refactor: rename thumbnail_rendering queue to asset_pipeline

The queue handles far more than thumbnails: OCC tessellation, USD master
generation, GLB production, order line renders, and workflow renders.
asset_pipeline better reflects its role as the render-worker's primary queue.

Updated all references in: task decorators, celery_app.py, beat_tasks.py,
docker-compose.yml worker command, worker.py MONITORED_QUEUES, admin.py,
CLAUDE.md, LEARNINGS.md, Dockerfile, helpTexts.ts, test files,
and all .claude/commands/*.md skill files.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

CLAUDE.md

@@ -22,7 +22,7 @@ Automatisiertes Render-System für Schaeffler-Produktbilder. Kunden (intern) lad
 | `redis` | 6379 | Celery Broker |
 | `backend` | 8888 | FastAPI App (uvicorn) |
 | `worker` | – | Celery Worker, Queue: `step_processing`, concurrency=8 |
-| `worker-thumbnail` | – | Celery Worker, Queue: `thumbnail_rendering`, **concurrency=1** |
+| `worker-thumbnail` | – | Celery Worker, Queue: `asset_pipeline`, **concurrency=1** |
 | `beat` | – | Celery Beat (Scheduler) |
 | `blender-renderer` | 8100 | Blender HTTP-Service (STEP→PNG, STEP→STL) |
 | `threejs-renderer` | 8101 | Three.js/Playwright HTTP-Service |
@@ -105,10 +105,10 @@ docker compose exec backend alembic current
 | Queue | Worker | Concurrency | Tasks |
 |---|---|---|---|
 | `step_processing` | `worker` | 8 | `process_step_file`, `render_order_line_task`, `dispatch_order_line_render` |
-| `thumbnail_rendering` | `worker-thumbnail` | 1 | `render_step_thumbnail`, `regenerate_thumbnail`, `generate_stl_cache` |
+| `asset_pipeline` | `worker-thumbnail` | 1 | `render_step_thumbnail`, `regenerate_thumbnail`, `generate_stl_cache` |
 | `ai_validation` | `worker` | 8 | Azure AI Validierung |
 
-**Wichtig**: `thumbnail_rendering` läuft mit concurrency=1, weil der blender-renderer nur 1 Request gleichzeitig verarbeiten kann. Mehr parallele Requests führen zu Timeouts.
+**Wichtig**: `asset_pipeline` läuft mit concurrency=1, weil der blender-renderer nur 1 Request gleichzeitig verarbeiten kann. Mehr parallele Requests führen zu Timeouts.
 
 ## STEP-Processing-Pipeline
 
@@ -118,7 +118,7 @@ docker compose exec backend alembic current
    - `parsed_objects` in DB speichern
    - glTF konvertieren (falls konfiguriert)
    - Status: `processing` → queut `render_step_thumbnail`
-3. **Thumbnail** (`render_step_thumbnail` auf `thumbnail_rendering`):
+3. **Thumbnail** (`render_step_thumbnail` auf `asset_pipeline`):
    - Blender oder Three.js renderer aufrufen
    - STL-Cache erstellen: `{step_stem}_low.stl`, `{step_stem}_high.stl`
    - Status: `completed` oder `failed`

LEARNINGS.md

@@ -52,7 +52,7 @@ Im Template-Modus (Mode B) lief Auto-Licht/World-Setup bedingungslos → übersc
 
 ### 2026-02-15 | Celery | Blender-Queue-Flooding durch falsche Concurrency
 Alle Tasks auf `step_processing` (concurrency=8) → 8 Workers gleichzeitig an blender-renderer (max 1) → 7× Timeout.
-**Lösung:** `process_step_file` (step_processing, concurrency=8) nur schnelle Metadata; `render_step_thumbnail` (thumbnail_rendering, concurrency=1) für Blender. HTTP-Services mit max 1 Request immer auf eigener Queue mit concurrency=1.
+**Lösung:** `process_step_file` (step_processing, concurrency=8) nur schnelle Metadata; `render_step_thumbnail` (asset_pipeline, concurrency=1) für Blender. HTTP-Services mit max 1 Request immer auf eigener Queue mit concurrency=1.
 
 ### 2026-02-18 | Frontend | Tailwind CSS-Variablen inkompatibel mit opacity-Syntax
 `bg-surface/50` erzeugt `rgb(var(--color-bg-surface) / 0.5)` — invalides CSS wenn Variable ein Hex-Wert ist.
@@ -64,7 +64,7 @@ Three.js schrieb STL in tempfile und löschte es → Download-Endpoint fand nich
 
 ### 2026-02-20 | STL-Cache | blender-renderer fehlte /convert-stl Endpoint
 Blender renderte + konvertierte in einem Schritt, persistierte STL nicht.
-**Lösung:** Neuer `/convert-stl` Endpoint (STEP→STL ohne Render), Celery-Task `generate_stl_cache` auf `thumbnail_rendering`, Admin-Batch-Funktion "Generate Missing STLs".
+**Lösung:** Neuer `/convert-stl` Endpoint (STEP→STL ohne Render), Celery-Task `generate_stl_cache` auf `asset_pipeline`, Admin-Batch-Funktion "Generate Missing STLs".
 
 ### 2026-02-22 | Material-System | Fehlender Alias blockiert Material-Replacement
 `"Stahl v2"` in DB nicht in materials noch in material_aliases → keine Ersetzung, Silent-Fail.
@@ -179,9 +179,9 @@ Celery-Task importierte nur `AssetLibrary` → `Material.creator` → String-Ref
 
 ### 2026-03-06 | Render-Pipeline | render_order_line_task auf falschem Worker (kein Blender)
 Task auf `step_processing` → `worker`-Container (kein Blender) → `is_blender_available()` = False → Pillow-Placeholder, kein Fehler.
-**Lösung:** Queue auf `thumbnail_rendering` → `render-worker`-Container (Blender 5.0.1 + cadquery). Blender-Tasks IMMER auf `thumbnail_rendering`.
+**Lösung:** Queue auf `asset_pipeline` → `render-worker`-Container (Blender 5.0.1 + cadquery). Blender-Tasks IMMER auf `asset_pipeline`.
 
-### 2026-03-06 | Docker | worker-thumbnail vs render-worker — beide auf thumbnail_rendering
+### 2026-03-06 | Docker | worker-thumbnail vs render-worker — beide auf asset_pipeline
 Zwei Services unterschiedlicher Capabilities auf gleicher Queue → round-robin → 50% Silent-Fail.
 **Lösung:** `worker-thumbnail` aus docker-compose entfernt. `render-worker` ist alleiniger Consumer. Nie zwei Services mit unterschiedlichen Fähigkeiten auf dieselbe Queue.
 

PLAN.md

@@ -1418,8 +1418,8 @@ Nach Aktivierung von Multi-Tenancy (Migration 035/036) hatten mehrere Bugs die g
 
 | Fix | Problem | Lösung | Datei |
 |---|---|---|---|
-| B-Fix-1 | `worker-thumbnail` ohne Blender konkurrierte auf `thumbnail_rendering` → 50% Silent-Fails | `worker-thumbnail` aus docker-compose.yml entfernt | `docker-compose.yml` |
-| B-Fix-2 | `render_order_line_task` auf `step_processing` Queue → `worker` ohne Blender → Pillow-Fallback | Queue zu `thumbnail_rendering` geändert | `step_tasks.py:247` |
+| B-Fix-1 | `worker-thumbnail` ohne Blender konkurrierte auf `asset_pipeline` → 50% Silent-Fails | `worker-thumbnail` aus docker-compose.yml entfernt | `docker-compose.yml` |
+| B-Fix-2 | `render_order_line_task` auf `step_processing` Queue → `worker` ohne Blender → Pillow-Fallback | Queue zu `asset_pipeline` geändert | `step_tasks.py:247` |
 | B-Fix-3 | Circular Import `template_service.py` ↔ `domains/rendering/service.py` → `resolve_template()` nie aufrufbar | Volle sync SQLAlchemy Implementierung in `template_service.py` wiederhergestellt | `services/template_service.py` |
 | B-Fix-4 | `audit_log.tenant_id NOT NULL` → Broadcast-Notifications scheiterten → Order Submit 500 | `ALTER TABLE audit_log ALTER COLUMN tenant_id DROP NOT NULL` | DB direkt |
 | B-Fix-5 | Shared System-Tabellen (`output_types`, `materials`, etc.) `tenant_id NOT NULL` → Create-Endpoints schlugen fehl | `tenant_id DROP NOT NULL` für alle System-Tabellen | DB direkt |
@@ -1433,7 +1433,7 @@ Nach Aktivierung von Multi-Tenancy (Migration 035/036) hatten mehrere Bugs die g
 **`GET /api/worker/health/render`** — Render Health Endpoint:
 - Render-Worker connected (Celery inspect)
 - Blender erreichbar (HTTP GET blender-renderer:8100/health)
-- `thumbnail_rendering` Queue Tiefe < 10
+- `asset_pipeline` Queue Tiefe < 10
 - Letzter Render < 30 min alt und erfolgreich
 - Response: `{ status: "ok"|"degraded"|"down", render_worker_connected, blender_available, thumbnail_queue_depth, last_render_at, ... }`
 
@@ -1449,7 +1449,7 @@ python scripts/test_render_pipeline.py --full      # Alle Output-Types (langsam)
 | Queue | Worker | Concurrency | Tasks |
 |---|---|---|---|
 | `step_processing` | `worker` | 8 | `process_step_file`, `dispatch_order_line_render` |
-| `thumbnail_rendering` | `render-worker` (Blender 5.0.1) | 1 | `render_step_thumbnail`, `regenerate_thumbnail`, `render_order_line_task`, `generate_stl_cache` |
+| `asset_pipeline` | `render-worker` (Blender 5.0.1) | 1 | `render_step_thumbnail`, `regenerate_thumbnail`, `render_order_line_task`, `generate_stl_cache` |
 | `ai_validation` | `worker` | 8 | Azure AI Validierung |
 
-**Schlüsselprinzip**: Alles was Blender aufruft → `thumbnail_rendering` Queue → nur `render-worker` → kein Timeout durch parallele Requests.
+**Schlüsselprinzip**: Alles was Blender aufruft → `asset_pipeline` Queue → nur `render-worker` → kein Timeout durch parallele Requests.

PLAN_REFACTOR.md

@@ -15,7 +15,7 @@ Schaeffler Automat is a working Blender-based media production pipeline with:
 - 7 Docker services with GPU render-worker
 - PostgreSQL with tenant_id columns + Row Level Security (RLS) enabled but inconsistently
   applied at the application layer
-- Celery task queues with two workers (step_processing + thumbnail_rendering)
+- Celery task queues with two workers (step_processing + asset_pipeline)
 - WebSocket real-time events via Redis Pub/Sub
 - React/Vite frontend with workflow editor (ReactFlow), media browser, notifications
 
@@ -584,7 +584,7 @@ internal teams), RLS + tenant_id partitioning is sufficient.
 - Each tenant gets own PostgreSQL schema (not separate DB) with schema-based routing
 - Shared MinIO with per-tenant bucket policies
 - Separate Redis database (0-15) per tenant (max 16 tenants)
-- Celery routing: per-tenant queue prefix `{tenant_slug}.thumbnail_rendering`
+- Celery routing: per-tenant queue prefix `{tenant_slug}.asset_pipeline`
 
 ### 4.4 Per-Tenant Feature Flags
 
@@ -751,7 +751,7 @@ export const HELP_TEXTS: Record<string, HelpText> = {
   },
   "action.regenerate_thumbnails": {
     title: "Regenerate All Thumbnails",
-    body: "Re-renders thumbnails for all STEP files using current settings. This queues all files on the thumbnail_rendering worker. Expected time: N × 30s. Only needed after changing renderer settings.",
+    body: "Re-renders thumbnails for all STEP files using current settings. This queues all files on the asset_pipeline worker. Expected time: N × 30s. Only needed after changing renderer settings.",
     warning: "This will queue a large number of tasks. Only run during off-peak hours.",
   },
   // ... all settings
@@ -889,7 +889,7 @@ rejection UI. `rejected_at` column exists but there is no rejection reason field
 ### 8.1 Current Concurrency Controls
 
 - `worker` (step_processing): `CELERY_WORKER_CONCURRENCY` env var, default 8
-- `render-worker` (thumbnail_rendering): hardcoded 1 (Blender serial access)
+- `render-worker` (asset_pipeline): hardcoded 1 (Blender serial access)
 - Both require Docker service restart to change concurrency
 
 ### 8.2 Dynamic Worker Scaling
@@ -901,7 +901,7 @@ Use Celery's built-in `autoscale` option:
 render-worker:
   command: celery -A app.tasks.celery_app worker
     --loglevel=info
-    -Q thumbnail_rendering
+    -Q asset_pipeline
     --autoscale=1,1   # min=1, max=1 (single Blender concurrency)
     --concurrency=1
 ```
@@ -984,7 +984,7 @@ After Phase 2 decomposition, update `celery_app.conf.update(task_routes={...})`:
 ```python
 task_routes = {
     "app.domains.pipeline.tasks.*": {"queue": "step_processing"},
-    "app.domains.rendering.tasks.*": {"queue": "thumbnail_rendering"},
+    "app.domains.rendering.tasks.*": {"queue": "asset_pipeline"},
     "app.domains.media.tasks.*": {"queue": "step_processing"},
     "app.tasks.ai_tasks.*": {"queue": "ai_validation"},
     "app.tasks.beat_tasks.*": {"queue": "step_processing"},

backend/alembic/versions/054_worker_configs.py

@@ -26,7 +26,7 @@ def upgrade() -> None:
         INSERT INTO worker_configs (queue_name, max_concurrency, min_concurrency, enabled)
         VALUES
             ('step_processing', 8, 2, true),
-            ('thumbnail_rendering', 1, 1, true),
+            ('asset_pipeline', 1, 1, true),
             ('ai_validation', 4, 1, true)
         ON CONFLICT DO NOTHING
     """)

backend/alembic/versions/6ebfe2737531_rename_tessellation_settings_gltf_.py

@@ -0,0 +1,44 @@
+"""rename_tessellation_settings_gltf_production_to_scene
+
+Revision ID: 6ebfe2737531
+Revises: 062
+Create Date: 2026-03-12 20:39:36.880236
+
+"""
+from typing import Sequence, Union
+
+from alembic import op
+import sqlalchemy as sa
+
+
+# revision identifiers, used by Alembic.
+revision: str = '6ebfe2737531'
+down_revision: Union[str, None] = '062'
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    op.execute("""
+        UPDATE system_settings
+        SET key = 'scene_linear_deflection'
+        WHERE key = 'gltf_production_linear_deflection'
+    """)
+    op.execute("""
+        UPDATE system_settings
+        SET key = 'scene_angular_deflection'
+        WHERE key = 'gltf_production_angular_deflection'
+    """)
+
+
+def downgrade() -> None:
+    op.execute("""
+        UPDATE system_settings
+        SET key = 'gltf_production_linear_deflection'
+        WHERE key = 'scene_linear_deflection'
+    """)
+    op.execute("""
+        UPDATE system_settings
+        SET key = 'gltf_production_angular_deflection'
+        WHERE key = 'scene_angular_deflection'
+    """)

backend/app/api/routers/admin.py

@@ -366,7 +366,7 @@ async def update_settings(
     await db.commit()
 
     # Note: blender-renderer HTTP service removed; concurrency is now controlled
-    # via render-worker Docker concurrency setting (thumbnail_rendering queue).
+    # via render-worker Docker concurrency setting (asset_pipeline queue).
 
     return _settings_to_out(await _load_settings(db))
 

backend/app/api/routers/cad.py

@@ -1,4 +1,5 @@
 """CAD file router - serve thumbnails, glTF models, parsed objects, and trigger reprocessing."""
+import logging
 import uuid
 from datetime import datetime
 from pathlib import Path
@@ -20,6 +21,7 @@ from app.utils.auth import get_current_user, is_privileged
 from app.services.product_service import link_cad_to_product, lookup_product
 
 router = APIRouter(prefix="/cad", tags=["cad"])
+logger = logging.getLogger(__name__)
 
 
 # ---------------------------------------------------------------------------
@@ -273,6 +275,7 @@ async def get_objects(
         "cad_file_id": str(cad.id),
         "original_name": cad.original_name,
         "processing_status": cad.processing_status.value,
+        "step_hash": cad.step_file_hash,
         "parsed_objects": cad.parsed_objects,
     }
 
@@ -318,6 +321,11 @@ async def generate_gltf_production(
     if not cad.stored_path:
         raise HTTPException(status_code=404, detail="STEP file not uploaded for this CAD file")
 
+    logger.warning(
+        "generate_gltf_production called for cad %s — "
+        "deprecated: renders now consume usd_master directly",
+        id,
+    )
     from app.tasks.step_tasks import generate_gltf_production_task
     task = generate_gltf_production_task.delay(str(id))
     return {"status": "queued", "task_id": task.id, "cad_file_id": str(id)}

backend/app/api/routers/orders.py

@@ -1000,6 +1000,53 @@ async def cancel_line_render(
     }
 
 
+class RejectLineBody(BaseModel):
+    reason: str = ""
+
+
+@router.post("/{order_id}/lines/{line_id}/reject", status_code=200)
+async def reject_order_line(
+    order_id: uuid.UUID,
+    line_id: uuid.UUID,
+    body: RejectLineBody,
+    user: User = Depends(get_current_user),
+    db: AsyncSession = Depends(get_db),
+):
+    """Reject a single order line (admin/PM only).
+
+    Sets item_status to 'rejected' and stores the reason in the notes field.
+    """
+    if not _is_privileged(user):
+        raise HTTPException(status_code=403, detail="Insufficient permissions")
+
+    result = await db.execute(select(Order).where(Order.id == order_id))
+    order = result.scalar_one_or_none()
+    if not order:
+        raise HTTPException(404, detail="Order not found")
+
+    line_result = await db.execute(
+        select(OrderLine).where(OrderLine.id == line_id, OrderLine.order_id == order_id)
+    )
+    line = line_result.scalar_one_or_none()
+    if not line:
+        raise HTTPException(404, detail="Order line not found")
+
+    from sqlalchemy import update as sql_update
+
+    notes_value = body.reason.strip() if body.reason and body.reason.strip() else line.notes
+    await db.execute(
+        sql_update(OrderLine)
+        .where(OrderLine.id == line.id)
+        .values(
+            item_status="rejected",
+            notes=notes_value,
+        )
+    )
+    await db.commit()
+
+    return {"rejected": True, "line_id": str(line.id), "reason": body.reason}
+
+
 @router.post("/{order_id}/cancel-renders")
 async def cancel_order_renders(
     order_id: uuid.UUID,

backend/app/api/routers/worker.py

@@ -237,7 +237,7 @@ async def reprocess_cad_file(
 # Queue inspection + control
 # ---------------------------------------------------------------------------
 
-MONITORED_QUEUES = ["step_processing", "thumbnail_rendering", "ai_validation"]
+MONITORED_QUEUES = ["step_processing", "asset_pipeline", "ai_validation"]
 
 
 def _parse_redis_task(raw: str) -> dict | None:
@@ -515,7 +515,7 @@ async def render_health(
 
     details: dict = {}
 
-    # 1. Check if render-worker (thumbnail_rendering queue) is connected + has Blender
+    # 1. Check if render-worker (asset_pipeline queue) is connected + has Blender
     render_worker_connected = False
     blender_available = False
 
@@ -534,10 +534,10 @@ async def render_health(
     else:
         all_workers = list(inspect_result.get("ping", {}).keys())
         details["workers"] = all_workers
-        # Find any worker consuming thumbnail_rendering queue
+        # Find any worker consuming asset_pipeline queue
         for worker_name, queues in inspect_result.get("active_queues", {}).items():
             queue_names = [q.get("name") for q in (queues or [])]
-            if "thumbnail_rendering" in queue_names:
+            if "asset_pipeline" in queue_names:
                 render_worker_connected = True
                 # render-worker always has Blender — it starts Blender successfully
                 blender_available = True
@@ -547,11 +547,11 @@ async def render_health(
             render_worker_connected = True
             details["worker_detection"] = "fallback"
 
-    # 3. Queue depth for thumbnail_rendering
+    # 3. Queue depth for asset_pipeline
     thumbnail_queue_depth = 0
     try:
         r = redis_lib.from_url(app_settings.redis_url, decode_responses=True)
-        thumbnail_queue_depth = r.llen("thumbnail_rendering") or 0
+        thumbnail_queue_depth = r.llen("asset_pipeline") or 0
     except Exception as exc:
         details["redis_error"] = str(exc)
 

backend/app/domains/materials/tasks.py

@@ -18,7 +18,7 @@ CATALOG_SCRIPT = Path(os.environ.get("RENDER_SCRIPTS_DIR", "/render-scripts")) /
 
 @celery_app.task(
     name="app.domains.materials.tasks.refresh_asset_library_catalog",
-    queue="thumbnail_rendering",
+    queue="asset_pipeline",
     bind=True,
     max_retries=2,
     default_retry_delay=30,

backend/app/domains/media/models.py

@@ -14,8 +14,8 @@ class MediaAssetType(str, enum.Enum):
     turntable = "turntable"
     stl_low = "stl_low"
     stl_high = "stl_high"
-    gltf_geometry = "gltf_geometry"
-    gltf_production = "gltf_production"
+    gltf_geometry = "gltf_geometry"      # DEPRECATED: use usd_master — viewer GLB auto-generated as part of USD pipeline
+    gltf_production = "gltf_production"  # DEPRECATED: use usd_master — high-quality production GLB superseded by USD master
     blend_production = "blend_production"
     usd_master = "usd_master"
 

backend/app/domains/pipeline/tasks/export_glb.py

@@ -13,7 +13,7 @@ from app.core.pipeline_logger import PipelineLogger
 logger = logging.getLogger(__name__)
 
 
-@celery_app.task(bind=True, name="app.tasks.step_tasks.generate_gltf_geometry_task", queue="thumbnail_rendering", max_retries=1)
+@celery_app.task(bind=True, name="app.tasks.step_tasks.generate_gltf_geometry_task", queue="asset_pipeline", max_retries=1)
 def generate_gltf_geometry_task(self, cad_file_id: str):
     """Export a geometry GLB directly from STEP via OCC (no STL intermediary).
 
@@ -83,25 +83,47 @@ def generate_gltf_geometry_task(self, cad_file_id: str):
             settings_rows = session.execute(_select(_SysSetting)).scalars().all()
             sys_settings = {s.key: s.value for s in settings_rows}
 
-            # Hash-based cache check: skip tessellation if file hasn't changed
-            step_file_hash = cad_file.step_file_hash
-            if step_file_hash:
-                from app.domains.media.models import MediaAsset, MediaAssetType
-                import uuid as _uuid_check
+            linear_deflection = float(sys_settings.get("scene_linear_deflection", "0.1"))
+            angular_deflection = float(sys_settings.get("scene_angular_deflection", "0.1"))
+            tessellation_engine = sys_settings.get("tessellation_engine", "occ")
+
+            # Hash-based cache check: skip tessellation if file and settings haven't changed
+            from app.domains.products.cache_service import compute_step_hash as _compute_step_hash
+            from app.domains.media.models import MediaAsset, MediaAssetType
+            import uuid as _uuid_check
+            _current_hash = _compute_step_hash(str(step_path_str))
+            _cache_hit_asset_id = None
+
+            # Composite cache key includes deflection settings so changing them invalidates cache
+            effective_cache_key = (
+                f"{_current_hash}:{linear_deflection}:{angular_deflection}:{tessellation_engine}"
+                if _current_hash else None
+            )
+
+            if effective_cache_key:
                 existing_geo = session.execute(
                     _select(MediaAsset).where(
                         MediaAsset.cad_file_id == _uuid_check.UUID(cad_file_id),
                         MediaAsset.asset_type == MediaAssetType.gltf_geometry,
                     )
                 ).scalars().first()
-                if existing_geo:
-                    logger.info("[CACHE] hash match — skipping geometry GLB tessellation for %s", cad_file_id)
-                    pl.step_done("export_glb_geometry", result={"cached": True, "asset_id": str(existing_geo.id)})
-                    _cache_hit_asset_id = str(existing_geo.id)
+                stored_key = (existing_geo.render_config or {}).get("cache_key", "") if existing_geo else ""
+                if stored_key == effective_cache_key:
+                    _asset_disk_path = _Path(app_settings.upload_dir) / existing_geo.storage_key
+                    if _asset_disk_path.exists():
+                        logger.info("[CACHE] cache key match — skipping geometry GLB tessellation for %s", cad_file_id)
+                        pl.step_done("export_glb_geometry", result={"cached": True, "asset_id": str(existing_geo.id)})
+                        _cache_hit_asset_id = str(existing_geo.id)
+                    else:
+                        logger.info("[CACHE] cache key match but asset missing on disk — re-running tessellation for %s", cad_file_id)
                 else:
-                    _cache_hit_asset_id = None
+                    # Cache miss: update stored hash so next run can use it
+                    cad_file.step_file_hash = _current_hash
+                    session.commit()
             else:
-                _cache_hit_asset_id = None
+                # No hash available: update stored hash and proceed
+                cad_file.step_file_hash = _current_hash
+                session.commit()
         eng.dispose()
 
         if _cache_hit_asset_id is not None:
@@ -112,10 +134,6 @@ def generate_gltf_geometry_task(self, cad_file_id: str):
                 logger.debug("Could not queue generate_usd_master_task from cache-hit path (non-fatal)")
             return {"cached": True, "asset_id": _cache_hit_asset_id}
 
-        linear_deflection = float(sys_settings.get("scene_linear_deflection", "0.1"))
-        angular_deflection = float(sys_settings.get("scene_angular_deflection", "0.1"))
-        tessellation_engine = sys_settings.get("tessellation_engine", "occ")
-
         step = _Path(step_path_str)
 
         if not step.exists():
@@ -197,6 +215,7 @@ def generate_gltf_geometry_task(self, cad_file_id: str):
                 existing.storage_key = _key
                 existing.mime_type = "model/gltf-binary"
                 existing.file_size_bytes = _file_size
+                existing.render_config = {"cache_key": effective_cache_key}
                 if product_id:
                     existing.product_id = _uuid.UUID(product_id)
                 _sess.commit()
@@ -209,6 +228,7 @@ def generate_gltf_geometry_task(self, cad_file_id: str):
                     storage_key=_key,
                     mime_type="model/gltf-binary",
                     file_size_bytes=_file_size,
+                    render_config={"cache_key": effective_cache_key},
                 )
                 _sess.add(asset)
                 _sess.commit()
@@ -234,7 +254,7 @@ def generate_gltf_geometry_task(self, cad_file_id: str):
 @celery_app.task(
     bind=True,
     name="app.tasks.step_tasks.generate_gltf_production_task",
-    queue="thumbnail_rendering",
+    queue="asset_pipeline",
     max_retries=2,
 )
 def generate_gltf_production_task(self, cad_file_id: str, product_id: str | None = None) -> dict:
@@ -511,7 +531,7 @@ def generate_gltf_production_task(self, cad_file_id: str, product_id: str | None
 @celery_app.task(
     bind=True,
     name="app.tasks.step_tasks.generate_usd_master_task",
-    queue="thumbnail_rendering",
+    queue="asset_pipeline",  # needs pxr (usd-core) + OCC — both only in render-worker
     max_retries=1,
 )
 def generate_usd_master_task(self, cad_file_id: str) -> dict:
@@ -583,19 +603,44 @@ def generate_usd_master_task(self, cad_file_id: str) -> dict:
             settings_rows = sess.execute(_sel(SystemSetting)).scalars().all()
             sys_settings = {s.key: s.value for s in settings_rows}
 
-            # Hash-based cache check: skip tessellation if file hasn't changed
-            step_file_hash = cad_file.step_file_hash
-            if step_file_hash:
+            linear_deflection = float(sys_settings.get("render_linear_deflection", "0.03"))
+            angular_deflection = float(sys_settings.get("render_angular_deflection", "0.05"))
+            sharp_threshold = float(sys_settings.get("sharp_edge_threshold", "20.0"))
+
+            # Hash-based cache check: skip tessellation if file and settings haven't changed
+            from app.domains.products.cache_service import compute_step_hash as _compute_step_hash_usd
+            _current_hash_usd = _compute_step_hash_usd(str(step_path))
+
+            # Composite cache key includes deflection settings so changing them invalidates cache
+            effective_cache_key = (
+                f"{_current_hash_usd}:{linear_deflection}:{angular_deflection}:{sharp_threshold}"
+                if _current_hash_usd else None
+            )
+
+            if effective_cache_key:
                 existing_usd = sess.execute(
                     _sel(MediaAsset).where(
                         MediaAsset.cad_file_id == cad_file.id,
                         MediaAsset.asset_type == MediaAssetType.usd_master,
                     )
                 ).scalars().first()
-                if existing_usd:
-                    logger.info("[CACHE] hash match — skipping USD master tessellation for %s", cad_file_id)
-                    pl.step_done("usd_master", result={"cached": True, "asset_id": str(existing_usd.id)})
-                    _cache_hit_asset_id = str(existing_usd.id)
+                stored_key = (existing_usd.render_config or {}).get("cache_key", "") if existing_usd else ""
+                if stored_key == effective_cache_key:
+                    _usd_disk_path = _Path(app_settings.upload_dir) / existing_usd.storage_key
+                    if _usd_disk_path.exists():
+                        logger.info("[CACHE] cache key match — skipping USD master tessellation for %s", cad_file_id)
+                        pl.step_done("usd_master", result={"cached": True, "asset_id": str(existing_usd.id)})
+                        _cache_hit_asset_id = str(existing_usd.id)
+                    else:
+                        logger.info("[CACHE] cache key match but USD asset missing on disk — re-running tessellation for %s", cad_file_id)
+                else:
+                    # Cache miss: update stored hash so next run can use it
+                    cad_file.step_file_hash = _current_hash_usd
+                    sess.commit()
+            else:
+                # No hash available: update stored hash and proceed
+                cad_file.step_file_hash = _current_hash_usd
+                sess.commit()
         eng.dispose()
 
         if _cache_hit_asset_id is not None:
@@ -606,10 +651,6 @@ def generate_usd_master_task(self, cad_file_id: str) -> dict:
             pl.step_error("usd_master", err, None)
             raise RuntimeError(err)
 
-        linear_deflection = float(sys_settings.get("render_linear_deflection", "0.03"))
-        angular_deflection = float(sys_settings.get("render_angular_deflection", "0.05"))
-        sharp_threshold = float(sys_settings.get("sharp_edge_threshold", "20.0"))
-
         output_path = step_path.parent / f"{step_path.stem}_master.usd"
         scripts_dir = _Path(_os.environ.get("RENDER_SCRIPTS_DIR", "/render-scripts"))
         script_path = scripts_dir / "export_step_to_usd.py"
@@ -675,6 +716,7 @@ def generate_usd_master_task(self, cad_file_id: str) -> dict:
                 existing.storage_key = _key
                 existing.mime_type = "model/vnd.usd"
                 existing.file_size_bytes = _file_size
+                existing.render_config = {"cache_key": effective_cache_key}
                 sess2.commit()
                 asset_id = str(existing.id)
             else:
@@ -684,6 +726,7 @@ def generate_usd_master_task(self, cad_file_id: str) -> dict:
                     storage_key=_key,
                     mime_type="model/vnd.usd",
                     file_size_bytes=_file_size,
+                    render_config={"cache_key": effective_cache_key},
                 )
                 sess2.add(asset)
                 sess2.commit()

backend/app/domains/pipeline/tasks/extract_metadata.py

@@ -104,7 +104,7 @@ def process_step_file(self, cad_file_id: str):
         pl.info("process_step_file", f"Processing STEP file (metadata only): {cad_file_id}")
         try:
             from app.services.step_processor import extract_cad_metadata
-            extract_cad_metadata(cad_file_id)
+            extract_cad_metadata(cad_file_id, tenant_id=_tenant_id)
         except Exception as exc:
             pl.step_error("process_step_file", f"STEP metadata extraction failed: {exc}", exc)
             r.delete(lock_key)  # release lock so a retry can proceed
@@ -119,7 +119,7 @@ def process_step_file(self, cad_file_id: str):
     render_step_thumbnail.delay(cad_file_id)
 
 
-def _auto_populate_materials_for_cad(cad_file_id: str) -> None:
+def _auto_populate_materials_for_cad(cad_file_id: str, tenant_id: str | None = None) -> None:
     """Sync helper: auto-populate cad_part_materials from Excel for newly-processed CAD files.
 
     Only fills products where cad_part_materials is empty or all-blank,
@@ -132,10 +132,12 @@ def _auto_populate_materials_for_cad(cad_file_id: str) -> None:
     from app.models.product import Product
     from app.api.routers.products import build_materials_from_excel
     from app.services.step_processor import build_part_colors
+    from app.core.tenant_context import set_tenant_context_sync
 
     sync_url = app_settings.database_url.replace("+asyncpg", "")
     eng = create_engine(sync_url)
     with Session(eng) as session:
+        set_tenant_context_sync(session, tenant_id)
         # Load the CAD file to get parsed objects
         cad_file = session.execute(
             sql_select(CadFile).where(CadFile.id == cad_file_id)
@@ -201,7 +203,7 @@ def _auto_populate_materials_for_cad(cad_file_id: str) -> None:
     eng.dispose()
 
 
-@celery_app.task(name="app.tasks.step_tasks.reextract_cad_metadata", queue="thumbnail_rendering")
+@celery_app.task(name="app.tasks.step_tasks.reextract_cad_metadata", queue="asset_pipeline")
 def reextract_cad_metadata(cad_file_id: str):
     """Re-extract bounding-box dimensions for an already-completed CAD file.
 

backend/app/domains/pipeline/tasks/render_order_line.py

@@ -20,7 +20,7 @@ def dispatch_order_line_render(order_line_id: str):
     render_order_line_task.delay(order_line_id)
 
 
-@celery_app.task(bind=True, name="app.tasks.step_tasks.render_order_line_task", queue="thumbnail_rendering", max_retries=3)
+@celery_app.task(bind=True, name="app.tasks.step_tasks.render_order_line_task", queue="asset_pipeline", max_retries=3)
 def render_order_line_task(self, order_line_id: str):
     """Render a specific output type for an order line.
 

backend/app/domains/pipeline/tasks/render_thumbnail.py

@@ -14,11 +14,11 @@ from app.core.pipeline_logger import PipelineLogger
 logger = logging.getLogger(__name__)
 
 
-@celery_app.task(bind=True, name="app.tasks.step_tasks.render_step_thumbnail", queue="thumbnail_rendering")
+@celery_app.task(bind=True, name="app.tasks.step_tasks.render_step_thumbnail", queue="asset_pipeline")
 def render_step_thumbnail(self, cad_file_id: str):
     """Render the thumbnail for a freshly-processed STEP file.
 
-    Runs on the dedicated thumbnail_rendering queue (concurrency=1) so the
+    Runs on the dedicated asset_pipeline queue (concurrency=1) so the
     blender-renderer service is never overwhelmed by concurrent requests.
     On success, also auto-populates materials and marks the CadFile as completed.
     """
@@ -139,7 +139,7 @@ def render_step_thumbnail(self, cad_file_id: str):
     # Auto-populate materials now that parsed_objects are available
     try:
         from app.domains.pipeline.tasks.extract_metadata import _auto_populate_materials_for_cad
-        _auto_populate_materials_for_cad(cad_file_id)
+        _auto_populate_materials_for_cad(cad_file_id, tenant_id=_tenant_id)
     except Exception:
         logger.exception(
             f"Auto material population failed for cad_file {cad_file_id} (non-fatal)"
@@ -180,7 +180,7 @@ def render_step_thumbnail(self, cad_file_id: str):
     pl.step_done("render_step_thumbnail")
 
 
-@celery_app.task(bind=True, name="app.tasks.step_tasks.regenerate_thumbnail", queue="thumbnail_rendering")
+@celery_app.task(bind=True, name="app.tasks.step_tasks.regenerate_thumbnail", queue="asset_pipeline")
 def regenerate_thumbnail(self, cad_file_id: str, part_colors: dict):
     """Regenerate thumbnail with per-part colours."""
     pl = PipelineLogger(task_id=self.request.id)

backend/app/domains/rendering/tasks.py

@@ -1,6 +1,6 @@
 """Rendering domain tasks — Celery tasks for Blender-based rendering.
 
-These tasks run on the `thumbnail_rendering` queue in the render-worker
+These tasks run on the `asset_pipeline` queue in the render-worker
 container, which has Blender and cadquery available.
 
 Phase A2: Initial implementation replacing the blender-renderer HTTP service.
@@ -48,7 +48,7 @@ def _update_workflow_run_status(order_line_id: str, status: str, error: str | No
 @celery_app.task(
     bind=True,
     name="app.domains.rendering.tasks.render_still_task",
-    queue="thumbnail_rendering",
+    queue="asset_pipeline",
     max_retries=2,
 )
 def render_still_task(
@@ -150,7 +150,7 @@ def render_still_task(
 @celery_app.task(
     bind=True,
     name="app.domains.rendering.tasks.render_turntable_task",
-    queue="thumbnail_rendering",
+    queue="asset_pipeline",
     max_retries=2,
 )
 def render_turntable_task(
@@ -391,7 +391,7 @@ def _resolve_step_path_for_order_line(order_line_id: str) -> tuple[str | None, s
 @celery_app.task(
     bind=True,
     name="app.domains.rendering.tasks.render_order_line_still_task",
-    queue="thumbnail_rendering",
+    queue="asset_pipeline",
     max_retries=2,
 )
 def render_order_line_still_task(self, order_line_id: str, **params) -> dict:
@@ -509,7 +509,7 @@ def render_order_line_still_task(self, order_line_id: str, **params) -> dict:
 @celery_app.task(
     bind=True,
     name="app.domains.rendering.tasks.export_gltf_for_order_line_task",
-    queue="thumbnail_rendering",
+    queue="asset_pipeline",
     max_retries=1,
 )
 def export_gltf_for_order_line_task(self, order_line_id: str) -> dict:
@@ -555,7 +555,7 @@ def export_gltf_for_order_line_task(self, order_line_id: str) -> dict:
 @celery_app.task(
     bind=True,
     name="app.domains.rendering.tasks.export_blend_for_order_line_task",
-    queue="thumbnail_rendering",
+    queue="asset_pipeline",
     max_retries=1,
 )
 def export_blend_for_order_line_task(self, order_line_id: str) -> dict:
@@ -646,7 +646,7 @@ def export_blend_for_order_line_task(self, order_line_id: str) -> dict:
 @celery_app.task(
     bind=True,
     name="app.domains.rendering.tasks.apply_asset_library_materials_task",
-    queue="thumbnail_rendering",
+    queue="asset_pipeline",
     max_retries=1,
 )
 def apply_asset_library_materials_task(self, order_line_id: str, asset_library_id: str) -> dict:

backend/app/services/part_key_service.py

@@ -105,6 +105,10 @@ def build_scene_manifest(cad_file, usd_asset=None) -> dict:
         object_names: list[str] = cad_file.parsed_objects.get("objects") or []
         seen_keys: set[str] = set()
         for source_name in object_names:
+            # Fallback: USD master not yet generated. Use source_name as xcaf_path proxy.
+            # Note: slugs produced here may differ from what export_step_to_usd.py will
+            # produce for unnamed parts (which use sha256 of the XCAF hierarchy path).
+            # Named parts will match once USD master is generated.
             part_key = generate_part_key(source_name, source_name, seen_keys)
             effective_material, provenance = _resolve_material(
                 part_key, source_name, manual, resolved, source

backend/app/services/render_blender.py

@@ -15,8 +15,8 @@ from pathlib import Path
 logger = logging.getLogger(__name__)
 
 
-def _glb_from_step(step_path: Path, glb_path: Path) -> None:
-    """Convert STEP → GLB via OCC (export_step_to_gltf.py, no Blender needed)."""
+def _glb_from_step(step_path: Path, glb_path: Path, tessellation_engine: str = "occ") -> None:
+    """Convert STEP → GLB via OCC or GMSH (export_step_to_gltf.py, no Blender needed)."""
     import subprocess
     import sys as _sys
 
@@ -32,12 +32,13 @@ def _glb_from_step(step_path: Path, glb_path: Path) -> None:
         "--output_path", str(glb_path),
         "--linear_deflection", str(linear_deflection),
         "--angular_deflection", str(angular_deflection),
+        "--tessellation_engine", tessellation_engine,
     ]
     result = subprocess.run(cmd, capture_output=True, text=True, timeout=120)
     for line in result.stdout.splitlines():
-        logger.info("[occ-gltf] %s", line)
+        logger.info("[export-gltf] %s", line)
     for line in result.stderr.splitlines():
-        logger.warning("[occ-gltf stderr] %s", line)
+        logger.warning("[export-gltf stderr] %s", line)
     if result.returncode != 0 or not glb_path.exists() or glb_path.stat().st_size == 0:
         raise RuntimeError(
             f"export_step_to_gltf.py failed (exit {result.returncode}).\n"
@@ -90,8 +91,9 @@ def render_still(
     mesh_attributes: dict | None = None,
     log_callback: "Callable[[str], None] | None" = None,
     usd_path: "Path | None" = None,
+    tessellation_engine: str = "occ",
 ) -> dict:
-    """Convert STEP → GLB (OCC) → PNG (Blender subprocess).
+    """Convert STEP → GLB (OCC or GMSH) → PNG (Blender subprocess).
 
     When usd_path is provided and the file exists, the GLB conversion step is
     skipped and Blender imports the USD stage directly (--usd-path flag).
@@ -125,7 +127,7 @@ def render_still(
         glb_size_bytes = 0
     else:
         if not glb_path.exists() or glb_path.stat().st_size == 0:
-            _glb_from_step(step_path, glb_path)
+            _glb_from_step(step_path, glb_path, tessellation_engine)
         else:
             logger.info("GLB local hit: %s (%d KB)", glb_path.name, glb_path.stat().st_size // 1024)
         glb_size_bytes = glb_path.stat().st_size if glb_path.exists() else 0
@@ -310,6 +312,7 @@ def render_turntable_to_file(
     rotation_y: float = 0.0,
     rotation_z: float = 0.0,
     usd_path: "Path | None" = None,
+    tessellation_engine: str = "occ",
 ) -> dict:
     """Render a turntable animation: STEP → STL → N frames (Blender) → mp4 (ffmpeg).
 
@@ -349,7 +352,7 @@ def render_turntable_to_file(
         logger.info("[render_blender] turntable using USD path: %s", usd_path)
     else:
         if not glb_path.exists() or glb_path.stat().st_size == 0:
-            _glb_from_step(step_path, glb_path)
+            _glb_from_step(step_path, glb_path, tessellation_engine)
         else:
             logger.info("GLB local hit: %s (%d KB)", glb_path.name, glb_path.stat().st_size // 1024)
     glb_duration_s = round(time.monotonic() - t_glb, 2)

backend/app/services/step_processor.py

@@ -79,7 +79,7 @@ def match_cad_to_items(
     return matched
 
 
-def extract_cad_metadata(cad_file_id: str) -> None:
+def extract_cad_metadata(cad_file_id: str, tenant_id: str | None = None) -> None:
     """
     Fast metadata extraction for a CAD file (no thumbnail generation).
 
@@ -94,9 +94,11 @@ def extract_cad_metadata(cad_file_id: str) -> None:
     from sqlalchemy import create_engine
     from sqlalchemy.orm import Session
     from app.models.cad_file import CadFile, ProcessingStatus
+    from app.core.tenant_context import set_tenant_context_sync
 
     engine = create_engine(settings.database_url_sync)
     with Session(engine) as session:
+        set_tenant_context_sync(session, tenant_id)
         cad_file = session.get(CadFile, uuid.UUID(cad_file_id))
         if not cad_file:
             logger.error(f"CAD file not found: {cad_file_id}")
@@ -452,6 +454,7 @@ def _get_all_settings() -> dict[str, str]:
         "thumbnail_format":       "jpg",
         "blender_smooth_angle":   "30",
         "cycles_device":          "auto",
+        "tessellation_engine":    "occ",
     }
     try:
         from app.config import settings as app_settings
@@ -533,6 +536,7 @@ def _generate_thumbnail(
                     samples=samples,
                     smooth_angle=int(settings["blender_smooth_angle"]),
                     cycles_device=settings["cycles_device"],
+                    tessellation_engine=settings["tessellation_engine"],
                 )
                 rendered_png = tmp_png if tmp_png.exists() else None
             except Exception as exc:
@@ -642,6 +646,7 @@ def render_to_file(
     denoising_use_gpu: str = "",
     order_line_id: str | None = None,
     usd_path: "Path | None" = None,
+    tessellation_engine: str | None = None,
 ) -> tuple[bool, dict]:
     """Render a STEP file to a specific output path using current system settings.
 
@@ -777,6 +782,7 @@ def render_to_file(
                     denoising_use_gpu=denoising_use_gpu,
                     log_callback=_log_cb,
                     usd_path=usd_path,
+                    tessellation_engine=tessellation_engine or settings["tessellation_engine"],
                 )
                 rendered_png = tmp_png if tmp_png.exists() else None
             except Exception as exc:

backend/app/tasks/beat_tasks.py

@@ -73,7 +73,7 @@ def broadcast_queue_status() -> None:
         r = sync_redis.from_url(settings.redis_url, decode_responses=True)
         depths = {
             "step_processing": r.llen("step_processing"),
-            "thumbnail_rendering": r.llen("thumbnail_rendering"),
+            "asset_pipeline": r.llen("asset_pipeline"),
         }
         event = {"type": "queue_update", "depths": depths}
         r.publish("__broadcast__", json.dumps(event))

backend/app/tasks/celery_app.py

@@ -30,7 +30,7 @@ celery_app.conf.update(
     enable_utc=True,
     task_routes={
         "app.domains.pipeline.tasks.*": {"queue": "step_processing"},
-        "app.domains.rendering.tasks.*": {"queue": "thumbnail_rendering"},
+        "app.domains.rendering.tasks.*": {"queue": "asset_pipeline"},
         "app.tasks.beat_tasks.*": {"queue": "step_processing"},
         "app.tasks.ai_tasks.*": {"queue": "ai_validation"},
         # Legacy task names (shim) — keep until old queued tasks drain

backend/app/tasks/gpu_tasks.py

@@ -5,7 +5,7 @@ from app.tasks.celery_app import celery_app
 logger = logging.getLogger(__name__)
 
 
-@celery_app.task(name="app.tasks.gpu_tasks.probe_gpu", queue="thumbnail_rendering")
+@celery_app.task(name="app.tasks.gpu_tasks.probe_gpu", queue="asset_pipeline")
 def probe_gpu() -> dict:
     """Run Blender GPU probe on the render-worker. Stores result in system_settings."""
     import subprocess

backend/tests/conftest.py

@@ -230,7 +230,6 @@ def mock_celery_tasks(monkeypatch):
         "app.domains.materials.tasks.refresh_asset_library_catalog",
         "app.tasks.step_tasks.process_step_file",
         "app.tasks.step_tasks.render_step_thumbnail",
-        "app.tasks.step_tasks.generate_stl_cache",
         "app.domains.imports.tasks.validate_excel_import",
         "app.domains.rendering.tasks.render_still_task",
         "app.domains.rendering.tasks.render_turntable_task",

backend/tests/domains/test_rendering_service.py

@@ -99,14 +99,14 @@ def test_generate_gltf_geometry_task_importable():
 def test_render_order_line_still_task_importable():
     from app.domains.rendering.tasks import render_order_line_still_task
     assert render_order_line_still_task.name == "app.domains.rendering.tasks.render_order_line_still_task"
-    assert render_order_line_still_task.queue == "thumbnail_rendering"
+    assert render_order_line_still_task.queue == "asset_pipeline"
 
 
 def test_export_gltf_for_order_line_task_importable():
     from app.domains.rendering.tasks import export_gltf_for_order_line_task
-    assert export_gltf_for_order_line_task.queue == "thumbnail_rendering"
+    assert export_gltf_for_order_line_task.queue == "asset_pipeline"
 
 
 def test_export_blend_for_order_line_task_importable():
     from app.domains.rendering.tasks import export_blend_for_order_line_task
-    assert export_blend_for_order_line_task.queue == "thumbnail_rendering"
+    assert export_blend_for_order_line_task.queue == "asset_pipeline"

docker-compose.yml

@@ -121,7 +121,7 @@ services:
       dockerfile: render-worker/Dockerfile
       args:
         - BLENDER_VERSION=${BLENDER_VERSION:-5.0.1}
-    command: bash -c "python3 /check_version.py && celery -A app.tasks.celery_app worker --loglevel=info -Q thumbnail_rendering --autoscale=1,1 --concurrency=1"
+    command: bash -c "python3 /check_version.py && celery -A app.tasks.celery_app worker --loglevel=info -Q asset_pipeline --autoscale=1,1 --concurrency=1"
     environment:
       - POSTGRES_DB=${POSTGRES_DB:-schaeffler}
       - POSTGRES_USER=${POSTGRES_USER:-schaeffler}

frontend/src/__tests__/pages/WorkerActivity.test.tsx

@@ -25,18 +25,18 @@ describe('worker API types', () => {
   test('CeleryWorker interface shape', () => {
     const worker = {
       name: 'celery@worker1',
-      queues: ['thumbnail_rendering'],
+      queues: ['asset_pipeline'],
       active_task_count: 2,
       active_tasks: [{ name: 'render_still_task', id: 'abc' }],
       total_tasks_processed: { render_still_task: 42 },
     }
-    expect(worker.queues).toContain('thumbnail_rendering')
+    expect(worker.queues).toContain('asset_pipeline')
     expect(worker.active_tasks).toHaveLength(1)
   })
 
   test('QueueStatus interface shape', () => {
     const qs = {
-      queue_depths: { step_processing: 3, thumbnail_rendering: 0 },
+      queue_depths: { step_processing: 3, asset_pipeline: 0 },
       pending_count: 3,
       active: [],
       reserved: [],

frontend/src/api/orders.ts

@@ -276,6 +276,18 @@ export async function generateLinesFromItems(
   return res.data
 }
 
+export async function rejectOrderLine(
+  orderId: string,
+  lineId: string,
+  reason: string,
+): Promise<{ rejected: boolean; line_id: string; reason: string }> {
+  const res = await api.post<{ rejected: boolean; line_id: string; reason: string }>(
+    `/orders/${orderId}/lines/${lineId}/reject`,
+    { reason },
+  )
+  return res.data
+}
+
 export async function rejectOrder(orderId: string, reason: string, notifyClient: boolean = true): Promise<Order> {
   const res = await api.post<Order>(`/orders/${orderId}/reject`, {
     reason,

frontend/src/components/cad/ThreeDViewer.tsx

@@ -537,6 +537,17 @@ export default function ThreeDViewer({
     const map = glbExtras.partKeyMap as Record<string, string> | undefined
     if (map && Object.keys(map).length > 0) {
       setPartKeyMap(map)
+      // Task 2: Stamp userData.partKey on every mesh (fallback for meshes whose
+      // GLB node extras were not populated — e.g. files generated before Task 1).
+      // For new GLBs, Three.js already set userData.partKey from node extras;
+      // the guard `if (obj.userData.partKey) return` avoids overwriting it.
+      sceneRef.current.traverse((obj) => {
+        if (!(obj instanceof THREE.Mesh)) return
+        if (obj.userData.partKey) return // already set by GLB node extras
+        const normalized = normalizeMeshName((obj.userData?.name as string) || obj.name)
+        const pk = map[normalized] ?? normalized
+        if (pk) obj.userData.partKey = pk
+      })
     }
 
     const names = new Set<string>()
@@ -679,8 +690,11 @@ export default function ThreeDViewer({
     e.stopPropagation()
     const mesh = e.object as THREE.Mesh
     const raw = (mesh?.userData?.name as string) || mesh?.name || (mesh?.parent?.userData?.name as string) || mesh?.parent?.name || ''
-    const name = normalizeMeshName(raw) || 'Part'
-    setHoverInfo({ name, x: e.nativeEvent.clientX, y: e.nativeEvent.clientY })
+    const normalized = normalizeMeshName(raw) || 'Part'
+    // Task 3: prefer userData.partKey (set by GLB node extras or Task 2 stamp) over
+    // raw normalized name so tooltip shows canonical slug (e.g. "ring_outer") not OCC name
+    const displayName = (mesh?.userData?.partKey as string | undefined) ?? resolvePartKey(normalized)
+    setHoverInfo({ name: displayName, x: e.nativeEvent.clientX, y: e.nativeEvent.clientY })
 
     // Restore previous hovered mesh (array-safe)
     if (hoveredMeshRef.current && hoveredMeshRef.current !== mesh) {
@@ -703,7 +717,7 @@ export default function ThreeDViewer({
       const mat = m as THREE.MeshStandardMaterial
       if (mat && 'emissive' in mat) { mat.emissive.set(0x333333); mat.emissiveIntensity = 0.5 }
     })
-  }, [showUnassigned])
+  }, [showUnassigned, resolvePartKey])
 
   const handlePointerOut = useCallback(() => {
     setHoverInfo(null)

frontend/src/components/layout/NotificationCenter.tsx

@@ -59,6 +59,41 @@ const ACTION_CONFIG: Record<string, { icon: typeof Bell; label: (d: Record<strin
   },
 }
 
+type NotifGroup =
+  | { kind: 'single'; item: Notification }
+  | { kind: 'batch'; count: number; failed: number; entityId: string | null; latest: Notification; ids: string[] }
+
+function groupNotifications(items: Notification[]): NotifGroup[] {
+  const result: NotifGroup[] = []
+  let i = 0
+  while (i < items.length) {
+    const n = items[i]
+    const isRender = n.action === 'render.completed' || n.action === 'render.failed'
+    if (isRender) {
+      const t0 = new Date(n.timestamp).getTime()
+      let j = i
+      let done = 0; let failed = 0
+      const batchIds: string[] = []
+      while (j < items.length) {
+        const m = items[j]
+        const isRenderM = m.action === 'render.completed' || m.action === 'render.failed'
+        const tM = new Date(m.timestamp).getTime()
+        if (!isRenderM || m.entity_id !== n.entity_id || Math.abs(tM - t0) > 5 * 60 * 1000) break
+        batchIds.push(m.id)
+        if (m.action === 'render.failed') failed++; else done++
+        j++
+      }
+      if (j - i >= 2) {
+        result.push({ kind: 'batch', count: done, failed, entityId: n.entity_id ?? null, latest: n, ids: batchIds })
+        i = j; continue
+      }
+    }
+    result.push({ kind: 'single', item: n })
+    i++
+  }
+  return result
+}
+
 function relativeTime(ts: string): string {
   const diff = Date.now() - new Date(ts).getTime()
   const seconds = Math.floor(diff / 1000)
@@ -182,7 +217,44 @@ export default function NotificationCenter() {
             {!data?.items.length && (
               <div className="py-8 text-center text-sm text-content-muted">No notifications</div>
             )}
-            {data?.items.map((n) => {
+            {data?.items && groupNotifications(data.items).map((group) => {
+              if (group.kind === 'batch') {
+                const { count, failed, entityId, latest, ids } = group
+                const BatchIcon = failed > 0 ? AlertTriangle : CheckCircle
+                const batchColor = failed > 0 ? 'text-red-500' : 'text-status-success-text'
+                const batchLabel = `Render batch: ${count} done${failed > 0 ? `, ${failed} failed` : ''}`
+                return (
+                  <button
+                    key={`batch-${latest.id}`}
+                    onClick={() => {
+                      if (!latest.read_at) {
+                        ids.forEach((id) => markOneMutation.mutate(id))
+                      }
+                      if (entityId) navigate(`/orders/${entityId}`)
+                      setOpen(false)
+                    }}
+                    className={clsx(
+                      'w-full flex items-start gap-3 px-4 py-3 text-left hover:bg-surface-hover transition-colors border-b border-border-light',
+                      !latest.read_at && 'bg-status-info-bg',
+                    )}
+                  >
+                    <BatchIcon size={16} className={clsx('mt-0.5 shrink-0', batchColor)} />
+                    <div className="flex-1 min-w-0">
+                      <p className={clsx('text-sm', !latest.read_at ? 'font-medium text-content' : 'text-content-secondary')}>
+                        {batchLabel}
+                      </p>
+                      {entityId && (
+                        <p className="text-xs text-content-muted mt-0.5">Click to view order</p>
+                      )}
+                      <p className="text-xs text-content-muted mt-0.5">{relativeTime(latest.timestamp)}</p>
+                    </div>
+                    {!latest.read_at && (
+                      <span className="mt-1.5 w-2 h-2 rounded-full bg-blue-500 shrink-0" />
+                    )}
+                  </button>
+                )
+              }
+              const n = group.item
               const cfg = ACTION_CONFIG[n.action] ?? {
                 icon: Bell,
                 label: () => n.action,

frontend/src/help/helpTexts.ts

@@ -37,7 +37,7 @@ export const HELP_TEXTS: Record<string, HelpText> = {
   },
   'action.regenerate_thumbnails': {
     title: 'Regenerate All Thumbnails',
-    body: 'Re-renders thumbnails for all STEP files using current renderer settings. Queues every file on the thumbnail_rendering worker.',
+    body: 'Re-renders thumbnails for all STEP files using current renderer settings. Queues every file on the asset_pipeline worker.',
     warning: 'This queues a large number of tasks. Only run during off-peak hours.',
   },
   'action.process_unprocessed': {

frontend/src/pages/Admin.tsx

@@ -138,6 +138,7 @@ export default function AdminPage() {
 
   const [tessellationDraft, setTessellationDraft] = useState<Partial<Settings>>({})
   const tess = { ...settings, ...tessellationDraft } as Settings
+  const [showAdvancedTess, setShowAdvancedTess] = useState(false)
 
   const { data: rendererStatus, refetch: refetchStatus } = useQuery({
     queryKey: ['renderer-status'],
@@ -1332,6 +1333,7 @@ export default function AdminPage() {
                 />
                 <span className="text-sm text-content">Apply Laplacian smoothing on export</span>
               </label>
+              <p className="text-xs text-content-muted mt-1">Smooths surface normals during GLB export for a less faceted look in the 3D viewer.</p>
             </div>
           </div>
 
@@ -1348,8 +1350,10 @@ export default function AdminPage() {
                 max="10000"
                 value={viewer3d.viewer_max_distance ?? 50}
                 onChange={e => setViewerDraft(d => ({ ...d, viewer_max_distance: parseFloat(e.target.value) }))}
+                title="Maximum camera distance from the model in the 3D viewer (in metres after mm→m conversion). Default: 50"
                 className="input w-full"
               />
+              <p className="text-xs text-content-muted mt-1">Maximum camera pull-back distance in the 3D viewer (metres).</p>
             </div>
             <div>
               <label className="text-sm font-medium text-content-muted block mb-1">
@@ -1362,8 +1366,10 @@ export default function AdminPage() {
                 max="1"
                 value={viewer3d.viewer_min_distance ?? 0.001}
                 onChange={e => setViewerDraft(d => ({ ...d, viewer_min_distance: parseFloat(e.target.value) }))}
+                title="Minimum camera distance from the model in the 3D viewer (in metres). Default: 0.001. Prevents clipping into the geometry."
                 className="input w-full"
               />
+              <p className="text-xs text-content-muted mt-1">Closest the camera can zoom in (metres). Prevents clipping through geometry.</p>
             </div>
           </div>
 
@@ -1376,11 +1382,13 @@ export default function AdminPage() {
               <select
                 value={viewer3d.gltf_material_quality ?? 'pbr_colors'}
                 onChange={e => setViewerDraft(d => ({ ...d, gltf_material_quality: e.target.value }))}
+                title="Controls what material data is embedded in exported GLB files. 'None' exports bare geometry; 'PBR Colors' bakes part colours into PBR materials."
                 className="input w-full"
               >
                 <option value="none">None (geometry only)</option>
                 <option value="pbr_colors">PBR Colors (from part colors)</option>
               </select>
+              <p className="text-xs text-content-muted mt-1">Material data embedded in exported GLB files.</p>
             </div>
             <div>
               <label className="text-sm font-medium text-content-muted block mb-1">
@@ -1393,8 +1401,10 @@ export default function AdminPage() {
                 max="1"
                 value={viewer3d.gltf_pbr_roughness ?? 0.4}
                 onChange={e => setViewerDraft(d => ({ ...d, gltf_pbr_roughness: parseFloat(e.target.value) }))}
+                title="Surface roughness for GLB PBR materials (0 = mirror-smooth, 1 = fully matte). Default: 0.4 — appropriate for brushed metal."
                 className="input w-full"
               />
+              <p className="text-xs text-content-muted mt-1">0 = mirror-smooth, 1 = fully matte. Default 0.4 suits brushed metal.</p>
             </div>
             <div>
               <label className="text-sm font-medium text-content-muted block mb-1">
@@ -1407,8 +1417,10 @@ export default function AdminPage() {
                 max="1"
                 value={viewer3d.gltf_pbr_metallic ?? 0.6}
                 onChange={e => setViewerDraft(d => ({ ...d, gltf_pbr_metallic: parseFloat(e.target.value) }))}
+                title="Metallic factor for GLB PBR materials (0 = dielectric/plastic, 1 = fully metallic). Default: 0.6 — suitable for steel parts."
                 className="input w-full"
               />
+              <p className="text-xs text-content-muted mt-1">0 = plastic/dielectric, 1 = fully metallic. Default 0.6 suits steel parts.</p>
             </div>
           </div>
 
@@ -1525,10 +1537,19 @@ export default function AdminPage() {
             </div>
           </div>
 
+          <button
+            onClick={() => setShowAdvancedTess(v => !v)}
+            className="text-xs text-accent hover:underline flex items-center gap-1 mt-1"
+          >
+            {showAdvancedTess ? <ChevronDown size={12} /> : <ChevronRight size={12} />}
+            {showAdvancedTess ? 'Hide manual values' : 'Advanced: manual deflection values'}
+          </button>
+
           {/* Manual inputs */}
+          {showAdvancedTess && (<>
           <div className="grid grid-cols-2 gap-6">
             <div className="space-y-4">
-              <p className="text-xs font-semibold text-content-secondary uppercase tracking-wide">Scene / Viewer</p>
+              <p className="text-xs font-semibold text-content-secondary uppercase tracking-wide">Scene (USD Master)</p>
               <div className="flex items-center gap-3">
                 <label className="text-sm text-content-secondary w-36 shrink-0">Linear deflection</label>
                 <input
@@ -1555,7 +1576,7 @@ export default function AdminPage() {
                 />
                 <span className="text-sm text-content-muted">rad</span>
               </div>
-              <p className="text-xs text-content-muted">Used for the 3D viewer (canonical scene). Smaller = smoother surfaces.</p>
+              <p className="text-xs text-content-muted">Used for the USD master + 3D viewer GLB (canonical scene). Smaller = smoother surfaces.</p>
             </div>
             <div className="space-y-4">
               <p className="text-xs font-semibold text-content-secondary uppercase tracking-wide">Render output</p>
@@ -1588,6 +1609,7 @@ export default function AdminPage() {
               <p className="text-xs text-content-muted">Used for final render output. Smaller = smoother surfaces, larger file sizes.</p>
             </div>
           </div>
+          </>)}
           <div className="flex gap-2">
             <button
               onClick={() => { updateSettingsMut.mutate(tessellationDraft); setTessellationDraft({}) }}

frontend/src/pages/MediaBrowser.tsx

@@ -504,13 +504,22 @@ export default function MediaBrowserPage() {
             <span className="text-sm">Loading assets…</span>
           </div>
         ) : items.length === 0 ? (
-          <div className="flex flex-col items-center justify-center h-64 text-content-muted gap-3">
-            <Image size={48} className="opacity-25" />
-            <p className="text-sm font-medium">No media assets found.</p>
-            <p className="text-xs text-center max-w-xs">
-              Renders will appear here once orders are completed. Try adjusting your filters.
-            </p>
-          </div>
+          (() => {
+            const hasActiveFilters = !!(q || assetType || categoryKey || renderStatus)
+            return (
+              <div className="flex flex-col items-center justify-center h-64 gap-3" style={{ color: 'var(--color-content-muted)' }}>
+                <Image size={48} style={{ opacity: 0.25 }} />
+                <p className="text-sm font-medium">
+                  {hasActiveFilters ? 'No assets match your filters.' : 'No assets yet — upload a STEP file to get started'}
+                </p>
+                {hasActiveFilters && (
+                  <p className="text-xs text-center max-w-xs">
+                    Try adjusting your search or filter settings.
+                  </p>
+                )}
+              </div>
+            )
+          })()
         ) : (
           <div className="grid grid-cols-2 sm:grid-cols-3 md:grid-cols-4 lg:grid-cols-5 xl:grid-cols-6 gap-4">
             {items.map(asset => (

frontend/src/pages/OrderDetail.tsx

@@ -1,6 +1,7 @@
 import { useQuery, useMutation, useQueryClient } from '@tanstack/react-query'
 import { useParams, useNavigate, Link } from 'react-router-dom'
 import { useState, useMemo, Fragment } from 'react'
+import { createPortal } from 'react-dom'
 import {
   ArrowLeft, Send, Trash2,
   FileBox, AlertTriangle, CheckCircle2, Image as ImageIcon, Unlink,
@@ -11,7 +12,7 @@ import {
   XCircle, RotateCw, Info,
 } from 'lucide-react'
 import { toast } from 'sonner'
-import { getOrder, submitOrder, deleteOrder, unlinkCadFile, regenerateItemThumbnail, patchOrderItem, removeOrderLine, dispatchRenders, cancelLineRender, cancelOrderRenders, splitMissingStep, generateLinesFromItems, downloadOrderRenders, rejectOrder, resubmitOrder } from '../api/orders'
+import { getOrder, submitOrder, deleteOrder, unlinkCadFile, regenerateItemThumbnail, patchOrderItem, removeOrderLine, dispatchRenders, cancelLineRender, cancelOrderRenders, splitMissingStep, generateLinesFromItems, downloadOrderRenders, rejectOrder, resubmitOrder, rejectOrderLine } from '../api/orders'
 import type { OrderItem, OrderLine } from '../api/orders'
 import { listOutputTypes } from '../api/outputTypes'
 import type { OutputType } from '../api/outputTypes'
@@ -827,6 +828,8 @@ function OrderLineRow({
 }) {
   const qc = useQueryClient()
   const [showInfo, setShowInfo] = useState(false)
+  const [rejectLineModalOpen, setRejectLineModalOpen] = useState(false)
+  const [rejectLineReason, setRejectLineReason] = useState('')
 
   const removeMut = useMutation({
     mutationFn: () => removeOrderLine(orderId, line.id),
@@ -843,7 +846,19 @@ function OrderLineRow({
     onError: (e: any) => toast.error(e.response?.data?.detail || 'Cancel failed'),
   })
 
+  const rejectLineMut = useMutation({
+    mutationFn: () => rejectOrderLine(orderId, line.id, rejectLineReason),
+    onSuccess: () => {
+      toast.success('Line rejected')
+      setRejectLineModalOpen(false)
+      setRejectLineReason('')
+      qc.invalidateQueries({ queryKey: ['order', orderId] })
+    },
+    onError: (e: any) => toast.error(e.response?.data?.detail || 'Reject failed'),
+  })
+
   const canCancel = isPrivileged && (line.render_status === 'processing' || line.render_status === 'pending') && line.output_type_id
+  const canRejectLine = isPrivileged && line.item_status !== 'rejected'
 
   const renderStatusColor: Record<string, string> = {
     pending: 'bg-surface-muted text-content-muted',
@@ -1003,7 +1018,21 @@ function OrderLineRow({
 
       {/* Item Status */}
       <td className="px-4 py-2">
-        <ItemStatusBadge status={line.item_status} />
+        <div className="flex items-center gap-1.5">
+          <ItemStatusBadge status={line.item_status} />
+          {canRejectLine && (
+            <button
+              onClick={(e) => {
+                e.stopPropagation()
+                setRejectLineModalOpen(true)
+              }}
+              className="text-content-muted hover:text-red-500 transition-colors"
+              title="Reject this line"
+            >
+              <XCircle size={12} />
+            </button>
+          )}
+        </div>
       </td>
 
       {/* Remove (draft only) */}
@@ -1027,6 +1056,61 @@ function OrderLineRow({
         renderStartedAt={line.render_started_at}
         renderCompletedAt={line.render_completed_at}
       />
+
+      {rejectLineModalOpen && createPortal(
+        <div
+          className="fixed inset-0 z-50 flex items-center justify-center"
+          style={{ backgroundColor: 'rgba(0,0,0,0.4)' }}
+          onClick={() => { setRejectLineModalOpen(false); setRejectLineReason('') }}
+        >
+          <div
+            className="rounded-xl shadow-2xl border border-border-default w-full max-w-md mx-4 p-6"
+            style={{ backgroundColor: 'var(--color-bg-surface)' }}
+            onClick={(e) => e.stopPropagation()}
+          >
+            <div className="flex items-center gap-3 mb-4">
+              <XCircle size={20} className="text-red-500 shrink-0" />
+              <h2 className="text-lg font-semibold text-content">Reject this item?</h2>
+              <button
+                onClick={() => { setRejectLineModalOpen(false); setRejectLineReason('') }}
+                className="ml-auto p-1 text-content-muted hover:text-content transition-colors rounded"
+              >
+                <X size={16} />
+              </button>
+            </div>
+            <div className="mb-4">
+              <p className="text-sm text-content-secondary mb-3">
+                Optionally provide a reason for rejecting this line.
+              </p>
+              <textarea
+                value={rejectLineReason}
+                onChange={(e) => setRejectLineReason(e.target.value)}
+                placeholder="Reason (optional)"
+                rows={3}
+                className="w-full rounded-lg border border-border-default px-3 py-2 text-sm text-content focus:outline-none focus:ring-2 focus:ring-accent resize-none"
+                style={{ backgroundColor: 'var(--color-bg-surface-alt)' }}
+              />
+            </div>
+            <div className="flex justify-end gap-3">
+              <button
+                onClick={() => { setRejectLineModalOpen(false); setRejectLineReason('') }}
+                className="btn-secondary"
+              >
+                Cancel
+              </button>
+              <button
+                onClick={() => rejectLineMut.mutate()}
+                disabled={rejectLineMut.isPending}
+                className="px-4 py-2 rounded-lg text-sm font-semibold bg-red-600 hover:bg-red-700 text-white transition-colors disabled:opacity-50 flex items-center gap-2"
+              >
+                <XCircle size={15} />
+                {rejectLineMut.isPending ? 'Rejecting…' : 'Confirm Rejection'}
+              </button>
+            </div>
+          </div>
+        </div>,
+        document.body,
+      )}
     </tr>
   )
 }
@@ -1611,10 +1695,12 @@ function SourceSpreadsheet({
               {sorted.map((item) => (
                 <tr
                   key={item.id}
-                  className={`hover:bg-surface-hover/30 group ${saving === item.id ? 'opacity-60' : ''}`}
+                  className={`group ${saving === item.id ? 'opacity-60' : ''}`}
+                  onMouseEnter={(e) => { (e.currentTarget as HTMLTableRowElement).style.backgroundColor = 'var(--color-bg-surface-hover)' }}
+                  onMouseLeave={(e) => { (e.currentTarget as HTMLTableRowElement).style.backgroundColor = '' }}
                 >
                   {/* Row # */}
-                  <td className="sticky left-0 z-10 bg-surface group-hover:bg-surface-hover/30 py-1.5 px-3 font-mono text-content-muted border-r border-b border-border-light">
+                  <td className="sticky left-0 z-10 bg-surface py-1.5 px-3 font-mono text-content-muted border-r border-b border-border-light">
                     {item.row_index}
                   </td>
 

frontend/src/pages/ProductLibrary.tsx

@@ -326,10 +326,19 @@ export default function ProductLibraryPage() {
           ))}
         </div>
       ) : !products?.length ? (
-        <div className="text-center py-16 text-content-muted">
-          <Library size={48} className="mx-auto mb-3 opacity-30" />
-          <p>No products found</p>
-          <p className="text-sm mt-1">Upload an Excel file to populate the library</p>
+        <div className="text-center py-16" style={{ color: 'var(--color-content-muted)' }}>
+          <Library size={48} className="mx-auto mb-3" style={{ opacity: 0.3 }} />
+          {debouncedSearch || categoryFilter || hasCadFilter || materialsFilter ? (
+            <>
+              <p className="font-medium">No products match your filters.</p>
+              <p className="text-sm mt-1">Try clearing the search or adjusting the filter criteria.</p>
+            </>
+          ) : (
+            <>
+              <p className="font-medium">No products yet</p>
+              <p className="text-sm mt-1">Upload an Excel file to populate the library.</p>
+            </>
+          )}
         </div>
       ) : view === 'grid' ? (
         /* ── Grid view ─────────────────────────────────────────────────── */

render-worker/Dockerfile

@@ -84,4 +84,4 @@ COPY backend/ .
 
 # Verify Blender version at build time if binary is available
 # (skipped during build since /opt/blender is a host mount)
-CMD ["bash", "-c", "python3 /check_version.py && celery -A app.tasks.celery_app worker --loglevel=info -Q thumbnail_rendering --concurrency=1"]
+CMD ["bash", "-c", "python3 /check_version.py && celery -A app.tasks.celery_app worker --loglevel=info -Q asset_pipeline --concurrency=1"]

render-worker/scripts/export_step_to_gltf.py

@@ -488,13 +488,19 @@ def _collect_part_key_map(shape_tool, free_labels) -> dict:
     return part_key_map
 
 
-def _inject_glb_extras(glb_path: Path, extras: dict) -> None:
+def _inject_glb_extras(glb_path: Path, extras: dict, part_key_map: dict | None = None) -> None:
     """Patch a GLB binary to add/update scenes[0].extras JSON field.
 
+    Also stamps per-node extras.partKey on each GLB node whose name maps to an
+    entry in part_key_map (the dict returned by _collect_part_key_map).  Three.js
+    GLTFLoader propagates node extras → object.userData, so every THREE.Mesh will
+    carry userData.partKey after load — no runtime lookup needed in the viewer.
+
     The GLB format stores a JSON chunk immediately after the 12-byte header.
     We re-serialize it with the new extras and update chunk + total lengths.
     No external dependencies — pure stdlib struct/json.
     """
+    import re as _re
     import struct as _struct
 
     data = glb_path.read_bytes()
@@ -515,6 +521,35 @@ def _inject_glb_extras(glb_path: Path, extras: dict) -> None:
     else:
         j.setdefault("extras", {}).update(extras)
 
+    # Stamp per-node extras.partKey so Three.js maps it to mesh.userData.partKey.
+    # part_key_map keys are raw OCC names with _AF\d+ stripped but not slugified
+    # (e.g. "GE360-HF_000_P_ASM_1" → "ge360_hf_000_p_asm_1").
+    # GLB node names are raw OCC names (may or may not have _AF\d+ suffix).
+    # Normalize both sides to slugified form for the lookup.
+    if part_key_map:
+        _norm_re = _re.compile(r'_AF\d+$', _re.IGNORECASE)
+
+        def _slugify(s: str) -> str:
+            return _re.sub(r'[^a-z0-9]+', '_', _norm_re.sub('', s).lower()).strip('_')
+
+        # Build a slug→partKey lookup from the part_key_map
+        # part_key_map: {raw_name_no_af_suffix: part_key_slug}
+        slug_to_part_key: dict = {}
+        for raw_key, part_key in part_key_map.items():
+            slug_to_part_key[_slugify(raw_key)] = part_key
+
+        n_stamped = 0
+        for node in j.get("nodes", []):
+            raw = node.get("name", "")
+            if not raw:
+                continue
+            slug = _slugify(raw)
+            part_key = slug_to_part_key.get(slug)
+            if part_key:
+                node.setdefault("extras", {})["partKey"] = part_key
+                n_stamped += 1
+        print(f"Stamped partKey extras on {n_stamped} GLB nodes")
+
     new_json = json.dumps(j, separators=(",", ":"))
     # Pad to 4-byte boundary with spaces (required by GLB spec)
     pad = (4 - len(new_json) % 4) % 4
@@ -720,7 +755,7 @@ def main() -> None:
         if part_key_map:
             extras_payload["partKeyMap"] = part_key_map
         if extras_payload:
-            _inject_glb_extras(out, extras_payload)
+            _inject_glb_extras(out, extras_payload, part_key_map=part_key_map if part_key_map else None)
             if sharp_pairs:
                 print(f"Injected {len(sharp_pairs)} sharp edge segment pairs into GLB extras")
             if part_key_map:

render-worker/scripts/export_step_to_usd.py

@@ -616,7 +616,10 @@ def main() -> None:
                 continue
 
             part_path = f"{node_path}/{part_key}"
-            mesh_path = f"{part_path}/Mesh"
+            # Name the Mesh prim after part_key so Blender imports it with the
+            # part name directly (Blender collapses single-child Xform+Mesh into
+            # just the Mesh object, using the mesh prim's leaf name as object name).
+            mesh_path = f"{part_path}/{part_key}"
 
             # ── Xform prim ────────────────────────────────────────────────
             xform = UsdGeom.Xform.Define(stage, part_path)

scripts/test_render_pipeline.py

@@ -125,7 +125,7 @@ def test_health(client: APIClient) -> bool:
     info(f"Overall status: {data['status']}")
     info(f"Render worker connected: {data['render_worker_connected']}")
     info(f"Blender available: {data['blender_available']}")
-    info(f"thumbnail_rendering queue depth: {data['thumbnail_queue_depth']}")
+    info(f"asset_pipeline queue depth: {data['thumbnail_queue_depth']}")
     if data.get("last_render_at"):
         info(f"Last render: {data['last_render_at']} ({'success' if data['last_render_success'] else 'FAILED'}, {data['last_render_age_minutes']}m ago)")
 
@@ -140,9 +140,9 @@ def test_health(client: APIClient) -> bool:
         fail("Blender renderer NOT reachable — thumbnail/order renders will fail")
 
     if data["thumbnail_queue_ok"]:
-        ok(f"thumbnail_rendering queue healthy (depth={data['thumbnail_queue_depth']})")
+        ok(f"asset_pipeline queue healthy (depth={data['thumbnail_queue_depth']})")
     else:
-        warn(f"thumbnail_rendering queue DEEP ({data['thumbnail_queue_depth']} tasks) — renders may be slow")
+        warn(f"asset_pipeline queue DEEP ({data['thumbnail_queue_depth']} tasks) — renders may be slow")
 
     return data["status"] != "down"