@forgeax/engine-render 0.1.21 → 0.1.24

package/src/gpu-driven/production-raster.ts

@@ -1,4 +1,5 @@
-import { frustum, mat4 } from '@forgeax/engine-math';
+import { deriveVertexBufferLayoutFromProjection } from '@forgeax/engine-geometry';
+import { frustum, mat3, mat4 } from '@forgeax/engine-math';
 import type {
   GraphAccess,
   GraphBuffer,
@@ -9,6 +10,7 @@ import type {
 import type {
   BindGroup,
   BindGroupLayout,
+  Buffer,
   RenderPipeline,
   Result,
   RhiDevice,
@@ -16,29 +18,166 @@ import type {
   TextureFormat,
 } from '@forgeax/engine-rhi';
 import { err, ok, RhiError } from '@forgeax/engine-rhi';
+import type { MaterialShaderArtifact } from '@forgeax/engine-shader';
 import type { MaterialRenderState } from '@forgeax/engine-types';
+import type { DeviceScope, LifecycleResourceSpec } from '../device/device-scope';
 import type { MeshGpuHandles } from '../device/gpu-residency';
 import { GPU_SCENE_WGSL } from '../gpu-scene-schema';
 import {
   GPU_BUFFER_USAGE_COPY_DST,
   GPU_BUFFER_USAGE_INDEX,
+  GPU_BUFFER_USAGE_STORAGE,
+  GPU_BUFFER_USAGE_UNIFORM,
   GPU_BUFFER_USAGE_VERTEX,
 } from '../gpu-usage';
+import { assembleMaterialWithSkylightEntries } from '../ibl/skylight-bind-group';
 import type { GpuDrivenProductionInspection } from '../inspection-types';
 import type { PipelineBuilderShaderModuleFactory } from '../pipeline-builder';
 import { worldEntityKey } from '../record/frame-snapshot';
 import { geometryRenderStateForTopology } from '../record/main-pass-material';
+import { MESH_SSBO_BYTES } from '../record/mesh-ssbo';
+import type { PipelineState } from '../record/render-context';
 import type { CameraSnapshot } from '../render-contract';
 import type { RenderPipelineFrame, RenderPipelineGpuDrivenProjection } from '../render-pipeline';
 import type { RenderableSnapshot } from '../render-system-extract';
 import type { PersistentGpuDrivenState } from '../scene/render-scene';
-import type { GpuDrivenBatch, SubmissionPlan } from './batch-topology';
-import { GpuDrivenView } from './view-gpu';
+import { selectLod } from '../scene/visibility/lod-selector';
+import type { OcclusionFrameProjection } from '../scene/visibility/occlusion-runtime';
+import {
+  type GpuDrivenBatch,
+  type GpuDrivenCandidate,
+  projectedHeightForCandidate,
+  type SubmissionPlan,
+} from './batch-topology';
+import { type GpuDrivenLodSelectionInspection, GpuDrivenView, selectGpuLodLane } from './view-gpu';
 
 const VERTEX_STAGE = 0x1;
 const FRAGMENT_STAGE = 0x2;
 const EMPTY_RESOURCE_CLASS = JSON.stringify({ textures: [], samplers: [], video: [] });
 
+export interface StandardPbrSceneCandidate {
+  readonly primitiveIndex: number;
+  readonly world: Float32Array;
+}
+
+export interface StandardPbrSceneProjection {
+  readonly entryPoint: string;
+  readonly layoutIdentity: string;
+  readonly receiptGeneration: number;
+  /** Packed Mesh rows: mat4 + padded mat3 + previous mat4 + temporal vec4. */
+  readonly rows: Float32Array;
+}
+
+/**
+ * Build the producer-owned Standard PBR scene-index projection. The receipt is
+ * the only source of the vertex entry and layout identity; callers cannot
+ * silently fall back to the legacy GPU-driven shader ABI.
+ */
+export function projectStandardPbrScene(input: {
+  readonly artifact: MaterialShaderArtifact;
+  readonly candidates: readonly StandardPbrSceneCandidate[];
+}):
+  | { readonly ok: true; readonly value: StandardPbrSceneProjection }
+  | {
+      readonly ok: false;
+      readonly error: { readonly code: 'missing-material-receipt' | 'invalid-material-receipt' };
+    } {
+  const receipt = input.artifact.receipt;
+  if (receipt === undefined) return { ok: false, error: { code: 'missing-material-receipt' } };
+  if (receipt.reflection.layoutIdentity !== input.artifact.layoutIdentity) {
+    return { ok: false, error: { code: 'invalid-material-receipt' } };
+  }
+  const rows = new Float32Array(input.candidates.length * 48);
+  const normal = mat3.create();
+  for (let index = 0; index < input.candidates.length; index += 1) {
+    const candidate = input.candidates[index];
+    if (candidate === undefined) continue;
+    const base = index * 48;
+    for (let value = 0; value < 16; value += 1) {
+      rows[base + value] = candidate.world[value] ?? 0;
+      rows[base + 28 + value] = candidate.world[value] ?? 0;
+    }
+    const normalMatrix = mat3.normalMatrix(normal, candidate.world);
+    for (let column = 0; column < 3; column += 1) {
+      const source = column * 3;
+      const target = base + 16 + column * 4;
+      rows[target] = normalMatrix[source] ?? 0;
+      rows[target + 1] = normalMatrix[source + 1] ?? 0;
+      rows[target + 2] = normalMatrix[source + 2] ?? 0;
+    }
+  }
+  return {
+    ok: true,
+    value: {
+      entryPoint: receipt.sceneIndexEntry,
+      layoutIdentity: input.artifact.layoutIdentity,
+      receiptGeneration: receipt.generation,
+      rows,
+    },
+  };
+}
+
+export interface StandardPbrFrameResourceBinding {
+  readonly entryPoint: string;
+  readonly layoutIdentity: string;
+  readonly receiptGeneration: number;
+  readonly bindGroups: readonly [BindGroup, BindGroup, BindGroup, BindGroup];
+  readonly resourceSlots: NonNullable<MaterialShaderArtifact['receipt']>['resourceSlots'];
+}
+
+/**
+ * Adapt the frame's already-created Standard PBR groups to the GPU-driven
+ * scene-index draw. This deliberately accepts group handles rather than
+ * rebuilding layouts or shader artifacts; the receipt remains the binding
+ * contract and the regular frame producers remain the resource owners.
+ */
+export function adaptStandardPbrFrameResources(input: {
+  readonly artifact: MaterialShaderArtifact;
+  readonly view: BindGroup;
+  readonly material: BindGroup;
+  readonly mesh: BindGroup;
+  readonly instances: BindGroup;
+}):
+  | { readonly ok: true; readonly value: StandardPbrFrameResourceBinding }
+  | {
+      readonly ok: false;
+      readonly error: { readonly code: 'missing-material-receipt' | 'invalid-material-receipt' };
+    } {
+  const receipt = input.artifact.receipt;
+  if (receipt === undefined) return { ok: false, error: { code: 'missing-material-receipt' } };
+  const reflected = receipt.reflection;
+  const sameSlots =
+    reflected.resourceSlots.length === receipt.resourceSlots.length &&
+    reflected.resourceSlots.every((slot, index) => {
+      const expected = receipt.resourceSlots[index];
+      return (
+        expected !== undefined &&
+        slot.name === expected.name &&
+        slot.parameter === expected.parameter &&
+        slot.kind === expected.kind &&
+        slot.group === expected.group &&
+        slot.binding === expected.binding
+      );
+    });
+  if (
+    reflected.layoutIdentity !== input.artifact.layoutIdentity ||
+    !sameSlots ||
+    receipt.resourceSlots.some((slot) => slot.group !== 1)
+  ) {
+    return { ok: false, error: { code: 'invalid-material-receipt' } };
+  }
+  return {
+    ok: true,
+    value: {
+      entryPoint: receipt.sceneIndexEntry,
+      layoutIdentity: input.artifact.layoutIdentity,
+      receiptGeneration: receipt.generation,
+      bindGroups: [input.view, input.material, input.mesh, input.instances],
+      resourceSlots: receipt.resourceSlots,
+    },
+  };
+}
+
 export const GPU_DRIVEN_RIGID_UNLIT_WGSL = /* wgsl */ `
 ${GPU_SCENE_WGSL}
 
@@ -87,6 +226,11 @@ interface PreparedBatch {
   readonly batch: GpuDrivenBatch;
   readonly mesh: MeshGpuHandles;
   readonly renderState: MaterialRenderState | undefined;
+  /** Stable World attachment key used to join same-submit GPU counters. */
+  readonly worldKey: number;
+  /** Stable representative slot identity for this homogeneous batch. */
+  readonly primitiveSlot: number;
+  readonly slotGeneration: number;
 }
 
 interface ProjectedBatch extends PreparedBatch {
@@ -118,8 +262,8 @@ function activeFrustum(camera: CameraSnapshot): Float32Array {
       projection,
       camera.orthoLeft,
       camera.orthoRight,
-      camera.orthoBottom,
       camera.orthoTop,
+      camera.orthoBottom,
       camera.near,
       camera.far,
     );
@@ -133,55 +277,172 @@ function activeFrustum(camera: CameraSnapshot): Float32Array {
   );
 }
 
+/**
+ * LOD projected height is a view fact, not a scene-topology fact. Keep the
+ * filtered plan cache scene-stable while still rebuilding its candidate
+ * heights whenever a camera input used by the selector changes.
+ */
+function lodProjectionFingerprint(camera: CameraSnapshot): string {
+  return JSON.stringify({
+    position: Array.from(camera.position),
+    world: Array.from(camera.world),
+    projection: camera.projection,
+    fov: camera.fov,
+    aspect: camera.aspect,
+    orthoHeight: camera.orthoTop - camera.orthoBottom,
+  });
+}
+
 function productionEligible(snapshot: RenderableSnapshot, batch: GpuDrivenBatch): boolean {
   const material = snapshot.materials[batch.key.materialSlot] ?? snapshot.material;
   const shader = material.materialShaderId ?? 'forgeax::default-unlit';
   return (
     shader === 'forgeax::default-unlit' &&
+    // The production rigid shader owns triangle-list raster only. Lines and
+    // points stay with their CPU/specialized topology owners.
+    batch.key.topology === 'triangle-list' &&
     batch.key.materialResourceClass === EMPTY_RESOURCE_CLASS &&
     material.baseColorTexture === undefined &&
     material.transparent !== true
   );
 }
 
+/**
+ * Prepared Standard PBR draws have an explicit receipt identity on their
+ * batch key. Unprepared batches stay on the rigid production shader; the
+ * presence of a frame-wide PBR artifact is not permission to reinterpret an
+ * unlit batch with the PBR bind-group layout.
+ */
+function isStandardPbrBatch(batch: GpuDrivenBatch): boolean {
+  return batch.key.preparedIdentity !== undefined;
+}
+
 function meshSupportsBatch(mesh: MeshGpuHandles, batch: GpuDrivenBatch): boolean {
-  if (mesh.layoutProjection.arrayStride !== 48) return false;
+  if (
+    batch.candidates.some((candidate) => {
+      const lodCount = candidate.lodCoverages?.length ?? 0;
+      return (
+        lodCount > 1 &&
+        (mesh.lodRanges === undefined ||
+          mesh.lodRanges.length < lodCount ||
+          mesh.lodRanges.some((ranges) => ranges[candidate.drawItemIndex] === undefined))
+      );
+    })
+  ) {
+    return false;
+  }
   return batch.key.drawKind === 'indexed'
     ? mesh.indexed && mesh.indexBuffer !== null
     : !mesh.indexed;
 }
 
+function selectedLodLevel(candidate: GpuDrivenCandidate): number {
+  const coverages = candidate.lodCoverages;
+  if (coverages === undefined || coverages.length <= 1) return 0;
+  return selectLod({
+    levels: coverages.slice(1).map((screenCoverage) => ({ screenCoverage })),
+    projectedHeight: candidate.projectedHeight ?? Number.NaN,
+    previousLevel: 0,
+    hysteresis: candidate.lodHysteresis ?? 0.08,
+    ready: coverages.map(() => true),
+    historyValid: false,
+  }).level;
+}
+
+function selectedLodRange(
+  batch: GpuDrivenBatch,
+  candidate: GpuDrivenCandidate,
+): { readonly first: number; readonly count: number; readonly baseVertex: number } {
+  const level = selectedLodLevel(candidate);
+  return (
+    (level === 0 ? undefined : candidate.lodRanges?.[level - 1]) ?? {
+      first: batch.key.first,
+      count: batch.key.count,
+      baseVertex: batch.key.baseVertex,
+    }
+  );
+}
+
+/**
+ * One raster indirect command is emitted for each selector batch that has at
+ * least one admitted candidate. Selector-only batches remain in the compute
+ * and readback plan, but they must not inflate the raster command count.
+ */
+function admittedRasterBatchCount(plan: SubmissionPlan): number {
+  return plan.batches.reduce((count, batch) => count + (batch.visibleCapacity > 0 ? 1 : 0), 0);
+}
+
 function filteredPlan(
   source: SubmissionPlan,
   slots: PersistentGpuDrivenState['slots'],
   meshes: ReadonlyMap<number, MeshGpuHandles>,
+  camera: CameraSnapshot,
+  activeEntityKeys?: ReadonlySet<number>,
+  worldKeys?: readonly number[],
 ): {
   readonly plan: SubmissionPlan;
   readonly batches: readonly PreparedBatch[];
   readonly entityKeys: ReadonlySet<number>;
+  /** No source primitive can enter this lane without a topology/material change. */
+  readonly topologyInvariantEmpty: boolean;
 } {
   const slotByPrimitive = new Map(slots.map((slot) => [slot.slot, slot] as const));
+  const participatingByPrimitive = new Map<number, Set<GpuDrivenBatch>>();
+  for (const batch of source.batches) {
+    for (const candidate of batch.candidates) {
+      const participating = participatingByPrimitive.get(candidate.primitiveIndex);
+      if (participating === undefined) {
+        participatingByPrimitive.set(candidate.primitiveIndex, new Set([batch]));
+      } else {
+        participating.add(batch);
+      }
+    }
+  }
   const entityKeys = new Set<number>();
+  const worldKeyFor = (slot: PersistentGpuDrivenState['slots'][number]): number =>
+    worldKeys?.[slot.snapshot.worldId] ?? slot.snapshot.worldId;
+  const activeKey = (slot: PersistentGpuDrivenState['slots'][number]): number =>
+    worldEntityKey(worldKeyFor(slot), slot.snapshot.entityKey);
   const eligiblePrimitives = new Set<number>();
+  let topologyEligiblePrimitiveCount = 0;
   for (const slot of slots) {
     const mesh = meshes.get(slot.snapshot.assetHandle);
-    const participating = source.batches.filter((batch) =>
-      batch.candidates.some((candidate) => candidate.primitiveIndex === slot.slot),
-    );
+    const participating = participatingByPrimitive.get(slot.slot);
+    const participatingBatches = participating === undefined ? [] : [...participating];
+    const topologyEligible =
+      participatingBatches.length > 0 &&
+      participatingBatches.length === (slot.snapshot.gpuDrivenDraws?.length ?? 0) &&
+      participatingBatches.every((batch) => productionEligible(slot.snapshot, batch));
+    if (topologyEligible) topologyEligiblePrimitiveCount += 1;
     if (
       mesh !== undefined &&
-      participating.length === (slot.snapshot.gpuDrivenDraws?.length ?? 0) &&
-      participating.every(
-        (batch) => productionEligible(slot.snapshot, batch) && meshSupportsBatch(mesh, batch),
-      )
+      topologyEligible &&
+      participatingBatches.every((batch) => meshSupportsBatch(mesh, batch))
     ) {
       eligiblePrimitives.add(slot.slot);
-      entityKeys.add(worldEntityKey(slot.snapshot.worldId, slot.snapshot.entityKey));
+      // Record validation still keys extracted renderables by their current
+      // stable world/entity key. Suppressed candidates remain in the GPU
+      // selector, but only admitted entities are owned by the compacted draw.
+      if (activeEntityKeys === undefined || activeEntityKeys.has(activeKey(slot))) {
+        entityKeys.add(activeKey(slot));
+      }
     }
   }
-  const batches: PreparedBatch[] = [];
-  let visibleBase = 0;
-  const filteredBatches: GpuDrivenBatch[] = [];
+  // BatchTopology deliberately keeps LOD candidates individually addressable
+  // so topology updates never alias a candidate's selector state.  Once the
+  // camera has projected the candidates, the renderer can safely coalesce
+  // candidates that have identical immutable geometry/LOD facts and the same
+  // selected range.  The GPU selector uses the same no-history inputs below,
+  // so one indirect range remains correct for every candidate in a group.
+  const groups = new Map<
+    string,
+    {
+      readonly source: GpuDrivenBatch;
+      readonly mesh: MeshGpuHandles;
+      readonly renderState: MaterialRenderState | undefined;
+      readonly candidates: GpuDrivenCandidate[];
+    }
+  >();
   for (const batch of source.batches) {
     const candidates = batch.candidates.filter((candidate) => {
       const slot = slotByPrimitive.get(candidate.primitiveIndex);
@@ -190,24 +451,84 @@ function filteredPlan(
       return mesh !== undefined && meshSupportsBatch(mesh, batch);
     });
     if (candidates.length === 0) continue;
+    for (const candidate of candidates) {
+      const slot = slotByPrimitive.get(candidate.primitiveIndex);
+      if (slot === undefined) continue;
+      const mesh = meshes.get(slot.snapshot.assetHandle);
+      if (mesh === undefined) continue;
+      const physicalLodRanges = mesh.lodRanges
+        ?.slice(1)
+        .map((ranges) => ranges[candidate.drawItemIndex]);
+      const projected = Object.freeze({
+        ...candidate,
+        submitAdmission: activeEntityKeys === undefined || activeEntityKeys.has(activeKey(slot)),
+        projectedHeight: projectedHeightForCandidate(slot, camera),
+        ...(physicalLodRanges?.every((range) => range !== undefined)
+          ? { lodRanges: physicalLodRanges as NonNullable<typeof candidate.lodRanges> }
+          : {}),
+      });
+      const selectedRange = selectedLodRange(batch, projected);
+      const groupKey = JSON.stringify({
+        key: batch.key,
+        worldKey: worldKeyFor(slot),
+        lodCoverages: projected.lodCoverages ?? null,
+        lodRanges: projected.lodRanges ?? null,
+        selectedRange,
+        selectedLevel: selectedLodLevel(projected),
+      });
+      const existing = groups.get(groupKey);
+      if (existing === undefined) {
+        const material = slot.snapshot.materials[batch.key.materialSlot] ?? slot.snapshot.material;
+        groups.set(groupKey, {
+          source: batch,
+          mesh,
+          renderState: geometryRenderStateForTopology(batch.key.topology, material.renderState),
+          candidates: [projected],
+        });
+      } else {
+        existing.candidates.push(projected);
+      }
+    }
+  }
+
+  const batches: PreparedBatch[] = [];
+  let visibleBase = 0;
+  const filteredBatches: GpuDrivenBatch[] = [];
+  let batchId = 0;
+  for (const group of groups.values()) {
     visibleBase = Math.ceil(visibleBase / 64) * 64;
+    const admittedCount = group.candidates.reduce(
+      (count, candidate) =>
+        count +
+        ((candidate as GpuDrivenCandidate & { readonly submitAdmission?: boolean })
+          .submitAdmission === false
+          ? 0
+          : 1),
+      0,
+    );
     const filtered: GpuDrivenBatch = Object.freeze({
-      ...batch,
-      candidates: Object.freeze(candidates),
+      ...group.source,
+      batchId,
+      candidates: Object.freeze(group.candidates),
       visibleBase,
-      visibleCapacity: candidates.length,
+      visibleCapacity: admittedCount,
+      indirectOffset: batchId * 20,
     });
-    visibleBase += candidates.length;
-    const first = slotByPrimitive.get(candidates[0]?.primitiveIndex ?? -1);
-    const mesh = first === undefined ? undefined : meshes.get(first.snapshot.assetHandle);
-    if (mesh === undefined || first === undefined) continue;
+    visibleBase += admittedCount;
     filteredBatches.push(filtered);
-    const material = first.snapshot.materials[batch.key.materialSlot] ?? first.snapshot.material;
+    const firstCandidate = group.candidates[0];
+    const firstSlot =
+      firstCandidate === undefined ? undefined : slotByPrimitive.get(firstCandidate.primitiveIndex);
+    if (firstCandidate === undefined || firstSlot === undefined) continue;
     batches.push({
       batch: filtered,
-      mesh,
-      renderState: geometryRenderStateForTopology(batch.key.topology, material.renderState),
+      mesh: group.mesh,
+      renderState: group.renderState,
+      worldKey: worldKeyFor(firstSlot),
+      primitiveSlot: firstSlot.slot,
+      slotGeneration: firstSlot.generation,
     });
+    batchId += 1;
   }
   return {
     plan: Object.freeze({
@@ -218,11 +539,32 @@ function filteredPlan(
     }),
     batches: Object.freeze(batches),
     entityKeys,
+    topologyInvariantEmpty: topologyEligiblePrimitiveCount === 0,
   };
 }
 
 type FilteredProductionPlan = ReturnType<typeof filteredPlan>;
 
+export interface GpuDrivenWorldLodSelectionInspection {
+  readonly worldKey: number;
+  readonly primitiveSlot: number;
+  readonly slotGeneration: number;
+  readonly candidateCount: number;
+  readonly visible: number;
+  readonly occluded: number;
+  readonly lodHistogram: readonly { readonly level: number; readonly count: number }[];
+}
+
+function sameEntityKeySet(
+  left: ReadonlySet<number> | undefined,
+  right: ReadonlySet<number> | undefined,
+): boolean {
+  if (left === right) return true;
+  if (left === undefined || right === undefined || left.size !== right.size) return false;
+  for (const key of left) if (!right.has(key)) return false;
+  return true;
+}
+
 class GpuDrivenRigidRaster {
   private readonly pipelines = new Map<string, RenderPipeline>();
 
@@ -375,10 +717,212 @@ class GpuDrivenRigidRaster {
   }
 }
 
+/** The GPU-driven bridge to the existing four-group Standard PBR pipeline. */
+class StandardPbrRasterAdapter {
+  private readonly pipelines = new Map<string, RenderPipeline>();
+  private readonly projectionBuffers: Buffer[] = [];
+
+  private constructor(
+    private readonly device: RhiDevice,
+    private readonly pipelineLayout: import('@forgeax/engine-rhi').PipelineLayout,
+    private readonly meshLayout: BindGroupLayout,
+    private readonly module: ShaderModule,
+    private readonly materialBindGroup: BindGroup,
+    private readonly instancesBindGroup: BindGroup,
+    private readonly artifact: MaterialShaderArtifact,
+  ) {}
+
+  static create(input: {
+    readonly device: RhiDevice;
+    readonly shaderModuleFactory: PipelineBuilderShaderModuleFactory;
+    readonly pipelineState: PipelineState;
+    readonly artifact: MaterialShaderArtifact;
+  }): Result<StandardPbrRasterAdapter, RhiError> {
+    const pipelineLayout = input.pipelineState.pbrPipelineLayout;
+    const fallback = input.pipelineState.skylightFallback;
+    if (pipelineLayout === null || fallback === null) {
+      return err(
+        new RhiError({
+          code: 'rhi-not-available',
+          expected: 'Standard PBR pipeline layout and Skylight fallback',
+          hint: 'initialize the Standard PBR pipeline before enabling GPU-driven raster',
+        }),
+      );
+    }
+    const module = input.shaderModuleFactory.createShaderModule({
+      label: 'standard-pbr-gpu-driven-scene-index',
+      code: input.artifact.wgsl,
+    });
+    if (!module.ok) return module;
+    const materialBuffer = input.device.createBuffer({
+      label: 'standard-pbr-gpu-driven-material-row',
+      size: 384,
+      usage: GPU_BUFFER_USAGE_UNIFORM | GPU_BUFFER_USAGE_COPY_DST,
+    });
+    if (!materialBuffer.ok) return materialBuffer;
+    const materialPayload = new Float32Array(96);
+    materialPayload[0] = 1;
+    materialPayload[1] = 1;
+    materialPayload[2] = 1;
+    materialPayload[3] = 1;
+    materialPayload[5] = 1;
+    materialPayload[6] = 1;
+    const materialWrite = input.device.queue.writeBuffer(materialBuffer.value, 0, materialPayload);
+    if (!materialWrite.ok) return materialWrite;
+    const baseEntries = [
+      { binding: 0, resource: bufferBinding(materialBuffer.value, 0, 384) },
+      ...Array.from({ length: 8 }, (_, index) => [
+        {
+          binding: index * 2 + 1,
+          resource: { kind: 'sampler' as const, value: fallback.sampler },
+        },
+        {
+          binding: index * 2 + 2,
+          resource: {
+            kind: 'textureView' as const,
+            value:
+              index === 2
+                ? input.pipelineState.defaultNormalTextureView
+                : input.pipelineState.defaultWhiteTextureView,
+          },
+        },
+      ]).flat(),
+    ];
+    const materialEntries = assembleMaterialWithSkylightEntries(
+      baseEntries,
+      {
+        irradianceView: fallback.irradianceView,
+        irradianceSampler: fallback.sampler,
+        prefilterView: fallback.prefilterView,
+        prefilterSampler: fallback.sampler,
+        brdfLutView: fallback.brdfLutView,
+        brdfLutSampler: fallback.sampler,
+        intensityBuffer: fallback.intensityBuffer,
+      },
+      { sampler: fallback.sampler, backdropView: input.pipelineState.defaultWhiteTextureView },
+    );
+    const materialBindGroup = input.device.createBindGroup({
+      label: 'standard-pbr-gpu-driven-material-bg',
+      layout: input.pipelineState.materialBindGroupLayout,
+      entries: materialEntries,
+    });
+    if (!materialBindGroup.ok) return materialBindGroup;
+    const instancesBindGroup = input.device.createBindGroup({
+      label: 'standard-pbr-gpu-driven-identity-instances-bg',
+      layout: input.pipelineState.instancesBindGroupLayout,
+      entries: [
+        {
+          binding: 0,
+          resource: bufferBinding(input.pipelineState.identityInstanceBuffer),
+        },
+      ],
+    });
+    if (!instancesBindGroup.ok) return instancesBindGroup;
+    return ok(
+      new StandardPbrRasterAdapter(
+        input.device,
+        pipelineLayout,
+        input.pipelineState.meshBindGroupLayout,
+        module.value,
+        materialBindGroup.value,
+        instancesBindGroup.value,
+        input.artifact,
+      ),
+    );
+  }
+
+  pipeline(
+    format: TextureFormat,
+    sampleCount: 1 | 4,
+    layoutProjection: MeshGpuHandles['layoutProjection'],
+    topology: GpuDrivenBatch['key']['topology'],
+    stripIndexFormat: 'uint16' | 'uint32' | undefined,
+    renderState: MaterialRenderState | undefined,
+  ): Result<RenderPipeline, RhiError> {
+    const projectedLayout = deriveVertexBufferLayoutFromProjection(
+      layoutProjection,
+      this.artifact.uvSetCount === undefined
+        ? undefined
+        : { shaderUvSetCount: this.artifact.uvSetCount },
+    )[0];
+    const vertexStride = projectedLayout?.arrayStride ?? layoutProjection.arrayStride;
+    const attributes = projectedLayout?.attributes ?? layoutProjection.attributes;
+    const key = `${format}|${sampleCount}|${vertexStride}|${topology}|${stripIndexFormat ?? ''}|${JSON.stringify(renderState ?? null)}|${this.artifact.layoutIdentity}`;
+    const cached = this.pipelines.get(key);
+    if (cached !== undefined) return ok(cached);
+    const pipeline = this.device.createRenderPipeline({
+      label: `standard-pbr-gpu-driven.${key}`,
+      layout: this.pipelineLayout,
+      vertex: {
+        module: this.module,
+        entryPoint: this.artifact.receipt?.sceneIndexEntry ?? 'vs_scene_index',
+        buffers: [
+          {
+            arrayStride: vertexStride,
+            stepMode: 'vertex',
+            attributes: attributes.map(({ shaderLocation, offset, format }) => ({
+              shaderLocation,
+              offset,
+              format: format as GPUVertexFormat,
+            })),
+          },
+        ],
+      },
+      fragment: {
+        module: this.module,
+        entryPoint: 'fs_main',
+        targets: [{ format }],
+      },
+      primitive: {
+        topology,
+        cullMode: renderState?.cullMode ?? 'back',
+        frontFace: renderState?.frontFace ?? 'ccw',
+        ...(stripIndexFormat === undefined ? {} : { stripIndexFormat }),
+      },
+      depthStencil: {
+        format: 'depth24plus-stencil8',
+        depthWriteEnabled: renderState?.depthWriteEnabled ?? true,
+        depthCompare: renderState?.depthCompare ?? 'less',
+      },
+      ...(sampleCount === 4 ? { multisample: { count: 4 } } : {}),
+    });
+    if (!pipeline.ok) return pipeline;
+    this.pipelines.set(key, pipeline.value);
+    return pipeline;
+  }
+
+  projectionBindGroup(rows: Float32Array): Result<BindGroup, RhiError> {
+    const buffer = this.device.createBuffer({
+      label: 'standard-pbr-gpu-driven-scene-meshes',
+      size: Math.max(MESH_SSBO_BYTES, rows.byteLength),
+      usage: GPU_BUFFER_USAGE_STORAGE | GPU_BUFFER_USAGE_COPY_DST,
+    });
+    if (!buffer.ok) return buffer;
+    const write = this.device.queue.writeBuffer(buffer.value, 0, rows);
+    if (!write.ok) return write;
+    this.projectionBuffers.push(buffer.value);
+    return this.device.createBindGroup({
+      label: 'standard-pbr-gpu-driven-scene-mesh-bg',
+      layout: this.meshLayout,
+      entries: [{ binding: 0, resource: bufferBinding(buffer.value, 0, rows.byteLength) }],
+    });
+  }
+
+  dispose(): void {
+    for (const buffer of this.projectionBuffers) this.device.destroyBuffer(buffer);
+    this.projectionBuffers.length = 0;
+  }
+
+  get frameGroups(): readonly [BindGroup, BindGroup] {
+    return [this.materialBindGroup, this.instancesBindGroup];
+  }
+}
+
 export interface PreparedGpuDrivenFrame {
   readonly topologySignature: string;
   readonly entityKeys: ReadonlySet<number>;
   readonly ownsAllRenderables: boolean;
+  readonly occlusion?: OcclusionFrameProjection;
   /** @internal Commits a replacement generation only after graph promotion. */
   readonly _commitResourceReplacement: () => void;
   project(
@@ -388,15 +932,74 @@ export interface PreparedGpuDrivenFrame {
   ): Result<RenderPipelineGpuDrivenProjection, RenderGraphError | RhiError>;
 }
 
+export interface RecoveryCapabilityFacts {
+  readonly generation: number;
+  readonly shaderCompilation: boolean;
+  readonly compute: boolean;
+  readonly storageBuffer: boolean;
+  readonly indirectDrawing: boolean;
+  readonly multisample: boolean;
+}
+
+export type RecoveryCapabilityOutcome =
+  | {
+      readonly status: 'ready' | 'fallback';
+      readonly generation: number;
+      readonly disabled: readonly string[];
+    }
+  | {
+      readonly status: 'disabled';
+      readonly generation: number;
+      readonly disabled: readonly string[];
+      readonly allocations: 0;
+    }
+  | {
+      readonly status: 'refused';
+      readonly generation: number;
+      readonly failedOwner: 'shader-material-pipeline';
+      readonly resourceKind: 'pipeline';
+    };
+
+export function classifyRecoveryCapability(
+  facts: RecoveryCapabilityFacts,
+): RecoveryCapabilityOutcome {
+  if (!facts.shaderCompilation) {
+    return {
+      status: 'refused',
+      generation: facts.generation,
+      failedOwner: 'shader-material-pipeline',
+      resourceKind: 'pipeline',
+    };
+  }
+  if (!facts.compute || !facts.storageBuffer || !facts.indirectDrawing) {
+    return {
+      status: 'disabled',
+      generation: facts.generation,
+      disabled: ['gpu-driven'],
+      allocations: 0,
+    };
+  }
+  return {
+    status: facts.multisample ? 'ready' : 'fallback',
+    generation: facts.generation,
+    disabled: [],
+  };
+}
+
 export class GpuDrivenProduction {
+  private disposed = false;
   private view: GpuDrivenView | undefined;
   private raster: GpuDrivenRigidRaster | undefined;
+  private standardRaster: StandardPbrRasterAdapter | undefined;
   private readonly sceneIdentities = new WeakMap<object, number>();
   private nextSceneIdentity = 1;
   private filteredCache:
     | {
         readonly source: SubmissionPlan;
         readonly meshResidencyEpoch: number;
+        readonly lodProjectionFingerprint: string;
+        readonly activeEntityRevision?: number;
+        readonly activeEntityKeys?: ReadonlySet<number>;
         readonly value: FilteredProductionPlan;
       }
     | undefined;
@@ -406,6 +1009,13 @@ export class GpuDrivenProduction {
   private batchBindGroupCreates = 0;
   private validatedGpuOwnedRows = 0;
   private cpuFallbackDrawItems = 0;
+  private telemetryPrepared = false;
+  private telemetryCandidateCount = 0;
+  private telemetryGeometryWork = 0;
+  private telemetryRootGeometryWork = 0;
+  private indirectDrawCount = 0;
+  private sceneRowsRequired = true;
+  private standardPbrProjection: StandardPbrSceneProjection | undefined;
   private signatureCache:
     | {
         readonly filtered: FilteredProductionPlan;
@@ -421,45 +1031,171 @@ export class GpuDrivenProduction {
     private readonly shaderModuleFactory: PipelineBuilderShaderModuleFactory,
   ) {}
 
+  static forDevice(
+    device: RhiDevice,
+    shaderModuleFactory: PipelineBuilderShaderModuleFactory,
+  ): GpuDrivenProduction {
+    return new GpuDrivenProduction(device, shaderModuleFactory);
+  }
+
+  createRecoveryRoot(scope: DeviceScope): LifecycleResourceSpec<unknown> {
+    return {
+      kind: 'scene-table',
+      create: () => {
+        if (!scope.isAlive()) throw new Error('GPU-driven candidate scope is not active.');
+        // Keep the actual detached production owner in the generation
+        // aggregate. The recovery transaction, not a marker object, is the
+        // owner that is published or discarded exactly once.
+        return this;
+      },
+      cleanup: () => undefined,
+    };
+  }
+
   prepare(input: {
     readonly scene: PersistentGpuDrivenState | undefined;
     readonly camera: CameraSnapshot;
     readonly meshes: ReadonlyMap<number, MeshGpuHandles>;
     readonly viewBindGroupLayout: BindGroupLayout;
     readonly meshResidencyEpoch: number;
+    readonly hdrp: boolean;
+    /** Entity keys admitted by the persistent visibility facet for this view. */
+    readonly activeEntityKeys?: ReadonlySet<number>;
+    /** Collision-free renderer-owned revision for the active visibility set. */
+    readonly activeEntityRevision?: number;
+    /** Total extracted candidates, including facet-suppressed rows. */
+    readonly telemetryCandidateCount?: number;
+    readonly occlusion?: OcclusionFrameProjection;
+    /** Cooked producer artifact selected from the frame material catalog. */
+    readonly standardPbrArtifact?: MaterialShaderArtifact;
+    readonly standardPbrPipelineState?: PipelineState;
   }): Result<PreparedGpuDrivenFrame | undefined, RhiError> {
     this.gpuOwnedSnapshotsMaterialized = 0;
     this.filteredPlanBuilds = 0;
     this.batchBindGroupCreates = 0;
     this.validatedGpuOwnedRows = 0;
     this.cpuFallbackDrawItems = 0;
+    this.telemetryPrepared = false;
+    this.telemetryCandidateCount = input.telemetryCandidateCount ?? 0;
+    this.telemetryGeometryWork = 0;
+    this.telemetryRootGeometryWork = 0;
+    this.indirectDrawCount = 0;
+    const capability = classifyRecoveryCapability({
+      generation: 0,
+      shaderCompilation: true,
+      compute: this.device.caps.compute,
+      storageBuffer: this.device.caps.storageBuffer,
+      indirectDrawing: this.device.caps.indirectDrawing,
+      multisample: true,
+    });
     if (
+      input.hdrp ||
       input.scene === undefined ||
-      !this.device.caps.compute ||
-      !this.device.caps.storageBuffer ||
-      !this.device.caps.indirectDrawing
+      capability.status === 'disabled' ||
+      selectGpuLodLane(this.device.caps) === 'cpu'
     ) {
       this.gpuOwnedEntityCount = 0;
+      this.sceneRowsRequired = false;
       return ok(undefined);
     }
     const scene = input.scene;
+    const projectionFingerprint = lodProjectionFingerprint(input.camera);
+    // An empty filtered plan means this topology/mesh-residency pair has no
+    // renderables that this production lane can own. Camera and visibility
+    // inputs only affect admitted candidates and LOD selection, so they cannot
+    // turn that empty topology into an eligible one. Keep the negative result
+    // resident instead of rescanning every scene slot as the camera moves.
+    const cachedPlanIsTopologyInvariant = this.filteredCache?.value.topologyInvariantEmpty === true;
+    const cachedPlanIsCameraInvariant = cachedPlanIsTopologyInvariant;
+    const cachedVisibilityMatches =
+      cachedPlanIsCameraInvariant || this.filteredCache === undefined
+        ? true
+        : input.activeEntityRevision !== undefined &&
+            this.filteredCache.activeEntityRevision !== undefined
+          ? input.activeEntityRevision === this.filteredCache.activeEntityRevision
+          : sameEntityKeySet(this.filteredCache.activeEntityKeys, input.activeEntityKeys);
     let filtered = this.filteredCache?.value;
     if (
       filtered === undefined ||
       this.filteredCache?.source !== scene.plan ||
-      this.filteredCache.meshResidencyEpoch !== input.meshResidencyEpoch
+      (!cachedPlanIsTopologyInvariant &&
+        this.filteredCache.meshResidencyEpoch !== input.meshResidencyEpoch) ||
+      (!cachedPlanIsCameraInvariant &&
+        this.filteredCache.lodProjectionFingerprint !== projectionFingerprint) ||
+      !cachedVisibilityMatches
     ) {
-      filtered = filteredPlan(scene.plan, scene.slots, input.meshes);
+      filtered = filteredPlan(
+        scene.plan,
+        scene.slots,
+        input.meshes,
+        input.camera,
+        input.activeEntityKeys,
+        scene.worldKeys,
+      );
       this.filteredCache = {
         source: scene.plan,
         meshResidencyEpoch: input.meshResidencyEpoch,
+        lodProjectionFingerprint: projectionFingerprint,
+        ...(input.activeEntityRevision === undefined
+          ? {}
+          : { activeEntityRevision: input.activeEntityRevision }),
+        ...(input.activeEntityKeys === undefined
+          ? {}
+          : { activeEntityKeys: input.activeEntityKeys }),
         value: filtered,
       };
       this.gpuOwnedSnapshotsMaterialized = scene.slots.length;
       this.filteredPlanBuilds = 1;
     }
     this.gpuOwnedEntityCount = filtered.entityKeys.size;
+    this.sceneRowsRequired = !filtered.topologyInvariantEmpty;
+    // Publish the command count from the current filtered plan on every
+    // prepare call. A cached graph does not call project() again on stable
+    // frames, so deriving this only while compiling the raster pass would
+    // transiently report zero after prepare() reset the per-frame counters.
+    this.indirectDrawCount = admittedRasterBatchCount(filtered.plan);
     if (filtered.batches.length === 0) return ok(undefined);
+    this.standardPbrProjection = undefined;
+    const usesStandardPbrRaster = filtered.batches.every(({ batch }) => isStandardPbrBatch(batch));
+    if (usesStandardPbrRaster && input.standardPbrArtifact !== undefined) {
+      const slotsByPrimitive = new Map(scene.slots.map((slot) => [slot.slot, slot] as const));
+      const projection = projectStandardPbrScene({
+        artifact: input.standardPbrArtifact,
+        candidates: filtered.batches.flatMap(({ batch }) =>
+          batch.candidates.flatMap((candidate) => {
+            const slot = slotsByPrimitive.get(candidate.primitiveIndex);
+            return slot === undefined
+              ? []
+              : [
+                  {
+                    primitiveIndex: candidate.primitiveIndex,
+                    world: slot.snapshot.transform.world,
+                  },
+                ];
+          }),
+        ),
+      });
+      if (!projection.ok) {
+        return err(
+          new RhiError({
+            code: 'shader-compile-failed',
+            expected: 'receipt-derived Standard PBR scene projection',
+            hint: 'publish one validated MaterialShaderArtifact receipt before entering the GPU lane',
+          }),
+        );
+      }
+      this.standardPbrProjection = projection.value;
+      if (input.standardPbrPipelineState !== undefined && this.standardRaster === undefined) {
+        const created = StandardPbrRasterAdapter.create({
+          device: this.device,
+          shaderModuleFactory: this.shaderModuleFactory,
+          pipelineState: input.standardPbrPipelineState,
+          artifact: input.standardPbrArtifact,
+        });
+        if (!created.ok) return created;
+        this.standardRaster = created.value;
+      }
+    }
     if (this.view === undefined) {
       const created = GpuDrivenView.create({
         device: this.device,
@@ -479,8 +1215,11 @@ export class GpuDrivenProduction {
     }
     const updated = this.view.update(filtered.plan, scene.scene, activeFrustum(input.camera));
     if (!updated.ok) return updated;
+    this.telemetryPrepared = true;
     const view = this.view;
     const raster = this.raster;
+    const standardRaster = this.standardRaster;
+    const standardArtifact = input.standardPbrArtifact;
     const sceneIdentity = this.sceneIdentity(scene.scene);
     const viewInspection = view.inspect();
     let signature = this.signatureCache?.value;
@@ -495,10 +1234,12 @@ export class GpuDrivenProduction {
         sceneIdentity,
         revision: filtered.plan.revision,
         meshResidencyEpoch: input.meshResidencyEpoch,
+        standardPbrLayoutIdentity: this.standardPbrProjection?.layoutIdentity,
         viewResourceGeneration: viewInspection.resourceGeneration,
         batches: filtered.batches.map(({ batch, mesh }) => [
           batch.batchId,
           batch.generation,
+          batch.visibleCapacity,
           batch.visibleBase,
           batch.candidates.map((candidate) => [
             candidate.primitiveIndex,
@@ -512,6 +1253,7 @@ export class GpuDrivenProduction {
         ]),
         capacities: [
           viewInspection.candidateCapacity,
+          viewInspection.visibleBufferCapacity,
           viewInspection.batchCapacity,
           viewInspection.indirectCapacity,
         ],
@@ -527,11 +1269,165 @@ export class GpuDrivenProduction {
     return ok({
       topologySignature: signature,
       entityKeys: filtered.entityKeys,
-      ownsAllRenderables: filtered.entityKeys.size === scene.slots.length,
+      ownsAllRenderables:
+        input.activeEntityKeys === undefined
+          ? filtered.entityKeys.size === scene.slots.length
+          : filtered.entityKeys.size === input.activeEntityKeys.size,
+      ...(input.occlusion === undefined ? {} : { occlusion: input.occlusion }),
       _commitResourceReplacement: () => view._commitResourceReplacement(),
       project: (graph, format, sampleCount) => {
+        this.indirectDrawCount = admittedRasterBatchCount(filtered.plan);
         const outputs = view.addPasses(graph);
         if (!outputs.ok) return outputs;
+        if (
+          usesStandardPbrRaster &&
+          standardRaster !== undefined &&
+          standardArtifact !== undefined
+        ) {
+          const standardProjected: Array<{
+            readonly batch: GpuDrivenBatch;
+            readonly mesh: MeshGpuHandles;
+            readonly vertex: GraphBuffer;
+            readonly index?: GraphBuffer;
+            readonly pipeline: RenderPipeline;
+            readonly meshBindGroup: BindGroup;
+          }> = [];
+          const standardAccesses: GraphAccess[] = [
+            { resource: outputs.value.primitive, usage: 'storage-read' },
+            { resource: outputs.value.instance, usage: 'storage-read' },
+            { resource: outputs.value.material, usage: 'storage-read' },
+            { resource: outputs.value.visible, usage: 'storage-read' },
+            { resource: outputs.value.transform, usage: 'storage-read' },
+            { resource: outputs.value.indirect, usage: 'indirect-read' },
+          ];
+          const slotsByPrimitive = new Map(scene.slots.map((slot) => [slot.slot, slot] as const));
+          const meshBuffers = new Map<
+            number,
+            { readonly vertex: GraphBuffer; readonly index?: GraphBuffer }
+          >();
+          for (const prepared of filtered.batches) {
+            const candidates = prepared.batch.candidates.flatMap((candidate) => {
+              const slot = slotsByPrimitive.get(candidate.primitiveIndex);
+              return slot === undefined
+                ? []
+                : [
+                    {
+                      primitiveIndex: candidate.primitiveIndex,
+                      world: slot.snapshot.transform.world,
+                    },
+                  ];
+            });
+            const projection = projectStandardPbrScene({
+              artifact: standardArtifact,
+              candidates,
+            });
+            if (!projection.ok) {
+              return err(
+                new RhiError({
+                  code: 'shader-compile-failed',
+                  expected: 'receipt-derived Standard PBR scene projection',
+                  hint: 'keep the frame artifact receipt and projected rows on one identity',
+                }),
+              );
+            }
+            const meshBindGroup = standardRaster.projectionBindGroup(projection.value.rows);
+            if (!meshBindGroup.ok) return meshBindGroup;
+            let buffers = meshBuffers.get(prepared.batch.key.assetHandle);
+            if (buffers === undefined) {
+              const vertex = graph.importBuffer(
+                `gpu-driven.standard.${prepared.batch.key.assetHandle}.vertex`,
+                {
+                  size: prepared.mesh.vboBytes,
+                  usage: GPU_BUFFER_USAGE_VERTEX | GPU_BUFFER_USAGE_COPY_DST,
+                },
+                () => prepared.mesh.vertexBuffer.handle,
+              );
+              if (!vertex.ok) return vertex;
+              let index: GraphBuffer | undefined;
+              if (prepared.mesh.indexBuffer !== null) {
+                const indexHandle = prepared.mesh.indexBuffer;
+                const imported = graph.importBuffer(
+                  `gpu-driven.standard.${prepared.batch.key.assetHandle}.index`,
+                  {
+                    size: prepared.mesh.iboBytes,
+                    usage: GPU_BUFFER_USAGE_INDEX | GPU_BUFFER_USAGE_COPY_DST,
+                  },
+                  () => indexHandle.handle,
+                );
+                if (!imported.ok) return imported;
+                index = imported.value;
+              }
+              buffers = { vertex: vertex.value, ...(index === undefined ? {} : { index }) };
+              meshBuffers.set(prepared.batch.key.assetHandle, buffers);
+              standardAccesses.push({ resource: buffers.vertex, usage: 'vertex-read' });
+              if (buffers.index !== undefined) {
+                standardAccesses.push({ resource: buffers.index, usage: 'index-read' });
+              }
+            }
+            const pipeline = standardRaster.pipeline(
+              format,
+              sampleCount,
+              prepared.mesh.layoutProjection,
+              prepared.batch.key.topology,
+              prepared.batch.key.drawKind === 'indexed' &&
+                (prepared.batch.key.topology === 'line-strip' ||
+                  prepared.batch.key.topology === 'triangle-strip')
+                ? prepared.mesh.indexFormat
+                : undefined,
+              prepared.renderState,
+            );
+            if (!pipeline.ok) return pipeline;
+            standardProjected.push({
+              batch: prepared.batch,
+              mesh: prepared.mesh,
+              ...buffers,
+              pipeline: pipeline.value,
+              meshBindGroup: meshBindGroup.value,
+            });
+          }
+          const [materialBindGroup, instancesBindGroup] = standardRaster.frameGroups;
+          return ok({
+            accesses: standardAccesses,
+            encode: (viewBindGroup, pass, resources: GraphResourceResolver) => {
+              // pbr-view-bgl has two dynamic uniform bindings: the frame view
+              // slot and the optional Points/Lines style slot.  Standard
+              // draws do not use the latter, but WebGPU still requires one
+              // offset per dynamic binding.
+              pass.setBindGroup(0, viewBindGroup, [0, 0]);
+              pass.setBindGroup(1, materialBindGroup, [0]);
+              pass.setBindGroup(3, instancesBindGroup);
+              let currentPipeline: RenderPipeline | undefined;
+              let currentVertex: GraphBuffer | undefined;
+              let currentIndex: GraphBuffer | undefined;
+              for (const batch of standardProjected) {
+                if (currentPipeline !== batch.pipeline) {
+                  pass.setPipeline(batch.pipeline);
+                  currentPipeline = batch.pipeline;
+                }
+                pass.setBindGroup(2, batch.meshBindGroup, [0]);
+                if (currentVertex !== batch.vertex) {
+                  const vertex = resources.buffer(batch.vertex);
+                  if (!vertex.ok) throw vertex.error;
+                  pass.setVertexBuffer(0, vertex.value);
+                  currentVertex = batch.vertex;
+                }
+                if (batch.index !== undefined && currentIndex !== batch.index) {
+                  const index = resources.buffer(batch.index);
+                  if (!index.ok) throw index.error;
+                  pass.setIndexBuffer(index.value, batch.mesh.indexFormat);
+                  currentIndex = batch.index;
+                }
+                const indirect = resources.buffer(outputs.value.indirect);
+                if (!indirect.ok) throw indirect.error;
+                if (batch.batch.key.drawKind === 'indexed') {
+                  pass.drawIndexedIndirect(indirect.value, batch.batch.indirectOffset);
+                } else {
+                  pass.drawIndirect(indirect.value, batch.batch.indirectOffset);
+                }
+              }
+            },
+          });
+        }
         const projected: ProjectedBatch[] = [];
         const accesses: GraphAccess[] = [
           { resource: outputs.value.primitive, usage: 'storage-read' },
@@ -546,6 +1442,10 @@ export class GpuDrivenProduction {
           { readonly vertex: GraphBuffer; readonly index?: GraphBuffer }
         >();
         for (const prepared of filtered.batches) {
+          // A selector-only batch can have zero admitted candidates. Keep it
+          // in the GPU selector and readback counters, but do not create a
+          // raster bind group or encode a zero-instance indirect draw.
+          if (prepared.batch.visibleCapacity === 0) continue;
           let buffers = meshBuffers.get(prepared.batch.key.assetHandle);
           if (buffers === undefined) {
             const vertex = graph.importBuffer(
@@ -581,7 +1481,7 @@ export class GpuDrivenProduction {
           const pipeline = raster.pipeline(
             format,
             sampleCount,
-            48 + (prepared.mesh.uvSetCount - 1) * 8,
+            prepared.mesh.layoutProjection.arrayStride,
             prepared.batch.key.topology,
             prepared.batch.key.drawKind === 'indexed' &&
               (prepared.batch.key.topology === 'line-strip' ||
@@ -601,6 +1501,7 @@ export class GpuDrivenProduction {
             bindings: bindings.value,
           });
         }
+        this.indirectDrawCount = projected.length;
         return ok({
           accesses,
           encode: (viewBindGroup, pass, resources: GraphResourceResolver) => {
@@ -633,11 +1534,14 @@ export class GpuDrivenProduction {
               }
               const indirect = resources.buffer(outputs.value.indirect);
               if (!indirect.ok) throw indirect.error;
-              if (batch.batch.key.drawKind === 'indexed') {
-                pass.drawIndexedIndirect(indirect.value, batch.batch.indirectOffset);
-              } else {
-                pass.drawIndirect(indirect.value, batch.batch.indirectOffset);
-              }
+              const draw = (): void => {
+                if (batch.batch.key.drawKind === 'indexed') {
+                  pass.drawIndexedIndirect(indirect.value, batch.batch.indirectOffset);
+                } else {
+                  pass.drawIndirect(indirect.value, batch.batch.indirectOffset);
+                }
+              };
+              draw();
             }
           },
         });
@@ -646,11 +1550,19 @@ export class GpuDrivenProduction {
   }
 
   dispose(): void {
+    if (this.disposed) return;
+    this.disposed = true;
     this.view?.dispose();
     this.view = undefined;
     this.raster = undefined;
     this.filteredCache = undefined;
     this.signatureCache = undefined;
+    this.sceneRowsRequired = true;
+  }
+
+  /** Whether the current production topology can consume GPU Scene row updates. */
+  requiresSceneRows(): boolean {
+    return this.sceneRowsRequired;
   }
 
   inspect(): GpuDrivenProductionInspection {
@@ -667,17 +1579,112 @@ export class GpuDrivenProduction {
       topologyRevision: view?.topologyRevision,
       validatedGpuOwnedRows: this.validatedGpuOwnedRows,
       cpuFallbackDrawItems: this.cpuFallbackDrawItems,
+      geometryWork: this.telemetryGeometryWork,
+      rootGeometryWork: this.telemetryRootGeometryWork,
+      geometryWorkReduction:
+        this.telemetryRootGeometryWork > 0
+          ? 1 - this.telemetryGeometryWork / this.telemetryRootGeometryWork
+          : 0,
+      batchCount: view?.batchCount ?? 0,
+      indirectDrawCount: this.indirectDrawCount,
     };
   }
 
+  /** Read renderer-owned LOD counters copied by the last GPU submission. */
+  readLodSelection(): Promise<GpuDrivenLodSelectionInspection | undefined> {
+    if (!this.telemetryPrepared) return Promise.resolve(undefined);
+    return (this.view?.readLodSelection() ?? Promise.resolve(undefined)).then((selection) => {
+      if (selection !== undefined) {
+        this.telemetryGeometryWork = selection.geometryWork;
+        this.telemetryRootGeometryWork = selection.rootGeometryWork;
+      }
+      if (selection === undefined || this.telemetryCandidateCount <= 0) return selection;
+      const filtered = this.filteredCache?.value;
+      if (filtered === undefined) return selection;
+      const preparedByBatchId = new Map(
+        filtered.batches.map((prepared) => [prepared.batch.batchId, prepared] as const),
+      );
+      const worldSelections = new Map<
+        number,
+        {
+          readonly worldKey: number;
+          readonly primitiveSlot: number;
+          readonly slotGeneration: number;
+          candidateCount: number;
+          visible: number;
+          occluded: number;
+          readonly lodHistogram: Map<number, number>;
+        }
+      >();
+      for (const batchSelection of selection.batches) {
+        const prepared = preparedByBatchId.get(batchSelection.batchId);
+        if (prepared === undefined) return selection;
+        let aggregate = worldSelections.get(prepared.worldKey);
+        if (aggregate === undefined) {
+          aggregate = {
+            worldKey: prepared.worldKey,
+            primitiveSlot: prepared.primitiveSlot,
+            slotGeneration: prepared.slotGeneration,
+            candidateCount: 0,
+            visible: 0,
+            occluded: 0,
+            lodHistogram: new Map(),
+          };
+          worldSelections.set(prepared.worldKey, aggregate);
+        }
+        aggregate.candidateCount += batchSelection.candidateCount;
+        aggregate.visible += batchSelection.visible;
+        aggregate.occluded += batchSelection.occluded;
+        for (const row of batchSelection.lodHistogram) {
+          aggregate.lodHistogram.set(
+            row.level,
+            (aggregate.lodHistogram.get(row.level) ?? 0) + row.count,
+          );
+        }
+      }
+      return {
+        ...selection,
+        candidateCount: this.telemetryCandidateCount,
+        occluded: Math.max(0, this.telemetryCandidateCount - selection.visible),
+        worldSelections: Object.freeze(
+          [...worldSelections.values()].map((aggregate) =>
+            Object.freeze({
+              worldKey: aggregate.worldKey,
+              primitiveSlot: aggregate.primitiveSlot,
+              slotGeneration: aggregate.slotGeneration,
+              candidateCount: aggregate.candidateCount,
+              visible: aggregate.visible,
+              occluded: aggregate.occluded,
+              lodHistogram: Object.freeze(
+                [...aggregate.lodHistogram.entries()]
+                  .sort(([left], [right]) => left - right)
+                  .map(([level, count]) => Object.freeze({ level, count })),
+              ),
+            } satisfies GpuDrivenWorldLodSelectionInspection),
+          ),
+        ),
+      };
+    });
+  }
+
   recordCpuValidation(
     validated: readonly { readonly source: RenderableSnapshot }[],
     gpuOwned: ReadonlySet<number>,
+    worldKeys?: readonly number[],
   ): void {
     this.cpuFallbackDrawItems = validated.length;
     let gpuOwnedRows = 0;
     for (const row of validated) {
-      if (gpuOwned.has(worldEntityKey(row.source.worldId, row.source.entityKey))) gpuOwnedRows += 1;
+      if (
+        gpuOwned.has(
+          worldEntityKey(
+            worldKeys?.[row.source.worldId] ?? row.source.worldId,
+            row.source.entityKey,
+          ),
+        )
+      ) {
+        gpuOwnedRows += 1;
+      }
     }
     this.validatedGpuOwnedRows = gpuOwnedRows;
   }