembly`,
+      `asset-optimization`, `lod-pipeline`).
 - [ ] Build a reusable photoreal facade kit (window bays, cornices, storefronts, awnings, signage)
-      so buildings share materials and stay inside the venue triangle/material budgets.
+      with `environment-artist`/`hard-surface`/`geometry-nodes` so buildings share materials and
+      stay inside the venue triangle/material budgets.
 - [ ] Add an AI-assisted composition step for palette, facade rhythm and reference boards; record
       the model, prompt and output, and keep human review as the acceptance gate. AI output must
       not be treated as licensed map data or shipped without provenance.
@@ -188,9 +301,10 @@ reference-image licence and privacy gates above rather than replacing them.
       erase facade edges, signage or window detail, and keep the redacted input variant.
 - [ ] Align the splat to game coordinates from the OSM footprint plus surveyed points (door
       width, column spacing, step rise, storefront height) and record alignment residuals in a
-      Blender scene containing both the splat and the current venue geometry.
-- [ ] Judge quality from held-out views and a render from the game’s day/night camera poses,
-      not from training views alone; reject splats that only look correct from the capture path.
+      Blender scene containing both the splat and the current venue geometry (`scene-assembly`).
+- [ ] Judge quality from held-out views and a render from the game's day/night camera poses
+      (`camera-cinematography`, `lookdev`, `qa-review`), not from training views alone; reject
+      splats that only look correct from the capture path.
 - [ ] Maintain a coverage map per block: which facades, corners and rooflines have usable views,
       and which areas must fall back to the facade kit or stylized geometry.
 
@@ -200,11 +314,11 @@ reference-image licence and privacy gates above rather than replacing them.
       compare it with the realism captures; derive dimensions, signage placement, facade rhythm
       and material colours. No splat data reaches the runtime.
 - [ ] **Path B — Texture derivation:** project splat renders onto the facade kit to produce
-      albedo, normal, roughness and AO candidates; de-light and bake them, then verify UV scale,
-      tiling and readability under the Compatibility renderer.
+      albedo, normal, roughness and AO candidates; de-light and bake them (`texture-workflow`),
+      then verify UV scale, tiling and readability under the Compatibility renderer.
 - [ ] **Path C — Mesh proxy:** extract a mesh from the splat or dense reconstruction, retopo and
-      dress it in Blender, assign LODs, and keep it inside the venue triangle/material budgets
-      with mesh-based colliders and navigation.
+      dress it in Blender (`retopology`, `prop-artist`), assign LODs (`lod-pipeline`), and keep it
+      inside the venue triangle/material budgets with mesh-based colliders (`collision-proxy`).
 - [ ] **Path D — Distant backdrop:** test a baked panorama, impostor or billboard generated from
       the splat for skyline and opposite-street depth where the player cannot walk; measure file
       size, loading and visible pop at the venue boundary.
@@ -244,11 +358,87 @@ reference-image licence and privacy gates above rather than replacing them.
       facades, passes performance/navigation/visual QA and has complete provenance; or the
       pipeline is rejected with the findings recorded.
 
+## Maintenance and documentation hygiene
+
+Unversioned housekeeping that keeps the planning sources trustworthy. Do these when they block a
+task, not on a schedule.
+
+### Memory and status accuracy
+
+- [ ] Trim `MEMORY.md` from ~970 lines to the durable set: drop the pre-0.8 host/disk/NSIS/release
+      anecdotes, superseded v0.2.x handoff status and duplicated release notes, keep the gotchas
+      that still bite (async/event wiring, import order, colliders, RNG, contracts), and keep the
+      easter-egg note.
+- [ ] Re-check the remaining memory claims against code before deleting them; `MAX_LIVE_NPCS` is
+      still 16 for gameplay while the separate budget smoke spawns 24, so do not "fix" that.
+- [ ] Fix stale status claims as they are found (the step-collider and flashlight-beam claims were
+      corrected on 2026-11-03); add a documentation check or grep note only if the same class of
+      drift returns.
+
+### Blender skills upkeep
+
+- [ ] Re-check the installed Blender pack (`version.json`) when updating it and record the version
+      in `AGENTS.md` if it changes; keep `.agents/` gitignored and `.agents/.gdignore` in place.
+- [ ] Prune the skill routes in `AGENTS.md` if a skill stops matching the project (for example if
+      a future second-rig path makes `rigging`/`animation` guidance authoritative).
+
+## Gap Analysis and 0.x Roadmap
+
+Baseline: **v0.9.6** (2026-11-03). Three playable venues, tutorial/campaign/checkpoint resume,
+difficulty/help, audio/accessibility settings, a 24-NPC budget baseline, a complete
+escalation/access/radio loop, shift review, the Blender asset library, 272 headless + 108 Python
+tests, a 21/21 clean screen-design judge and an optional Forward+ Desktop HQ build. This is the
+baseline; 0.10.0-rc.1 is the current candidate source, with release approvals still open.
+
+### Documentation audit (completed 2026-11-01)
+
+- [x] Make `docs/architecture.md` the single architecture source: asset/content pipeline
+      contracts, runtime venue-switch/fallback flows, deployment/version constraints, quality
+      requirements, risk register and documentation ownership map.
+- [x] Keep command-heavy operational pages separate but subordinate: `docs/asset-tooling.md`,
+      `docs/blender-assets.md`, `docs/getting-started.md`, `docs/RELEASING.md` and
+      `docs/smoke-checklist.md` link to architecture rather than acting as parallel sources.
+- [x] Remove the obsolete `docs/plan-graphics-v0.6.md` and its links; completed graphics work is
+      represented by ADR-0013, the asset pipeline section and the changelog.
+- [x] Resolve stale claims: README release version/download row, development status
+      "Forward+ not implemented", and Blender props "not AO-baked".
+- [x] Record the documentation ownership rule in `docs/architecture.md` §12: design intent
+      (`PLAN.md`), mechanics (`gameplay.md`), system architecture/contracts (`architecture.md`),
+      decisions (ADRs), status (`development-status.md`), roadmap (`TODO.md`), operations
+      (tooling/release docs).
+
+### Gap register (remaining roadmap)
+
+| Area | Current baseline | Remaining gap | Risk | Milestone |
+| --- | --- | --- | --- | --- |
+| 3D model re-evaluation | 8 characters, 18 gear, 20+6 props audited against budgets/proportions; no skill-guided scorecard yet | Scorecard, ranked P0/P1 fixes, texel/weight/collider measurements and accept decisions | Unmeasured drift stays in the shipped library | Unversioned |
+| Character development | 8-variant library; source/licence review and Quaternius/MakeHuman evaluation complete | Four crowd variants, role pools, gear fit and runtime integration | Repetition remains or external assets break the contract | Unversioned |
+| Local map development | Stylized venue geometry and reference maps | OSM/3D data, photoreal references, AI-assisted composition and splat evaluation | Licensing/provenance or performance failure | Unversioned |
+| Content and balance | 3 venues with validated 12+ event matrices, recovery events and contextual dialogue | External playtest evidence, balance/fairness pass, event-weight tuning and final content lock | Repetition or unfair outcomes survive into the release candidate | 0.10.0 |
+| Dialog/UI | Audited terminology, copy, button states, focus/navigation, modal close, long strings and all 21 screen captures | Keep the capture/compare/judge gate current as content changes | Low; the remaining risk is drift after new UI is added | Complete (0.9.3) |
+| Animation/presentation | Procedural pose layers, portrait captures, reduced hem clipping, 8,000/8,000 NPC cap | Thorough NPC silhouette/face/gear/pose/LOD optimization before visual QA | Close-up jank reduces perceived quality | Complete (0.9.4) |
+| Audio/accessibility | Procedural ambience/SFX, volume controls, captions/reduce-motion settings, searchable Help | Full accessibility/localization acceptance review, audio caption coverage and P0/P1 fixes | Public release cannot claim accessible play | 0.10.0 |
+| Localization | English JSON table, validation and input-generated Help | Pseudo-locale/layout audit and translator workflow sign-off | Text expansion/clipping at release | 0.10.0 |
+| Performance/platform | 24-NPC headless budget (6.94 ms), Compatibility/Web/HQ paths | Target hardware matrix, Web/HQ manual profiling, frame-time/memory evidence | Platform-specific drops or startup failures | 0.9.6 / 0.10.0 |
+| Save lifecycle | Settings/campaign/history/checkpoint files, atomic writes | Migration fixtures from 0.8/0.9, fresh-install/upgrade verification, rollback instructions | Player progress loss on upgrade | 0.10.0 |
+| QA/CI | Headless suite, smokes, screenshot compare, 21/21 judge | Release-candidate smoke matrix, final gallery and release dry run | Regressions escape the release gate | 0.9.6 / 0.10.0 |
+| Release/legal | Unsigned desktop/macOS builds, Web/HQ packages | Signing/notarization decision, checksums, AI-asset provenance, credits, privacy/body-camera and rating review | Public release blocked by trust/policy | 0.10.0 |
+| Documentation | Architecture/ADR/status/roadmap ownership established | Final README/changelog/status/TODO/smoke-checklist sign-off and known-limitations page | Docs drift at launch | 0.9.6 / 0.10.0 |
+| Future 0.x | Stable 0.x feature set | Live replay, more venues, VO/music, image-to-mesh props and seasonal content | Scope creep before release | Unscheduled |
+
+**Current priority:** the unversioned **3D model re-evaluation** runs first because its findings
+gate the crowd-variant roster; **Character development** and **Local map development** follow.
+The **0.10.0 release candidate** is the next versioned release; content lock, external playtest,
+balance, save migration and release/legal sign-off remain tracked there. The source version is
+0.10.0-rc.1; the v0.9.6 release is published. Earlier map/NPC/UI/texture/integration passes are
+complete and in `DONE.md`.
+
 ## Milestone 0.10.0 — Release candidate
 
 Goal: freeze the feature set, prove it with external players, clear release/legal blockers and
 verify every distribution path. This remains the next versioned release after the 0.9.x patch
-line, but follows the unversioned character and local-map tracks in planning order.
+line, but follows the unversioned 3D re-evaluation, character and local-map tracks in planning
+order. Only a release-blocking defect from those tracks may enter this milestone before the tag.
 
 ### Content lock and external playtest
 
@@ -394,64 +584,6 @@ line, but follows the unversioned character and local-map tracks in planning ord
 - [ ] Exit: the release candidate is content-locked, externally played, legally reviewed,
       documented, reproducibly built, and approved as the 0.10.0 release candidate.
 
-## Gap Analysis and 0.x Roadmap
-
-Baseline: **v0.9.6** (2026-11-03). Three playable venues, tutorial/campaign/checkpoint resume,
-difficulty/help, audio/accessibility settings, a 24-NPC budget baseline, a complete
-escalation/access/radio loop, shift review, the Blender asset library, 272 headless + 101 Python
-tests, a 21/21 clean screen-design judge and an optional Forward+ Desktop HQ build. This is the
-baseline; 0.10.0-rc.1 is the current candidate source, with release approvals still open.
-
-### Documentation audit (completed 2026-11-01)
-
-- [x] Make `docs/architecture.md` the single architecture source: asset/content pipeline
-      contracts, runtime venue-switch/fallback flows, deployment/version constraints, quality
-      requirements, risk register and documentation ownership map.
-- [x] Keep command-heavy operational pages separate but subordinate: `docs/asset-tooling.md`,
-      `docs/blender-assets.md`, `docs/getting-started.md`, `docs/RELEASING.md` and
-      `docs/smoke-checklist.md` link to architecture rather than acting as parallel sources.
-- [x] Remove the obsolete `docs/plan-graphics-v0.6.md` and its links; completed graphics work is
-      represented by ADR-0013, the asset pipeline section and the changelog.
-- [x] Resolve stale claims: README release version/download row, development status
-      "Forward+ not implemented", and Blender props "not AO-baked".
-- [x] Record the documentation ownership rule in `docs/architecture.md` §12: design intent
-      (`PLAN.md`), mechanics (`gameplay.md`), system architecture/contracts (`architecture.md`),
-      decisions (ADRs), status (`development-status.md`), roadmap (`TODO.md`), operations
-      (tooling/release docs).
-
-### Gap register (remaining roadmap)
-
-| Area | Current baseline | Remaining gap | Risk | Milestone |
-| --- | --- | --- | --- | --- |
-| Character development | 8-variant library; source/licence review and Quaternius/MakeHuman evaluation complete | Four crowd variants, role pools, gear fit and runtime integration | Repetition remains or external assets break the contract | Unversioned |
-| Local map development | Stylized venue geometry and reference maps | OSM/3D data, photoreal references, AI-assisted composition and splat evaluation | Licensing/provenance or performance failure | Unversioned |
-| Content and balance | 3 venues with validated 12+ event matrices, recovery events and contextual dialogue | External playtest evidence, balance/fairness pass, event-weight tuning and final content lock | Repetition or unfair outcomes survive into the release candidate | 0.10.0 |
-| Dialog/UI | Audited terminology, copy, button states, focus/navigation, modal close, long strings and all 21 screen captures | Keep the capture/compare/judge gate current as content changes | Low; the remaining risk is drift after new UI is added | Complete (0.9.3) |
-| Animation/presentation | Procedural pose layers, portrait captures, reduced hem clipping, 8,000/8,000 NPC cap | Thorough NPC silhouette/face/gear/pose/LOD optimization before visual QA | Close-up jank reduces perceived quality | Complete (0.9.4) |
-| Audio/accessibility | Procedural ambience/SFX, volume controls, captions/reduce-motion settings, searchable Help | Full accessibility/localization acceptance review, audio caption coverage and P0/P1 fixes | Public release cannot claim accessible play | 0.10.0 |
-| Localization | English JSON table, validation and input-generated Help | Pseudo-locale/layout audit and translator workflow sign-off | Text expansion/clipping at release | 0.10.0 |
-| Performance/platform | 24-NPC headless budget (6.94 ms), Compatibility/Web/HQ paths | Target hardware matrix, Web/HQ manual profiling, frame-time/memory evidence | Platform-specific drops or startup failures | 0.9.6 / 0.10.0 |
-| Save lifecycle | Settings/campaign/history/checkpoint files, atomic writes | Migration fixtures from 0.8/0.9, fresh-install/upgrade verification, rollback instructions | Player progress loss on upgrade | 0.10.0 |
-| QA/CI | Headless suite, smokes, screenshot compare, 21/21 judge | Release-candidate smoke matrix, final gallery and release dry run | Regressions escape the release gate | 0.9.6 / 0.10.0 |
-| Release/legal | Unsigned desktop/macOS builds, Web/HQ packages | Signing/notarization decision, checksums, AI-asset provenance, credits, privacy/body-camera and rating review | Public release blocked by trust/policy | 0.10.0 |
-| Documentation | Architecture/ADR/status/roadmap ownership established | Final README/changelog/status/TODO/smoke-checklist sign-off and known-limitations page | Docs drift at launch | 0.9.6 / 0.10.0 |
-| Future 0.x | Stable 0.x feature set | Live replay, more venues, VO/music, image-to-mesh props and seasonal content | Scope creep before release | Unscheduled |
-
-**Current priority:** the unversioned **Character development** and **Local map development**
-tracks are the active planning order. The **0.10.0 release candidate** follows them in this list
-as the next versioned release; content lock, external playtest, balance, save migration and
-release/legal sign-off remain tracked there. The source version is 0.10.0-rc.1; the v0.9.6
-release is published. Earlier map/NPC/UI/texture/integration passes are complete and in
-`DONE.md`.
-
-## Future 0.x / non-blocking candidates
-
-- [ ] Live 3D body-camera replay.
-- [ ] Additional venues or seasonal events.
-- [ ] Real voice/dialogue audio.
-- [ ] Image-to-mesh prop pipeline.
-- [ ] Rare easter eggs/cameos (e.g. Bob Dobbs) only after core content and QA are stable.
-
 ## Roadmap dependencies and risks
 
 - Content volume is the largest schedule risk; tutorial and campaign work must not outrun the
@@ -468,23 +600,18 @@ release is published. Earlier map/NPC/UI/texture/integration passes are complete
 - Keep the de-escalation-first rules and rare-emergency design unchanged while adding progression,
       difficulty and feedback.
 
-## Realism follow-ups (deferred)
+- The Blender skill pack is local, unversioned guidance; it can drift from the project contract or
+      assume Blender MCP. Re-verify every adopted technique with `inspect`/`audit`/`fit` and never
+      make a skill, asset pack or tool version a build or CI dependency.
+- The 3D re-evaluation can expand without bound (retopo every head, rig every finger). Cap it with
+      the scorecard ranking: P0/P1 only, one asset family per commit, and an explicit accept for
+      everything else.
 
-- [x] Re-run the screen-design judge for `12-joke-shop`, `13-studio-arrival` and the
-      Forward+ A/B with the valid key: **13/13 match, 0 diff**; the HQ A/B is in the report.
-- [ ] Capture a hooded/winter NPC close-up once a scenario spawns types 5-7 (the explicit
-      `--portrait-type` override is already in `capture_realism.gd`).
-- [ ] Human-authored tree/hedge silhouettes: the whole scene now reads well except the foliage,
-      which is still flat-shaded clumps (`_shrub()`/`_tree()` in `tools/blender_assets.py`).
-      Only worth doing if a higher-poly organic builder is added to the library.
-- [ ] Camera height: the concept frames are shot at ~1.55 m with a slightly wider lens; the game
-      camera sits at 1.7 m. Decide whether the first-person height should match the art direction
-      or stay at the current gameplay value.
+## Future / deferred (not scheduled)
 
-## Future ideas (optional, not scheduled)
-
-- [ ] Live 3D body-camera replay (the timeline is text-only today).
-- [ ] Additional venues or seasonal events authored from the new `.tres` files.
-- [ ] Real voice/dialogue audio (the crowd bed and radio crackle are procedural).
+- [ ] Live 3D body-camera replay; the timeline is text-only today.
+- [ ] Additional venues or seasonal events authored from the `content/venues/` resources.
+- [ ] Real voice/dialogue audio; ambience, cues and radio crackle are procedural.
 - [ ] Swap the primitive prop builder for an image-to-mesh model API behind the same
       `assets/models/generated/<prop>.glb` contract.
+- [ ] Rare easter eggs/cameos (e.g. Bob Dobbs) only after core content and QA are stable.

docs/development-status.md

@@ -1,8 +1,10 @@
 # Development Status
 
-Current planning priorities are the unversioned **Character development** and **Local map
-development** tracks in [`TODO.md`](../TODO.md); neither is part of the implemented list below
-yet. The 0.10.0 release candidate remains the next versioned release.
+Current planning priorities are the unversioned **3D model re-evaluation**, **Character
+development** and **Local map development** tracks in [`TODO.md`](../TODO.md); none is part of the
+implemented list below yet. The re-evaluation scores the existing characters, gear, props and
+venues against the locally installed Blender skill guidance before the roster expands. The 0.10.0
+release candidate remains the next versioned release.
 
 Implemented so far:
 
@@ -48,7 +50,8 @@ Implemented so far:
 - NPC torso/hip transition (0.9.4 work): the pelvis and chest are slightly wider than the waist, giving shirts and jackets a readable taper into the hips without changing the build multipliers or the skeleton. The comparison is `sheet_npc_baseline_compare-torso.jpg`.
 - Release v0.9.6 (2026-11-03): the tagged commit `4bb2896` passed Gitea Actions run 1296 (checks + full build) and published seven artifacts (Linux, Windows, macOS, NSIS setup, AppImage, Web and Desktop HQ). Release page: <https://git.allucanget.biz/zwitschi/security-game/releases/tag/v0.9.6>.
 - Release candidate (0.10.0-rc.1, 2026-11-03): clean worktree `a9658ec` built Linux 161.5 MB, Windows 197.2 MB, macOS 148.0 MB, Web 97.8 MB and Desktop HQ 241.6 MB. The Web/HQ archives contain `CREDITS.md`, `PRIVACY.md`, `docs/accessibility.md` and `docs/known-limitations.md`; SHA-256 checksums were generated, the Linux and HQ binaries launched headless, and the Web index/wasm/pck/docs served with HTTP 200. The candidate is not published: external playtest, AI-provider/legal approval, macOS signing and the Windows/macOS target matrix remain open blockers.
-- Save migration and reliability (0.10.0 work): committed 0.8-era settings/campaign/history fixtures and a 0.9.0 checkpoint. The loader default-fills old settings, keeps campaign progress and validates checkpoints for all three venues with an open incident, NPC rows, door, flashlight and body-camera state. Corrupt and partial files fall back to defaults; a blocked atomic replace leaves the prior valid file intact. `_load_json` uses the non-logging JSON parser; save locations and rollback are documented in `getting-started.md`. Suite is 272/272.
+- Save migration and reliability (0.10.0 work): committed 0.8-era settings/campaign/history fixtures and a 0.9.0 checkpoint. The loader default-fills old settings, keeps campaign progress and validates checkpoints for all three venues with an open incident, NPC rows, door, flashlight and body-camera state. Corrupt and partial files fall back to defaults; a blocked atomic replace leaves the prior valid file intact. `_load_json` uses the non-logging JSON parser; save locations and rollback are documented in `getting-started.md`. Suite is 272/272;
+Python tooling is 108/108 after the GLB `inspect` intake coverage.
 - Integration checkpoint (0.9.6 work): the clean worktree at `f5b8ef2` rebuilt all five release packages and the Desktop HQ zip without ignored `tmp/`/`build/` content (Linux 161.5 MB, Windows 197.1 MB, macOS 148.0 MB, Web 97.8 MB, HQ 241.5 MB). The Web zip contains `index.html`, `index.wasm` and `index.pck`; the HQ zip contains `override.cfg` and launches headless with the Forward+ override. Current verification: Godot 263/263, Python 101/101, 14 smokes green, 24-NPC budget 6.90 ms / 168 meshes / 131,616 triangles, gallery 21 good / 0 bad, screen judge 21/21 match with 0 review/diff, and a pinned realism judge (`z-ai/glm-5.3-flash`, one pass per shot) at mean overall 2.1 and Pillow gap 0.291. No new gameplay or content changes.
 - Release v0.9.5 (2026-11-02): the tagged commit `4bc6390` passed Gitea Actions run 922 (checks + full build) and published seven artifacts (Linux, Windows, macOS, NSIS setup, AppImage, Web and Desktop HQ). Release page: <https://git.allucanget.biz/zwitschi/security-game/releases/tag/v0.9.5>.
 - Release v0.9.4 (2026-11-02): the tagged commit `a57b86c` passed the Gitea checks and full build and published seven artifacts (Linux, Windows, macOS, NSIS setup, AppImage, Web and Desktop HQ). Release page: <https://git.allucanget.biz/zwitschi/security-game/releases/tag/v0.9.4>.
@@ -66,7 +69,8 @@ Implemented so far:
 - NPC P1 portrait fixes (0.9.4 work): tourists now hide their photo-pose camera bag outside `photo`/`film` poses, removing the duplicate-backpack silhouette; the backpack gear was rebuilt with flattened, forward chest straps and a fresh AO map. The before/after sheet is [`blender_previews/sheet_npc_baseline_compare-p1.jpg`](blender_previews/sheet_npc_baseline_compare-p1.jpg); the remaining face/hand work was closed under the 0.9.4 acceptance limits.
 - NPC visual baseline (0.9.4 work): `src/test/capture_npc_baseline.gd` captures all eight library characters from front, three-quarter, profile, back, full-body and crowd views in a neutral studio (48 frames). The criteria, measurements and ranked gaps are in [`npc-visual-review.md`](npc-visual-review.md); the committed sheet is [`blender_previews/sheet_npc_baseline.jpg`](blender_previews/sheet_npc_baseline.jpg). No model changes were made in this baseline step.
 - UI-state capture pass (0.9.3 work): `src/test/capture_ui_states.gd` adds committed captures 14–21 for the start menu, pause, settings, shift review, emergency radio, action wheel, crowd panic and responder arrival. Every one of the 21 catalogue entries now has a current-build capture; `tools/check_screenshots.sh` reports 21 good / 0 bad and the automatic judge reports 21/21 match, 0 diff. Crowd panic and responder arrival are staged presentation frames, not gameplay changes.
-- Realism adoption (post-v0.6.0): v2 per-phase skylines and surface tiles with normal maps, retuned sun/exposure/fog, wet asphalt, denser foliage, a visible additive flashlight beam, and refreshed staging captures. `tools/check_screenshots.sh` is the repeatable capture/compare gate; judge is clean of `diff` verdicts.
+- Realism adoption (post-v0.6.0): v2 per-phase skylines and surface tiles with normal maps, retuned sun/exposure/fog, wet asphalt, denser foliage, a tight additive flashlight spot with no beam geometry, and refreshed
+staging captures. `tools/check_screenshots.sh` is the repeatable capture/compare gate; judge is clean of `diff` verdicts.
 - Input/reliability fixes: selecting a venue releases the start-menu input lock and recaptures the mouse; the body-camera overlay no longer covers the shift status. Mouse look now reads at `_input`, restores capture on window focus/click-back, and the HUD is mouse-transparent. Normal radio reports submit on click; emergency calls still require the hold-to-confirm path.
 - Production-quality pass (v0.9.0): mixer/accessibility settings, audio volume clamping, 24-NPC budget smoke, shortened coat/dress stride, and a unified Godot 4.7.2 CI/local support baseline with documented upgrade/rollback requirements.
 - Campaign and first-hour pass (v0.8.0): campaign progression, checkpoint resume, difficulty presets, a searchable Help overlay, recovery events, contextual dialogue, policy/debrief lines and tutorial/replay/skip. Scripted QA bypasses the tutorial unless `--tutorial` is passed.