fix(simulation): align LCC layers in PlayCanvas world

apps/control-station/src/components/laboratory/LaboratoryMetricEvidenceScene.tsx

@@ -26,6 +26,7 @@ export interface LaboratoryMetricObstacleVisual {
   state: "current" | "retained" | "held" | "expired";
   centroidBodyXyzM: LaboratoryMetricPoint3;
   cellCentersBodyXyzM: readonly LaboratoryMetricPoint3[];
+  occupancySource?: "baseline" | "mixed" | "additive-low-step";
 }
 
 export interface LaboratoryMetricRigVisual {
@@ -41,7 +42,7 @@ export interface LaboratoryMetricCorridorVisual {
 }
 
 export interface LaboratoryMetricLegendEntry {
-  id: LaboratoryMetricDecision | "context" | "local-surface" | "rolling";
+  id: LaboratoryMetricDecision | "context" | "local-surface" | "rolling" | "low-step";
   label: string;
 }
 
@@ -52,6 +53,7 @@ export function laboratoryMetricLegendEntries({
   showCurrentIncrement,
   showLocalSurface,
   showRollingMap,
+  showLowStep = true,
 }: {
   pointCloudCount: number;
   localSurfaceCount: number;
@@ -59,19 +61,29 @@ export function laboratoryMetricLegendEntries({
   showCurrentIncrement: boolean;
   showLocalSurface: boolean;
   showRollingMap: boolean;
+  showLowStep?: boolean;
 }): readonly LaboratoryMetricLegendEntry[] {
   const visibleObstacles = obstacles.filter((obstacle) => (
-    obstacle.state === "current"
+    (showLowStep || obstacle.occupancySource === undefined || obstacle.occupancySource === "baseline")
+    && (obstacle.state === "current"
       ? showCurrentIncrement
       : obstacle.state === "retained"
         ? showRollingMap
-        : false
+        : false)
   ));
   const decisions = new Set(visibleObstacles.map(({ decision }) => decision));
   const entries: LaboratoryMetricLegendEntry[] = [];
   if (decisions.has("threat")) entries.push({ id: "threat", label: "Угроза" });
   if (decisions.has("not-threat")) entries.push({ id: "not-threat", label: "Вне коридора" });
   if (decisions.has("unknown")) entries.push({ id: "unknown", label: "Неизвестно" });
+  if (
+    showLowStep
+    && visibleObstacles.some((obstacle) => (
+      obstacle.occupancySource !== undefined && obstacle.occupancySource !== "baseline"
+    ))
+  ) {
+    entries.push({ id: "low-step", label: "LOW-STEP · добавлено системой" });
+  }
   if (showCurrentIncrement && pointCloudCount > 0) {
     entries.push({ id: "context", label: "Текущий кадр" });
   }
@@ -153,6 +165,16 @@ function decisionColor(
   return tokenColor(host, "--nodedc-warning-rgb", [255, 197, 92]);
 }
 
+function obstacleColor(
+  host: HTMLElement,
+  obstacle: LaboratoryMetricObstacleVisual,
+): THREE.Color {
+  if (obstacle.occupancySource !== "baseline") {
+    return tokenColor(host, "--nodedc-accent-rgb", [232, 56, 126]);
+  }
+  return decisionColor(host, obstacle.decision);
+}
+
 export interface LaboratoryMetricEvidenceSceneHandle {
   resetView: () => void;
 }
@@ -171,6 +193,7 @@ LaboratoryMetricEvidenceSceneHandle,
   showCurrentIncrement: boolean;
   showLocalSurface: boolean;
   showRollingMap: boolean;
+  showLowStep?: boolean;
   pointSemanticClassIds?: readonly (number | null)[];
   semanticClasses?: readonly RecordedEvidenceSemanticClass[];
   semanticPalette?: readonly RecordedEvidenceSemanticPaletteEntry[];
@@ -187,6 +210,7 @@ LaboratoryMetricEvidenceSceneHandle,
   showCurrentIncrement,
   showLocalSurface,
   showRollingMap,
+  showLowStep = true,
   pointSemanticClassIds,
   semanticClasses,
   semanticPalette,
@@ -352,6 +376,12 @@ LaboratoryMetricEvidenceSceneHandle,
 
     for (const obstacle of obstacles) {
       if (
+        (
+          !showLowStep
+          && obstacle.occupancySource !== undefined
+          && obstacle.occupancySource !== "baseline"
+        )
+        ||
         (obstacle.state === "current" && !showCurrentIncrement)
         || (obstacle.state === "retained" && !showRollingMap)
         || obstacle.state === "held"
@@ -359,7 +389,7 @@ LaboratoryMetricEvidenceSceneHandle,
       ) {
         continue;
       }
-      const color = decisionColor(host, obstacle.decision);
+      const color = obstacleColor(host, obstacle);
       if (obstacle.state === "retained") {
         const geometry = new THREE.BoxGeometry(
           occupiedVoxelSizeM * 0.82,
@@ -424,6 +454,7 @@ LaboratoryMetricEvidenceSceneHandle,
     showCurrentIncrement,
     showLocalSurface,
     showRollingMap,
+    showLowStep,
   ]);
 
   useEffect(() => {
@@ -538,6 +569,7 @@ LaboratoryMetricEvidenceSceneHandle,
     showCurrentIncrement,
     showLocalSurface,
     showRollingMap,
+    showLowStep,
   });
 
   return (

apps/control-station/src/components/simulation/PlayCanvasRuntime.ts

@@ -1,10 +1,18 @@
 import {
   Application,
   Asset,
+  BLEND_NORMAL,
   Color,
+  CULLFACE_NONE,
   Entity,
+  FILLMODE_NONE,
+  Mat4,
+  RESOLUTION_AUTO,
+  StandardMaterial,
   Vec2,
   Vec3,
+  type ContainerResource,
+  type RenderComponent,
   type ScriptType,
 } from "playcanvas";
 import { CameraControls } from "playcanvas/scripts/esm/camera-controls.mjs";
@@ -12,18 +20,58 @@ import { CameraControls } from "playcanvas/scripts/esm/camera-controls.mjs";
 import type { SimulationWorldManifest } from "../../core/simulation/projects";
 
 export type SimulationViewMode = "visual" | "collision" | "combined";
-export type SimulationQuality = "auto" | "low" | "medium" | "high";
+export type SimulationQuality = "low" | "medium" | "high" | "ultra" | "maximum";
+export type SimulationLayer = "visual" | "collision";
+
+export const PLAYCANVAS_IDENTITY_TRANSFORM = [
+  1, 0, 0, 0,
+  0, 1, 0, 0,
+  0, 0, 1, 0,
+  0, 0, 0, 1,
+];
+
+export const PLAYCANVAS_X_180_TRANSFORM = [
+  1, 0, 0, 0,
+  0, -1, 0, 0,
+  0, 0, -1, 0,
+  0, 0, 0, 1,
+];
 
 export interface SimulationRuntime {
   mount(canvas: HTMLCanvasElement): Promise<void>;
   loadWorld(manifest: SimulationWorldManifest): Promise<void>;
-  setViewMode(mode: SimulationViewMode): void;
+  setViewMode(mode: SimulationViewMode): Promise<void>;
   setQuality(quality: SimulationQuality): void;
+  setLayerWorldTransform(layer: SimulationLayer, transform: number[]): void;
+  home(): void;
   focusBounds(): void;
   dispose(): void;
 }
 
-type CameraController = ScriptType & Pick<CameraControls, "reset" | "focus">;
+type DesktopCameraInput = {
+  read(): { mouse: number[] } & Record<string, number[]>;
+};
+
+type CameraController = ScriptType & Pick<CameraControls, "reset" | "focus"> & {
+  _desktopInput?: DesktopCameraInput;
+};
+
+const HOME_POSITION = new Vec3(0, 1, 0);
+const HOME_FOCUS = new Vec3(1, 1, 0);
+
+const QUALITY_PROFILES: Record<SimulationQuality, {
+  lodBaseDistance: number;
+  lodMultiplier: number;
+  lodRangeMin: number;
+  lodRangeMax: number;
+  pixelRatio: number;
+}> = {
+  low: { lodBaseDistance: 5, lodMultiplier: 2, lodRangeMin: 3, lodRangeMax: 5, pixelRatio: 0.75 },
+  medium: { lodBaseDistance: 5, lodMultiplier: 2, lodRangeMin: 2, lodRangeMax: 5, pixelRatio: 1 },
+  high: { lodBaseDistance: 5, lodMultiplier: 3, lodRangeMin: 1, lodRangeMax: 5, pixelRatio: 1.5 },
+  ultra: { lodBaseDistance: 7, lodMultiplier: 3, lodRangeMin: 0, lodRangeMax: 5, pixelRatio: 2 },
+  maximum: { lodBaseDistance: 7, lodMultiplier: 3, lodRangeMin: 0, lodRangeMax: 0, pixelRatio: 2 },
+};
 
 export class PlayCanvasRuntime implements SimulationRuntime {
   private app: Application | null = null;
@@ -32,15 +80,21 @@ export class PlayCanvasRuntime implements SimulationRuntime {
   private cameraController: CameraController | null = null;
   private visualEntity: Entity | null = null;
   private visualAsset: Asset | null = null;
+  private collisionEntity: Entity | null = null;
+  private collisionAsset: Asset | null = null;
+  private collisionMaterial: StandardMaterial | null = null;
+  private collisionSource: { meshUrl: string; projectId: string; worldTransform: number[] } | null = null;
+  private collisionLoadPromise: Promise<void> | null = null;
   private resizeObserver: ResizeObserver | null = null;
   private viewMode: SimulationViewMode = "visual";
-  private quality: SimulationQuality = "auto";
+  private quality: SimulationQuality = "maximum";
 
   async mount(canvas: HTMLCanvasElement): Promise<void> {
     if (this.app) throw new Error("PlayCanvas runtime уже смонтирован.");
     this.canvas = canvas;
     canvas.tabIndex = 0;
     canvas.addEventListener("contextmenu", preventContextMenu);
+    canvas.addEventListener("pointerdown", focusCanvas, true);
     const app = new Application(canvas, {
       graphicsDeviceOptions: {
         antialias: true,
@@ -50,6 +104,8 @@ export class PlayCanvasRuntime implements SimulationRuntime {
       },
     });
     this.app = app;
+    app.setCanvasFillMode(FILLMODE_NONE, 1, 1);
+    app.setCanvasResolution(RESOLUTION_AUTO);
     app.scene.ambientLight = new Color(0.35, 0.37, 0.42);
 
     const camera = new Entity("SimulationCamera");
@@ -59,8 +115,8 @@ export class PlayCanvasRuntime implements SimulationRuntime {
       farClip: 20_000,
       fov: 58,
     });
-    camera.setPosition(5, 3, 5);
+    camera.setPosition(HOME_POSITION);
-    camera.lookAt(0, 0, 0);
+    camera.lookAt(HOME_FOCUS);
     camera.addComponent("script");
     const controller = camera.script?.create(CameraControls, {
       properties: {
@@ -70,6 +126,7 @@ export class PlayCanvasRuntime implements SimulationRuntime {
         zoomRange: new Vec2(0.05, 20_000),
       },
     }) as CameraController | null;
+    if (controller) invertHorizontalCameraDrag(controller);
     app.root.addChild(camera);
     this.camera = camera;
     this.cameraController = controller;
@@ -103,6 +160,7 @@ export class PlayCanvasRuntime implements SimulationRuntime {
       const ready = (loaded: Asset) => {
         const visual = new Entity("GaussianWorld");
         visual.enabled = false;
+        applyWorldTransform(visual, manifest.transforms.worldFromVisual);
         app.root.addChild(visual);
         visual.addComponent("gsplat", {
           asset: loaded,
@@ -112,7 +170,6 @@ export class PlayCanvasRuntime implements SimulationRuntime {
         this.visualEntity = visual;
         this.applyQuality();
         this.applyViewMode();
-        this.focusBounds();
         resolve();
       };
       const failed = (error: unknown) => {
@@ -122,11 +179,24 @@ export class PlayCanvasRuntime implements SimulationRuntime {
       asset.once("error", failed);
       app.assets.load(asset);
     });
+    if (manifest.collision.available && manifest.collision.meshUrl) {
+      this.collisionSource = {
+        meshUrl: manifest.collision.meshUrl,
+        projectId: manifest.projectId,
+        worldTransform: manifest.transforms.worldFromCollision,
+      };
+    }
+    this.applyViewMode();
+    this.home();
   }
 
-  setViewMode(mode: SimulationViewMode): void {
+  async setViewMode(mode: SimulationViewMode): Promise<void> {
     this.viewMode = mode;
     this.applyViewMode();
+    if (mode !== "visual" && !this.collisionEntity) {
+      await this.ensureCollision();
+      this.applyViewMode();
+    }
   }
 
   setQuality(quality: SimulationQuality): void {
@@ -134,6 +204,28 @@ export class PlayCanvasRuntime implements SimulationRuntime {
     this.applyQuality();
   }
 
+  setLayerWorldTransform(layer: SimulationLayer, transform: number[]): void {
+    if (transform.length !== 16) {
+      throw new Error("World transform должен содержать матрицу 4×4.");
+    }
+    if (layer === "visual") {
+      if (this.visualEntity) applyWorldTransform(this.visualEntity, transform);
+      return;
+    }
+    if (this.collisionSource) this.collisionSource.worldTransform = [...transform];
+    if (this.collisionEntity) applyWorldTransform(this.collisionEntity, transform);
+  }
+
+  home(): void {
+    if (this.cameraController) {
+      this.cameraController.reset(HOME_FOCUS, HOME_POSITION);
+    } else if (this.camera) {
+      this.camera.setPosition(HOME_POSITION);
+      this.camera.lookAt(HOME_FOCUS);
+    }
+    this.canvas?.focus({ preventScroll: true });
+  }
+
   focusBounds(): void {
     const bounds = this.visualEntity?.gsplat?.customAabb;
     const focus = bounds?.center.clone() ?? new Vec3(0, 0, 0);
@@ -153,7 +245,10 @@ export class PlayCanvasRuntime implements SimulationRuntime {
     this.resizeObserver = null;
     this.unloadWorld();
     if (this.app) this.app.destroy();
-    if (this.canvas) this.canvas.removeEventListener("contextmenu", preventContextMenu);
+    if (this.canvas) {
+      this.canvas.removeEventListener("contextmenu", preventContextMenu);
+      this.canvas.removeEventListener("pointerdown", focusCanvas, true);
+    }
     this.app = null;
     this.canvas = null;
     this.camera = null;
@@ -175,22 +270,100 @@ export class PlayCanvasRuntime implements SimulationRuntime {
 
   private applyViewMode(): void {
     if (this.visualEntity) {
-      this.visualEntity.enabled = this.viewMode !== "collision";
+      this.visualEntity.enabled = this.viewMode !== "collision" || !this.collisionEntity;
+    }
+    if (this.collisionEntity) {
+      this.collisionEntity.enabled = this.viewMode !== "visual";
+    }
+    if (this.collisionMaterial) {
+      const combined = this.viewMode === "combined";
+      this.collisionMaterial.opacity = combined ? 0.34 : 0.82;
+      this.collisionMaterial.depthWrite = !combined;
+      this.collisionMaterial.update();
     }
   }
 
   private applyQuality(): void {
     const gsplat = this.visualEntity?.gsplat;
-    if (!gsplat) return;
+    const profile = QUALITY_PROFILES[this.quality];
-    const profiles: Record<SimulationQuality, [number, number]> = {
+    if (gsplat) {
-      auto: [5, 2],
+      gsplat.lodBaseDistance = profile.lodBaseDistance;
-      low: [2.5, 1.7],
+      gsplat.lodMultiplier = profile.lodMultiplier;
-      medium: [5, 2],
+      gsplat.lodRangeMin = profile.lodRangeMin;
-      high: [9, 2.3],
+      gsplat.lodRangeMax = profile.lodRangeMax;
-    };
+    }
-    const [baseDistance, multiplier] = profiles[this.quality];
+    if (this.app) {
-    gsplat.lodBaseDistance = baseDistance;
+      const devicePixelRatio = window.devicePixelRatio || 1;
-    gsplat.lodMultiplier = multiplier;
+      this.app.graphicsDevice.maxPixelRatio = Math.min(devicePixelRatio, profile.pixelRatio);
+      this.resize();
+    }
+  }
+
+  private async loadCollision(meshUrl: string, projectId: string, worldTransform: number[]): Promise<void> {
+    const app = this.requiredApp();
+    const asset = new Asset(
+      `SimulationCollision:${projectId}`,
+      "container",
+      { url: meshUrl, filename: "scene.collision.glb" },
+    );
+    app.assets.add(asset);
+    this.collisionAsset = asset;
+    await new Promise<void>((resolve, reject) => {
+      asset.ready((loaded) => {
+        const resource = loaded.resource as ContainerResource | null;
+        if (!resource) {
+          reject(new Error("PlayCanvas не открыл collision GLB."));
+          return;
+        }
+        const entity = resource.instantiateRenderEntity({
+          castShadows: false,
+          receiveShadows: false,
+        });
+        entity.name = "CollisionWorld";
+        entity.enabled = false;
+        applyWorldTransform(entity, worldTransform);
+        app.root.addChild(entity);
+
+        const material = new StandardMaterial();
+        material.name = "SimulationCollisionMaterial";
+        material.diffuse = new Color(0.2, 0.95, 0.42);
+        material.emissive = new Color(0.08, 0.36, 0.14);
+        material.opacity = 0.34;
+        material.blendType = BLEND_NORMAL;
+        material.depthWrite = false;
+        material.cull = CULLFACE_NONE;
+        material.update();
+        for (const render of entity.findComponents("render") as RenderComponent[]) {
+          for (const meshInstance of render.meshInstances) {
+            meshInstance.material = material;
+          }
+        }
+        this.collisionEntity = entity;
+        this.collisionMaterial = material;
+        resolve();
+      });
+      asset.once("error", (error: unknown) => {
+        reject(new Error(error instanceof Error ? error.message : "PlayCanvas не загрузил collision GLB."));
+      });
+      app.assets.load(asset);
+    });
+  }
+
+  private async ensureCollision(): Promise<void> {
+    if (this.collisionEntity) return;
+    if (!this.collisionSource) {
+      throw new Error("Collision GLB отсутствует в world manifest.");
+    }
+    if (!this.collisionLoadPromise) {
+      this.collisionLoadPromise = this.loadCollision(
+        this.collisionSource.meshUrl,
+        this.collisionSource.projectId,
+        this.collisionSource.worldTransform,
+      ).finally(() => {
+        this.collisionLoadPromise = null;
+      });
+    }
+    await this.collisionLoadPromise;
   }
 
   private unloadWorld(): void {
@@ -203,9 +376,46 @@ export class PlayCanvasRuntime implements SimulationRuntime {
       this.visualAsset.unload();
       this.visualAsset = null;
     }
+    if (this.collisionEntity) {
+      this.collisionEntity.destroy();
+      this.collisionEntity = null;
+    }
+    if (this.collisionAsset && this.app) {
+      this.app.assets.remove(this.collisionAsset);
+      this.collisionAsset.unload();
+      this.collisionAsset = null;
+    }
+    this.collisionMaterial?.destroy();
+    this.collisionMaterial = null;
+    this.collisionSource = null;
+    this.collisionLoadPromise = null;
   }
 }
 
 function preventContextMenu(event: Event): void {
   event.preventDefault();
 }
+
+function focusCanvas(event: Event): void {
+  (event.currentTarget as HTMLCanvasElement | null)?.focus({ preventScroll: true });
+}
+
+function invertHorizontalCameraDrag(controller: CameraController): void {
+  // CameraControls 2.21.4 deliberately exposes one pinned input source here.
+  // Keep its pointer-capture and fly/orbit behavior, changing only horizontal drag semantics.
+  const input = controller._desktopInput;
+  if (!input) return;
+  const read = input.read.bind(input);
+  input.read = () => {
+    const frame = read();
+    if (frame.mouse.length > 0) frame.mouse[0] *= -1;
+    return frame;
+  };
+}
+
+function applyWorldTransform(entity: Entity, values: number[]): void {
+  const transform = new Mat4().set(values);
+  entity.setLocalPosition(transform.getTranslation());
+  entity.setLocalEulerAngles(transform.getEulerAngles());
+  entity.setLocalScale(transform.getScale());
+}

apps/control-station/src/components/simulation/SimulationViewport.tsx

@@ -1,8 +1,19 @@
-import { useEffect, useRef, useState } from "react";
+import { useEffect, useId, useRef, useState } from "react";
-import { ActivityIndicator, Button, SegmentedControl, StatusBadge } from "@nodedc/ui-react";
+import {
+  ActivityIndicator,
+  Button,
+  Icon,
+  IconButton,
+  SegmentedControl,
+  Select,
+  StatusBadge,
+  Switch,
+} from "@nodedc/ui-react";
 
 import type { SimulationProject } from "../../core/simulation/projects";
 import {
+  PLAYCANVAS_IDENTITY_TRANSFORM,
+  PLAYCANVAS_X_180_TRANSFORM,
   PlayCanvasRuntime,
   type SimulationQuality,
   type SimulationViewMode,
@@ -11,10 +22,25 @@ import {
 export function SimulationViewport({ project }: { project: SimulationProject }) {
   const canvasRef = useRef<HTMLCanvasElement>(null);
   const runtimeRef = useRef<PlayCanvasRuntime | null>(null);
+  const settingsId = useId();
   const [state, setState] = useState<"mounting" | "loading" | "ready" | "failed">("mounting");
   const [error, setError] = useState<string | null>(null);
   const [viewMode, setViewMode] = useState<SimulationViewMode>("visual");
-  const [quality, setQuality] = useState<SimulationQuality>("auto");
+  const [quality, setQuality] = useState<SimulationQuality>("maximum");
+  const [collisionState, setCollisionState] = useState<"idle" | "loading" | "ready" | "failed">("idle");
+  const [settingsOpen, setSettingsOpen] = useState(false);
+  const [visualInverted, setVisualInverted] = useState(
+    () => isX180Transform(project.worldManifest?.transforms.worldFromVisual),
+  );
+  const [collisionInverted, setCollisionInverted] = useState(
+    () => isX180Transform(project.worldManifest?.transforms.worldFromCollision),
+  );
+
+  useEffect(() => {
+    setVisualInverted(isX180Transform(project.worldManifest?.transforms.worldFromVisual));
+    setCollisionInverted(isX180Transform(project.worldManifest?.transforms.worldFromCollision));
+    setSettingsOpen(false);
+  }, [project.projectId, project.worldManifest]);
 
   useEffect(() => {
     const canvas = canvasRef.current;
@@ -48,17 +74,53 @@ export function SimulationViewport({ project }: { project: SimulationProject })
     <section className="simulation-viewport" aria-label={`Сцена ${project.name}`}>
       <header className="simulation-viewport__toolbar">
         <div>
-          <StatusBadge tone={state === "ready" ? "success" : state === "failed" ? "warning" : "accent"}>
+          <StatusBadge tone={state === "failed" || collisionState === "failed" ? "warning" : state === "ready" && collisionState !== "loading" ? "success" : "accent"}>
-            {state === "ready" ? "Runtime готов" : state === "failed" ? "Ошибка runtime" : "Загрузка сцены"}
+            {collisionState === "loading"
+              ? "Загрузка collision"
+              : collisionState === "failed"
+                ? "Ошибка collision"
+                : state === "ready"
+                  ? "Runtime готов"
+                  : state === "failed"
+                    ? "Ошибка runtime"
+                    : "Загрузка сцены"}
           </StatusBadge>
           <span>PlayCanvas Engine 2.21.4</span>
         </div>
+        <div className="simulation-viewport__controls">
+          <Select
+            label="Качество Gaussian-сцены"
+            value={quality}
+            onChange={(next) => {
+              setQuality(next);
+              runtimeRef.current?.setQuality(next);
+            }}
+            options={[
+              { value: "maximum", label: "Максимум", description: "Только полный LOD 0, Retina до 2×" },
+              { value: "ultra", label: "Ультра", description: "LOD 0 и Retina до 2×" },
+              { value: "high", label: "Высокое", description: "LOD 1 и Retina до 1,5×" },
+              { value: "medium", label: "Среднее", description: "LOD 2 и обычное разрешение" },
+              { value: "low", label: "Низкое", description: "LOD 3 для слабых устройств" },
+            ]}
+            minMenuWidth={230}
+            menuWidth={230}
+          />
           <SegmentedControl
             label="Слой сцены"
             value={viewMode}
             onChange={(next) => {
               setViewMode(next);
-            runtimeRef.current?.setViewMode(next);
+              const runtime = runtimeRef.current;
+              if (!runtime) return;
+              if (next !== "visual" && collisionAvailable && collisionState !== "ready") {
+                setCollisionState("loading");
+              }
+              void runtime.setViewMode(next).then(() => {
+                if (next !== "visual") setCollisionState("ready");
+              }).catch((caught: unknown) => {
+                setCollisionState("failed");
+                setError(caught instanceof Error ? caught.message : "Не удалось открыть collision GLB.");
+              });
             }}
             items={[
               { value: "visual", label: "Визуал" },
@@ -66,23 +128,70 @@ export function SimulationViewport({ project }: { project: SimulationProject })
               { value: "combined", label: "Вместе", disabled: !collisionAvailable },
             ]}
           />
-        <SegmentedControl
+          <Button size="compact" variant="secondary" onClick={() => runtimeRef.current?.home()}>
-          label="Качество Streamed SOG"
+            Домой
-          value={quality}
-          onChange={(next) => {
-            setQuality(next);
-            runtimeRef.current?.setQuality(next);
-          }}
-          items={[
-            { value: "auto", label: "Auto" },
-            { value: "low", label: "Low" },
-            { value: "medium", label: "Med" },
-            { value: "high", label: "High" },
-          ]}
-        />
-        <Button size="compact" variant="secondary" onClick={() => runtimeRef.current?.focusBounds()}>
-          Вписать сцену
           </Button>
+          <div className="simulation-viewport__settings-anchor">
+            <IconButton
+              label="Настройки системы координат"
+              aria-controls={settingsId}
+              aria-expanded={settingsOpen}
+              aria-pressed={settingsOpen}
+              onClick={() => setSettingsOpen((open) => !open)}
+            >
+              <Icon name="settings" size={18} />
+            </IconButton>
+            {settingsOpen ? (
+              <div
+                id={settingsId}
+                className="simulation-viewport__settings"
+                role="dialog"
+                aria-label="Настройки системы координат"
+              >
+                <div className="simulation-viewport__settings-head">
+                  <div>
+                    <strong>Система координат</strong>
+                    <span>Мир PlayCanvas · Y вверх</span>
+                  </div>
+                  <IconButton label="Закрыть настройки" onClick={() => setSettingsOpen(false)}>
+                    <Icon name="close" size={16} />
+                  </IconButton>
+                </div>
+                <p>
+                  Коррекция применяется к слоям независимо и не меняет координаты камеры,
+                  навигации и будущей физики.
+                </p>
+                <div className="simulation-viewport__settings-switches">
+                  <Switch
+                    checked={visualInverted}
+                    disabled={state !== "ready"}
+                    label="Инверсия визуального слоя"
+                    onChange={(checked) => {
+                      setVisualInverted(checked);
+                      runtimeRef.current?.setLayerWorldTransform(
+                        "visual",
+                        checked ? PLAYCANVAS_X_180_TRANSFORM : PLAYCANVAS_IDENTITY_TRANSFORM,
+                      );
+                    }}
+                  />
+                  <Switch
+                    checked={collisionInverted}
+                    disabled={state !== "ready"}
+                    label="Инверсия collision-слоя"
+                    onChange={(checked) => {
+                      setCollisionInverted(checked);
+                      runtimeRef.current?.setLayerWorldTransform(
+                        "collision",
+                        checked ? PLAYCANVAS_X_180_TRANSFORM : PLAYCANVAS_IDENTITY_TRANSFORM,
+                      );
+                    }}
+                  />
+                </div>
+              </div>
+            ) : null}
+          </div>
+        </div>
+        <span className="simulation-viewport__toolbar-balance" aria-hidden="true" />
       </header>
       <div className="simulation-viewport__stage">
         <canvas ref={canvasRef} aria-label={`PlayCanvas сцена ${project.name}`} />
@@ -99,7 +208,17 @@ export function SimulationViewport({ project }: { project: SimulationProject })
           <StatusBadge tone="warning">Collision недоступен</StatusBadge>
           <span>Для этой сборки collision GLB не был запрошен; визуальный слой настоящий и не подменяется.</span>
         </footer>
+      ) : collisionState === "failed" ? (
+        <footer className="simulation-viewport__notice" role="alert">
+          <StatusBadge tone="warning">Collision не открылся</StatusBadge>
+          <span>{error ?? "PlayCanvas не смог загрузить collision GLB."}</span>
+        </footer>
       ) : null}
     </section>
   );
 }
+
+function isX180Transform(transform: number[] | undefined): boolean {
+  if (!transform || transform.length !== 16) return false;
+  return transform.every((value, index) => value === PLAYCANVAS_X_180_TRANSFORM[index]);
+}

apps/control-station/src/core/laboratory/advancedIndex.ts

@@ -43,6 +43,7 @@ import {
 } from "./m48ObjectCentricQuality";
 import { fetchM48SmallStaticRegression } from "./m48SmallStaticRegression";
 import { fetchM48StaticOccupancyQualification } from "./m48StaticOccupancyQualification";
+import { fetchM48R3StaticOccupancyShadow } from "./m48r3StaticOccupancyShadow";
 import { fetchM48SFixedClassDetectorResult } from "./m48sFixedClassDetector";
 import { fetchM48TRiskQualityResult } from "./m48tRiskQuality";
 
@@ -50,6 +51,7 @@ export type AdvancedLaboratoryWorkId =
   | "m48-object-centric-quality"
   | "m48-small-static-passage-regression"
   | "m48-static-occupancy-qualification"
+  | "m48r3-static-occupancy-shadow"
   | "m48s-fixed-class-detector"
   | "m48t-risk-quality-temporal"
   | "m47-reference-graph-shadow"
@@ -97,6 +99,7 @@ const WORK_IDS: readonly AdvancedLaboratoryWorkId[] = [
   "m48-object-centric-quality",
   "m48-small-static-passage-regression",
   "m48-static-occupancy-qualification",
+  "m48r3-static-occupancy-shadow",
   "m48s-fixed-class-detector",
   "m48t-risk-quality-temporal",
   "m47-reference-graph-shadow",
@@ -139,6 +142,7 @@ const RESULT_PREFIX: Readonly<Record<AdvancedLaboratoryWorkId, string>> = {
   "m48-object-centric-quality": "m48-object-quality-(?:pack|result)",
   "m48-small-static-passage-regression": "m48-small-static-passage-regression",
   "m48-static-occupancy-qualification": "m48-static-occupancy-qualification",
+  "m48r3-static-occupancy-shadow": "m48r3-static-occupancy-shadow",
   "m48s-fixed-class-detector": "m48s-fixed-class-detector-lab",
   "m48t-risk-quality-temporal": "(?:m48t-risk-quality-temporal-lab|m48q-native-risk-quality-lab)",
   "m47-reference-graph-shadow": "m47-reference-graph-lab",
@@ -189,6 +193,7 @@ export function emptyAdvancedLaboratoryResults(): AdvancedLaboratoryResults {
     m48: null,
     m48SmallStatic: null,
     m48StaticOccupancy: null,
+    m48r3StaticOccupancy: null,
     m48s: null,
     m48t: null,
     m4Threat: null,
@@ -318,6 +323,7 @@ export function advancedLaboratoryResultAvailable(
   return workId === "m48-object-centric-quality" ? results.m48 !== null
     : workId === "m48-small-static-passage-regression" ? results.m48SmallStatic !== null
     : workId === "m48-static-occupancy-qualification" ? results.m48StaticOccupancy !== null
+    : workId === "m48r3-static-occupancy-shadow" ? results.m48r3StaticOccupancy !== null
     : workId === "m48s-fixed-class-detector" ? results.m48s !== null
     : workId === "m48t-risk-quality-temporal" ? results.m48t !== null
     : workId === "m47-reference-graph-shadow" ? results.m47Graph !== null
@@ -378,6 +384,9 @@ export async function fetchAdvancedLaboratoryResult(
   } else if (workId === "m48-static-occupancy-qualification") {
     if (!resultId) throw new AdvancedLaboratoryContractError("M4.8R2 qualification identity не выбрана.");
     results.m48StaticOccupancy = await fetchM48StaticOccupancyQualification(resultId, { fetcher, signal });
+  } else if (workId === "m48r3-static-occupancy-shadow") {
+    if (!resultId) throw new AdvancedLaboratoryContractError("M4.8R3 Worker shadow identity не выбрана.");
+    results.m48r3StaticOccupancy = await fetchM48R3StaticOccupancyShadow(resultId, { fetcher, signal });
   } else if (workId === "m48s-fixed-class-detector") {
     if (!resultId) throw new AdvancedLaboratoryContractError("M4.8S LAB identity не выбрана.");
     results.m48s = await fetchM48SFixedClassDetectorResult(resultId, { fetcher, signal });

apps/control-station/src/core/laboratory/advancedLaboratoryResults.ts

@@ -37,6 +37,7 @@ import type { M47ReferenceGraphLabResult } from "./m47ReferenceGraph";
 import type { M48AdvancedResult } from "./m48ObjectCentricQuality";
 import type { M48SmallStaticRegressionResult } from "./m48SmallStaticRegression";
 import type { M48StaticOccupancyQualificationResult } from "./m48StaticOccupancyQualification";
+import type { M48R3StaticOccupancyShadowResult } from "./m48r3StaticOccupancyShadow";
 import type { M48SFixedClassDetectorResult } from "./m48sFixedClassDetector";
 import type { M48TRiskQualityResult } from "./m48tRiskQuality";
 
@@ -45,6 +46,7 @@ export interface AdvancedLaboratoryResults {
   m48: M48AdvancedResult | null;
   m48SmallStatic: M48SmallStaticRegressionResult | null;
   m48StaticOccupancy: M48StaticOccupancyQualificationResult | null;
+  m48r3StaticOccupancy: M48R3StaticOccupancyShadowResult | null;
   m48s: M48SFixedClassDetectorResult | null;
   m48t: M48TRiskQualityResult | null;
   m4Threat: M4ThreatReplayResult | null;

apps/control-station/src/core/laboratory/advancedResults.ts

@@ -968,6 +968,7 @@ export async function fetchAdvancedLaboratoryResults({
   const e40 = settledCatalogValue(settled[8]);
   return {
     m47Graph: null, m48: null, m48SmallStatic: null, m48StaticOccupancy: null,
+    m48r3StaticOccupancy: null,
     m48s: null, m48t: null, m4Threat: null,
     l3: null, l31: null, l32: null, l33: null,
     e31,

apps/control-station/src/core/laboratory/m48r3StaticOccupancyShadow.ts

@@ -0,0 +1,343 @@
+import type { LaboratoryFetch } from "./advancedResults";
+import type { M48Authority } from "./m48ObjectCentricQuality";
+
+const RESULT_ID = /^m48r3-static-occupancy-shadow-[a-f0-9]{64}$/;
+
+interface M48R3WorkerPerformance {
+  admittedFrames: number;
+  deliveredFrames: number;
+  fps: number;
+  worldP95Ms: number;
+  worldP99Ms: number;
+  geometryP95Ms: number;
+  geometryP99Ms: number;
+}
+
+export interface M48R3StaticOccupancyShadowResult {
+  resultId: string;
+  createdAtUtc: string;
+  accepted: boolean;
+  profile: {
+    id: string;
+    sha256: string;
+    minimumPoints: number;
+  };
+  metrics: {
+    frames: { expected: number; baselineDelivered: number; candidateDelivered: number };
+    performance: {
+      baseline: M48R3WorkerPerformance;
+      candidate: M48R3WorkerPerformance;
+      fpsRegressionFraction: number;
+      worldStateP95DeltaMs: number;
+    };
+    occupancy: {
+      baselineCellTotal: number;
+      candidateCellTotal: number;
+      addedCellTotal: number;
+      lostCellTotal: number;
+      baselineComponentTotal: number;
+      candidateComponentTotal: number;
+      meanCellGrowthFraction: number;
+      meanComponentGrowthFraction: number;
+      maximumAddedCellsPerFrame: number;
+      maximumCandidateComponentsPerFrame: number;
+    };
+    provider: {
+      additiveObservationCount: number;
+      additiveVoxelCount: number;
+      framesWithAdditions: number;
+      additiveMeanMsPerFrame: number;
+      peakTemporalComponents: number;
+      peakRollingCells: number;
+    };
+    anchors: {
+      count: number;
+      criticalNearCount: number;
+      criticalNearRecall: number;
+      matchedCount: number;
+      canonicalEngineeringRecall: number;
+      separation: readonly {
+        displayFrame: number;
+        expectedMinimumComponents: number;
+        observedComponents: number;
+        passed: boolean;
+        interpretation: string;
+      }[];
+    };
+    capacityDropCount: number;
+    falseFreeCount: number;
+  };
+  gates: Readonly<Record<string, boolean>>;
+  decision: {
+    state: "accepted-bounded-worker-shadow" | "rejected-bounded-worker-shadow";
+    candidateAccepted: boolean;
+    productionAccepted: false;
+    nextAction: string;
+  };
+  limitations: readonly string[];
+  authority: M48Authority;
+}
+
+export interface M48R3StaticOccupancyCase {
+  anchorId: string;
+  displayFrame: number;
+  sourceSequence: number;
+  extentXyxy: readonly [number, number, number, number];
+  distanceBand: "critical-near" | "approach" | "outside-qualified-bands";
+  componentCount: number;
+  matched: boolean;
+}
+
+export class M48R3StaticOccupancyContractError extends Error {}
+
+function objectValue(value: unknown, label: string): Record<string, unknown> {
+  if (!value || typeof value !== "object" || Array.isArray(value)) {
+    throw new M48R3StaticOccupancyContractError(`${label}: ожидался объект.`);
+  }
+  return value as Record<string, unknown>;
+}
+
+function text(value: unknown, label: string): string {
+  if (typeof value !== "string" || !value.trim()) {
+    throw new M48R3StaticOccupancyContractError(`${label}: ожидалась строка.`);
+  }
+  return value;
+}
+
+function numberValue(value: unknown, label: string): number {
+  if (typeof value !== "number" || !Number.isFinite(value)) {
+    throw new M48R3StaticOccupancyContractError(`${label}: ожидалось число.`);
+  }
+  return value;
+}
+
+function integer(value: unknown, label: string): number {
+  const parsed = numberValue(value, label);
+  if (!Number.isSafeInteger(parsed) || parsed < 0) {
+    throw new M48R3StaticOccupancyContractError(`${label}: ожидалось целое число.`);
+  }
+  return parsed;
+}
+
+function bool(value: unknown, label: string): boolean {
+  if (typeof value !== "boolean") {
+    throw new M48R3StaticOccupancyContractError(`${label}: ожидался флаг.`);
+  }
+  return value;
+}
+
+function exact(value: unknown, expected: string | boolean, label: string): void {
+  if (value !== expected) {
+    throw new M48R3StaticOccupancyContractError(`${label}: нарушен контракт.`);
+  }
+}
+
+function sha256(value: unknown, label: string): string {
+  const parsed = text(value, label);
+  if (!/^[a-f0-9]{64}$/.test(parsed)) {
+    throw new M48R3StaticOccupancyContractError(`${label}: неверный SHA-256.`);
+  }
+  return parsed;
+}
+
+function extent(value: unknown): readonly [number, number, number, number] {
+  if (!Array.isArray(value) || value.length !== 4) {
+    throw new M48R3StaticOccupancyContractError("M4.8R3 extent: нарушен контракт.");
+  }
+  const parsed = value.map((item) => numberValue(item, "M4.8R3 extent"));
+  return [parsed[0]!, parsed[1]!, parsed[2]!, parsed[3]!];
+}
+
+function authority(value: unknown): M48Authority {
+  const row = objectValue(value, "M4.8R3 authority");
+  exact(row.mode, "replay-simulated", "M4.8R3 authority.mode");
+  exact(row.physical_live, false, "M4.8R3 authority.physical_live");
+  exact(row.commands_enabled, false, "M4.8R3 authority.commands_enabled");
+  exact(row.actuation_allowed, false, "M4.8R3 authority.actuation_allowed");
+  exact(
+    row.navigation_or_safety_accepted,
+    false,
+    "M4.8R3 authority.navigation_or_safety_accepted",
+  );
+  return {
+    mode: "replay-simulated",
+    physicalLive: false,
+    commandsEnabled: false,
+    actuationAllowed: false,
+    navigationOrSafetyAccepted: false,
+  };
+}
+
+function workerPerformance(value: unknown, label: string): M48R3WorkerPerformance {
+  const row = objectValue(value, label);
+  return {
+    admittedFrames: integer(row.admitted_frames, `${label}.admitted_frames`),
+    deliveredFrames: integer(row.delivered_frames, `${label}.delivered_frames`),
+    fps: numberValue(row.fps, `${label}.fps`),
+    worldP95Ms: numberValue(row.world_p95_ms, `${label}.world_p95_ms`),
+    worldP99Ms: numberValue(row.world_p99_ms, `${label}.world_p99_ms`),
+    geometryP95Ms: numberValue(row.geometry_p95_ms, `${label}.geometry_p95_ms`),
+    geometryP99Ms: numberValue(row.geometry_p99_ms, `${label}.geometry_p99_ms`),
+  };
+}
+
+export async function fetchM48R3StaticOccupancyShadow(
+  resultId: string,
+  { fetcher = fetch, signal }: { fetcher?: LaboratoryFetch; signal?: AbortSignal } = {},
+): Promise<M48R3StaticOccupancyShadowResult> {
+  if (!RESULT_ID.test(resultId)) {
+    throw new M48R3StaticOccupancyContractError("M4.8R3 identity недопустима.");
+  }
+  const response = await fetcher(
+    `/api/v1/laboratory/m48r3/static-occupancy/${encodeURIComponent(resultId)}`,
+    { method: "GET", headers: { Accept: "application/json" }, signal },
+  );
+  if (!response.ok) {
+    throw new M48R3StaticOccupancyContractError(`M4.8R3 недоступен: HTTP ${response.status}.`);
+  }
+  const payload = objectValue(await response.json(), "M4.8R3");
+  exact(
+    payload.schema_version,
+    "missioncore.m48r3-static-occupancy-shadow-view/v1",
+    "M4.8R3 schema",
+  );
+  exact(payload.result_id, resultId, "M4.8R3 result");
+  exact(payload.ground_truth, false, "M4.8R3 ground truth");
+  const profile = objectValue(payload.profile, "M4.8R3 profile");
+  const componentization = objectValue(profile.componentization, "M4.8R3 componentization");
+  const metrics = objectValue(payload.metrics, "M4.8R3 metrics");
+  const frames = objectValue(metrics.frames, "M4.8R3 frames");
+  const performance = objectValue(metrics.performance, "M4.8R3 performance");
+  const occupancy = objectValue(metrics.occupancy, "M4.8R3 occupancy");
+  const provider = objectValue(metrics.provider, "M4.8R3 provider");
+  const anchors = objectValue(metrics.assisted_anchors, "M4.8R3 anchors");
+  if (!Array.isArray(anchors.separation)) {
+    throw new M48R3StaticOccupancyContractError("M4.8R3 separation: ожидался массив.");
+  }
+  const gates = objectValue(payload.gates, "M4.8R3 gates");
+  const parsedGates = Object.fromEntries(
+    Object.entries(gates).map(([key, value]) => [key, bool(value, `M4.8R3 gate ${key}`)]),
+  );
+  const decision = objectValue(payload.decision, "M4.8R3 decision");
+  const state = text(decision.state, "M4.8R3 decision.state");
+  if (state !== "accepted-bounded-worker-shadow" && state !== "rejected-bounded-worker-shadow") {
+    throw new M48R3StaticOccupancyContractError("M4.8R3 decision.state: неизвестное состояние.");
+  }
+  exact(decision.production_accepted, false, "M4.8R3 production acceptance");
+  if (!Array.isArray(payload.limitations)) {
+    throw new M48R3StaticOccupancyContractError("M4.8R3 limitations: ожидался массив.");
+  }
+  return {
+    resultId,
+    createdAtUtc: text(payload.created_at_utc, "M4.8R3 created"),
+    accepted: bool(payload.accepted, "M4.8R3 accepted"),
+    profile: {
+      id: text(profile.id, "M4.8R3 profile id"),
+      sha256: sha256(profile.sha256, "M4.8R3 profile sha"),
+      minimumPoints: integer(componentization.minimum_points, "M4.8R3 minimum points"),
+    },
+    metrics: {
+      frames: {
+        expected: integer(frames.expected, "M4.8R3 expected frames"),
+        baselineDelivered: integer(frames.baseline_delivered, "M4.8R3 baseline frames"),
+        candidateDelivered: integer(frames.candidate_delivered, "M4.8R3 candidate frames"),
+      },
+      performance: {
+        baseline: workerPerformance(performance.baseline, "M4.8R3 baseline"),
+        candidate: workerPerformance(performance.candidate, "M4.8R3 candidate"),
+        fpsRegressionFraction: numberValue(performance.fps_regression_fraction, "M4.8R3 FPS regression"),
+        worldStateP95DeltaMs: numberValue(performance.world_state_p95_delta_ms, "M4.8R3 p95 delta"),
+      },
+      occupancy: {
+        baselineCellTotal: integer(occupancy.baseline_cell_total, "M4.8R3 baseline cells"),
+        candidateCellTotal: integer(occupancy.candidate_cell_total, "M4.8R3 candidate cells"),
+        addedCellTotal: integer(occupancy.added_cell_total, "M4.8R3 added cells"),
+        lostCellTotal: integer(occupancy.lost_cell_total, "M4.8R3 lost cells"),
+        baselineComponentTotal: integer(occupancy.baseline_component_total, "M4.8R3 baseline components"),
+        candidateComponentTotal: integer(occupancy.candidate_component_total, "M4.8R3 candidate components"),
+        meanCellGrowthFraction: numberValue(occupancy.mean_cell_growth_fraction, "M4.8R3 cell growth"),
+        meanComponentGrowthFraction: numberValue(occupancy.mean_component_growth_fraction, "M4.8R3 component growth"),
+        maximumAddedCellsPerFrame: integer(occupancy.maximum_added_cells_per_frame, "M4.8R3 maximum added cells"),
+        maximumCandidateComponentsPerFrame: integer(occupancy.maximum_candidate_components_per_frame, "M4.8R3 maximum components"),
+      },
+      provider: {
+        additiveObservationCount: integer(provider.additive_observation_count, "M4.8R3 observations"),
+        additiveVoxelCount: integer(provider.additive_voxel_count, "M4.8R3 voxels"),
+        framesWithAdditions: integer(provider.frames_with_additions, "M4.8R3 added frames"),
+        additiveMeanMsPerFrame: numberValue(provider.additive_mean_ms_per_frame, "M4.8R3 additive mean"),
+        peakTemporalComponents: integer(provider.peak_temporal_components, "M4.8R3 temporal peak"),
+        peakRollingCells: integer(provider.peak_rolling_cells, "M4.8R3 rolling peak"),
+      },
+      anchors: {
+        count: integer(anchors.count, "M4.8R3 anchor count"),
+        criticalNearCount: integer(anchors.critical_near_count, "M4.8R3 near anchors"),
+        criticalNearRecall: numberValue(anchors.critical_near_recall, "M4.8R3 near recall"),
+        matchedCount: integer(anchors.matched_count, "M4.8R3 matched anchors"),
+        canonicalEngineeringRecall: numberValue(anchors.canonical_engineering_recall, "M4.8R3 canonical recall"),
+        separation: anchors.separation.map((value, index) => {
+          const row = objectValue(value, `M4.8R3 separation ${index}`);
+          return {
+            displayFrame: integer(row.display_frame, "M4.8R3 separation frame"),
+            expectedMinimumComponents: integer(row.expected_minimum_components, "M4.8R3 expected components"),
+            observedComponents: integer(row.observed_components, "M4.8R3 observed components"),
+            passed: bool(row.passed, "M4.8R3 separation passed"),
+            interpretation: text(row.interpretation, "M4.8R3 separation interpretation"),
+          };
+        }),
+      },
+      capacityDropCount: integer(metrics.capacity_drop_count, "M4.8R3 capacity drops"),
+      falseFreeCount: integer(metrics.false_free_count, "M4.8R3 false free"),
+    },
+    gates: parsedGates,
+    decision: {
+      state,
+      candidateAccepted: bool(decision.candidate_accepted, "M4.8R3 candidate accepted"),
+      productionAccepted: false,
+      nextAction: text(decision.next_action, "M4.8R3 next action"),
+    },
+    limitations: payload.limitations.map((value, index) => text(value, `M4.8R3 limitation ${index}`)),
+    authority: authority(payload.authority),
+  };
+}
+
+export async function fetchM48R3StaticOccupancyCases(
+  resultId: string,
+  { fetcher = fetch, signal }: { fetcher?: LaboratoryFetch; signal?: AbortSignal } = {},
+): Promise<readonly M48R3StaticOccupancyCase[]> {
+  if (!RESULT_ID.test(resultId)) {
+    throw new M48R3StaticOccupancyContractError("M4.8R3 identity недопустима.");
+  }
+  const response = await fetcher(
+    `/api/v1/laboratory/m48r3/static-occupancy/${encodeURIComponent(resultId)}/cases`,
+    { method: "GET", headers: { Accept: "application/json" }, signal },
+  );
+  if (!response.ok) {
+    throw new M48R3StaticOccupancyContractError(`M4.8R3 cases недоступны: HTTP ${response.status}.`);
+  }
+  const payload = objectValue(await response.json(), "M4.8R3 cases");
+  exact(
+    payload.schema_version,
+    "missioncore.m48r3-static-occupancy-shadow-cases/v1",
+    "M4.8R3 cases schema",
+  );
+  exact(payload.result_id, resultId, "M4.8R3 cases result");
+  if (!Array.isArray(payload.cases) || payload.cases.length !== integer(payload.case_count, "M4.8R3 case count")) {
+    throw new M48R3StaticOccupancyContractError("M4.8R3 cases: нарушен размер.");
+  }
+  return payload.cases.map((value, index) => {
+    const row = objectValue(value, `M4.8R3 case ${index}`);
+    const distanceBand = text(row.distance_band, "M4.8R3 distance band");
+    if (distanceBand !== "critical-near" && distanceBand !== "approach" && distanceBand !== "outside-qualified-bands") {
+      throw new M48R3StaticOccupancyContractError("M4.8R3 distance band: неизвестное значение.");
+    }
+    return {
+      anchorId: text(row.anchor_id, "M4.8R3 anchor id"),
+      displayFrame: integer(row.display_frame, "M4.8R3 display frame"),
+      sourceSequence: integer(row.source_sequence, "M4.8R3 source sequence"),
+      extentXyxy: extent(row.extent_xyxy),
+      distanceBand,
+      componentCount: integer(row.component_count, "M4.8R3 component count"),
+      matched: bool(row.matched, "M4.8R3 matched"),
+    };
+  });
+}

apps/control-station/src/core/laboratory/m4ReplayThreat.ts

@@ -1,5 +1,6 @@
 export type M4ThreatDecision = "threat" | "not-threat" | "unknown";
 export type M4ThreatMotion = "moving" | "stationary" | "unknown";
+export type M4OccupancySource = "baseline" | "mixed" | "additive-low-step";
 export type M4Point3 = readonly [number, number, number];
 export type M4Matrix3 = readonly [M4Point3, M4Point3, M4Point3];
 
@@ -76,6 +77,7 @@ export interface M4ThreatMetricVisual {
   centroidBodyXyzM: M4Point3;
   cellCentersBodyXyzM: readonly M4Point3[];
   assessment: M4ThreatAssessment;
+  occupancySource: M4OccupancySource;
 }
 
 export interface M4ThreatCameraProposal {
@@ -174,6 +176,7 @@ export interface M4ThreatTimeline {
   cameraPointWindowSeconds: number;
   cameraPointSampleLimit: number;
   worldStateDelivery: "source-paced-latest-wins" | null;
+  occupancyProvenanceDelivery: "baseline-versus-additive-component-diff" | null;
   worldStateFrameCount: number;
   supersededFrameCount: number;
   sourceRepresentationId: "registered-map-increment-v1";
@@ -331,6 +334,9 @@ function parseMetricVisual(value: unknown): M4ThreatMetricVisual {
   ) {
     throw new M4ThreatContractError("M4.6 temporal state: неизвестное состояние.");
   }
+  const occupancySource = item.occupancy_source === undefined
+    ? "baseline"
+    : memberOccupancySource(item.occupancy_source);
   return {
     componentId: text(item.component_id, "M4.6 visual component"),
     state,
@@ -340,9 +346,17 @@ function parseMetricVisual(value: unknown): M4ThreatMetricVisual {
       (point) => vector(point, 3, "M4.6 cell") as [number, number, number],
     ),
     assessment: parseAssessment(item.assessment),
+    occupancySource,
   };
 }
 
+function memberOccupancySource(value: unknown): M4OccupancySource {
+  if (value !== "baseline" && value !== "mixed" && value !== "additive-low-step") {
+    throw new M4ThreatContractError("M4.8R3 occupancy source: неизвестное состояние.");
+  }
+  return value;
+}
+
 export async function fetchM4ThreatReplayResult({
   resultId: requestedResultId,
   fetcher = fetch,
@@ -683,6 +697,13 @@ export async function fetchM4ThreatTimeline(
           "source-paced-latest-wins",
           "M4.6 world-state delivery",
         ),
+    occupancyProvenanceDelivery: payload.occupancy_provenance_delivery == null
+      ? null
+      : exact(
+          payload.occupancy_provenance_delivery,
+          "baseline-versus-additive-component-diff",
+          "M4.8R3 occupancy provenance",
+        ),
     worldStateFrameCount: payload.world_state_frame_count === undefined
       ? frameCount
       : integer(payload.world_state_frame_count, "M4.6 world-state frames"),

apps/control-station/src/styles/m4-replay-threat.css

@@ -334,6 +334,12 @@
   background: transparent;
 }
 
+.laboratory-metric-evidence-scene__legend span[data-decision="low-step"]::before {
+  box-sizing: border-box;
+  border: 1px solid rgb(var(--nodedc-foreground-rgb));
+  background: rgb(var(--nodedc-accent-rgb));
+}
+
 .laboratory-metric-evidence-scene__legend span[data-decision="local-surface"]::before {
   background: rgb(var(--nodedc-accent-rgb));
   opacity: 0.72;

apps/control-station/src/styles/simulation.css

@@ -332,6 +332,7 @@
 }
 
 .simulation-viewport {
+  container-type: inline-size;
   display: grid;
   min-height: 0;
   grid-template-rows: auto minmax(0, 1fr) auto;
@@ -341,7 +342,9 @@
 }
 
 .simulation-viewport__toolbar {
-  flex-wrap: wrap;
+  display: grid;
+  grid-template-columns: minmax(12rem, 1fr) auto minmax(12rem, 1fr);
+  align-items: center;
   gap: 0.55rem;
   border-bottom: 1px solid var(--station-hairline);
   background: var(--nodedc-glass-panel-bg-soft);
@@ -356,6 +359,74 @@
   gap: 0.45rem;
 }
 
+.simulation-viewport__controls {
+  display: flex;
+  min-width: 0;
+  align-items: center;
+  justify-content: center;
+  gap: 0.55rem;
+}
+
+.simulation-viewport__settings-anchor {
+  position: relative;
+  display: grid;
+  place-items: center;
+}
+
+.simulation-viewport__settings {
+  position: absolute;
+  z-index: 12;
+  top: calc(100% + 0.65rem);
+  right: 0;
+  display: grid;
+  width: min(19rem, calc(100cqw - 1.3rem));
+  gap: 0.75rem;
+  border-radius: 0.9rem;
+  background: rgb(25 27 31 / 0.98);
+  box-shadow: 0 1rem 3rem rgb(0 0 0 / 0.42);
+  padding: 0.8rem;
+}
+
+.simulation-viewport__settings-head {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  gap: 0.75rem;
+}
+
+.simulation-viewport__settings-head > div,
+.simulation-viewport__settings-switches {
+  display: grid;
+  gap: 0.32rem;
+}
+
+.simulation-viewport__settings strong {
+  color: var(--nodedc-text-primary);
+  font-size: 0.72rem;
+}
+
+.simulation-viewport__settings span,
+.simulation-viewport__settings p {
+  color: var(--nodedc-text-muted);
+  font-size: 0.56rem;
+  line-height: 1.45;
+}
+
+.simulation-viewport__settings p {
+  margin: 0;
+}
+
+.simulation-viewport__settings-switches {
+  gap: 0.55rem;
+  border-radius: 0.72rem;
+  background: rgb(255 255 255 / 0.035);
+  padding: 0.65rem;
+}
+
+.simulation-viewport__toolbar-balance {
+  min-width: 0;
+}
+
 .simulation-viewport__toolbar > div:first-child > span:last-child,
 .simulation-viewport__notice > span:last-child {
   color: var(--nodedc-text-muted);
@@ -364,14 +435,24 @@
 
 .simulation-viewport__stage {
   position: relative;
+  min-width: 0;
   min-height: 30rem;
+  overflow: hidden;
 }
 
 .simulation-viewport__stage canvas {
   display: block;
+  max-width: 100%;
   width: 100%;
   height: 100%;
   outline: none;
+  cursor: grab;
+  touch-action: none;
+  user-select: none;
+}
+
+.simulation-viewport__stage canvas:active {
+  cursor: grabbing;
 }
 
 .simulation-viewport__stage canvas:focus-visible {
@@ -407,6 +488,21 @@
   padding: 0.48rem 0.65rem;
 }
 
+@container (max-width: 68rem) {
+  .simulation-viewport__toolbar {
+    grid-template-columns: 1fr;
+  }
+
+  .simulation-viewport__controls {
+    flex-wrap: wrap;
+    justify-content: center;
+  }
+
+  .simulation-viewport__toolbar-balance {
+    display: none;
+  }
+}
+
 .simulation-workspace__processing {
   display: flex;
   min-height: 18rem;

apps/control-station/src/workspaces/laboratory/AdvancedLaboratoryResult.tsx

@@ -45,6 +45,7 @@ import { M47ReferenceGraphResultView } from "./M47ReferenceGraphResult";
 import { M48ObjectCentricQualityResultView } from "./M48ObjectCentricQualityResult";
 import { M48SmallStaticPassageRegressionResultView } from "./M48SmallStaticPassageRegressionResult";
 import { M48StaticOccupancyQualificationResultView } from "./M48StaticOccupancyQualificationResult";
+import { M48R3StaticOccupancyShadowResultView } from "./M48R3StaticOccupancyShadowResult";
 import { M48SFixedClassDetectorResultView } from "./M48SFixedClassDetectorResult";
 import { M48TRiskQualityResultView } from "./M48TRiskQualityResult";
 
@@ -98,6 +99,9 @@ export function AdvancedLaboratoryResult({
   if (workId === "m48-static-occupancy-qualification" && results.m48StaticOccupancy) {
     return <M48StaticOccupancyQualificationResultView rigLabel={rigLabel} result={results.m48StaticOccupancy} />;
   }
+  if (workId === "m48r3-static-occupancy-shadow" && results.m48r3StaticOccupancy) {
+    return <M48R3StaticOccupancyShadowResultView rigLabel={rigLabel} result={results.m48r3StaticOccupancy} />;
+  }
   if (workId === "m48s-fixed-class-detector" && results.m48s) {
     return <M48SFixedClassDetectorResultView rigLabel={rigLabel} result={results.m48s} />;
   }

apps/control-station/src/workspaces/laboratory/M48R3StaticOccupancyShadowEvidence.tsx

@@ -0,0 +1,81 @@
+import { useEffect, useMemo, useState } from "react";
+import { Icon } from "@nodedc/ui-react";
+
+import {
+  fetchM48R3StaticOccupancyCases,
+  type M48R3StaticOccupancyCase,
+  type M48R3StaticOccupancyShadowResult,
+} from "../../core/laboratory/m48r3StaticOccupancyShadow";
+import {
+  M4ReplayThreatVisual,
+  type M4ReplayThreatReviewAnchor,
+} from "./M4ReplayThreatVisual";
+
+function message(error: unknown): string {
+  return error instanceof Error && error.message.trim()
+    ? error.message
+    : "M4.8R3 timeline недоступен.";
+}
+
+export function M48R3StaticOccupancyShadowEvidence({
+  result,
+}: {
+  result: M48R3StaticOccupancyShadowResult;
+}) {
+  const [cases, setCases] = useState<readonly M48R3StaticOccupancyCase[]>([]);
+  const [loading, setLoading] = useState(true);
+  const [error, setError] = useState<string | null>(null);
+
+  useEffect(() => {
+    const controller = new AbortController();
+    setLoading(true);
+    setError(null);
+    void fetchM48R3StaticOccupancyCases(result.resultId, { signal: controller.signal })
+      .then((nextCases) => {
+        if (!controller.signal.aborted) setCases(nextCases);
+      })
+      .catch((caught: unknown) => {
+        if (!controller.signal.aborted) setError(message(caught));
+      })
+      .finally(() => {
+        if (!controller.signal.aborted) setLoading(false);
+      });
+    return () => controller.abort();
+  }, [result.resultId]);
+
+  const reviewAnchors = useMemo<readonly M4ReplayThreatReviewAnchor[]>(() => (
+    cases.map((item) => ({
+      id: item.anchorId,
+      sourceSequence: item.sourceSequence,
+      extentXyxyNormalized: item.extentXyxy,
+      matchedAtThreshold: item.matched,
+    }))
+  ), [cases]);
+
+  if (loading) {
+    return (
+      <div className="l3-visual-audit__state" role="status">
+        <span className="busy-indicator" aria-hidden="true" />
+        <span>Открываем полный M4.8R3 Worker timeline</span>
+      </div>
+    );
+  }
+  if (error) {
+    return (
+      <div className="l3-visual-audit__state" role="alert">
+        <Icon name="alert" size={18} />
+        <span>{error}</span>
+      </div>
+    );
+  }
+  return (
+    <M4ReplayThreatVisual
+      resultId={result.resultId}
+      reviewAnchors={reviewAnchors}
+      showReviewAnchorBoxes={false}
+      reviewLabel="Контрольные кадры M4.8R3 · без ручных рамок"
+      timelineEndpointRoot="/api/v1/laboratory/m48r3/static-occupancy"
+      evidenceLabel="M4.8R3"
+    />
+  );
+}

apps/control-station/src/workspaces/laboratory/M48R3StaticOccupancyShadowResult.tsx

@@ -0,0 +1,101 @@
+import {
+  LaboratoryEvidence,
+  LaboratoryResultSummary,
+  LaboratorySummary,
+  LaboratoryWorkTemplate,
+} from "../../components/laboratory/LaboratoryPresentation";
+import type { M48R3StaticOccupancyShadowResult } from "../../core/laboratory/m48r3StaticOccupancyShadow";
+import { M48R3StaticOccupancyShadowEvidence } from "./M48R3StaticOccupancyShadowEvidence";
+
+function number(value: number, digits = 2): string {
+  return value.toLocaleString("ru-RU", { maximumFractionDigits: digits });
+}
+
+function percent(value: number): string {
+  return `${number(value * 100, 1)}%`;
+}
+
+export function M48R3StaticOccupancyShadowResultView({
+  rigLabel,
+  result,
+}: {
+  rigLabel: string;
+  result: M48R3StaticOccupancyShadowResult;
+}) {
+  const candidate = result.metrics.performance.candidate;
+  const occupancy = result.metrics.occupancy;
+  const separation = result.metrics.anchors.separation;
+  const separated = separation.every((item) => item.passed);
+  const status = result.accepted
+    ? "Полный Worker shadow принят: realtime и раздельные препятствия сохранены"
+    : "Worker shadow не прошёл один или несколько предобъявленных gate";
+  const separatedLabel = separation.length
+    ? separation.map((item) => `${item.observedComponents}/${item.expectedMinimumComponents}`).join(" · ")
+    : "—";
+  return (
+    <LaboratoryWorkTemplate
+      summary={(
+        <LaboratorySummary
+          title="M4.8R3 · full Worker static occupancy shadow"
+          description="Полный 4 489-кадровый прогон проверяет additive low-step occupied-only слой внутри штатного graph pipeline. Камера и RF-DETR не получают дополнительного inference; ручные прямоугольники используются только для перехода к контрольным кадрам и не рисуются как системный результат."
+          status={status}
+          statusTone={result.accepted ? "success" : "warning"}
+          facts={[
+            { label: "Конфигурация", value: `${rigLabel} RIGHT · VIDEO/CAMERA/3D/PLAN · LOW-STEP provenance` },
+            { label: "Профиль", value: `${result.profile.id} · minimum points ${result.profile.minimumPoints}` },
+            { label: "Прогон", value: `${result.metrics.frames.candidateDelivered}/${result.metrics.frames.expected} · immutable ${result.resultId}` },
+            { label: "Нагрузка", value: `12 Hz source-paced · ${number(candidate.fps, 3)} effective FPS · +0 inference` },
+            { label: "Authority", value: "REPLAY-SIMULATED · production/navigation/actuation OFF" },
+          ]}
+          brief={{
+            question: "Можно ли добавить геометрическое обнаружение низких статических препятствий, не разрушив realtime и не склеив отдельные столбики/шары в один объект?",
+            approach: `Кандидат сравнен покадрово с native baseline на всех ${result.metrics.frames.expected} кадрах. Проверены latency, FPS, рост occupancy, capacity drops, отсутствие потерянных baseline-ячеек и отдельные компоненты на кадре 1856.`,
+            principalResult: `${number(candidate.fps, 3)} FPS; world-state p95 ${number(candidate.worldP95Ms, 2)} мс; geometry p95/p99 ${number(candidate.geometryP95Ms, 2)}/${number(candidate.geometryP99Ms, 2)} мс; разделение ${separatedLabel}.`,
+            limitation: "Это воспроизводимый Worker shadow, а не доказательство физической проходимости. Просвет между компонентами сохраняется как геометрия, но допустимость проезда зависит от будущего габарита шасси и отдельного free-space контракта.",
+          }}
+          method={{
+            completeness: "complete",
+            executionClass: "deterministic",
+            pipelineId: "m48r3-native-plus-low-step-reference-graph/v1",
+            components: [
+              { kind: "source", name: "native reference graph baseline", version: "M4.7/M4.8R2", role: "immutable occupied/unknown baseline", identitySha256: null },
+              { kind: "algorithm", name: "additive low-step occupied-only", version: "v1", role: "CPU geometry; never clearing; no semantic class", identitySha256: result.profile.sha256 },
+              { kind: "runtime", name: "full source-paced Worker shadow", version: "4 489 frames", role: "predeclared realtime, growth, separation and safety gates", identitySha256: result.resultId.split("-").at(-1) ?? null },
+            ],
+          }}
+        />
+      )}
+      evidence={(
+        <LaboratoryEvidence
+          eyebrow="M4.8R3 VISUAL EVIDENCE · SYSTEM COMPONENTS"
+          title="Полный timeline; LOW-STEP показывает добавленные системой компоненты, ручные рамки скрыты"
+          kind="recorded-replay"
+          resizable
+        >
+          <M48R3StaticOccupancyShadowEvidence result={result} />
+        </LaboratoryEvidence>
+      )}
+      result={(
+        <LaboratoryResultSummary
+          title="Что доказал прогон"
+          status={status}
+          statusTone={result.accepted ? "success" : "warning"}
+          metrics={[
+            { label: "Realtime", value: `${number(candidate.fps, 3)} FPS`, hint: `p95 ${number(candidate.worldP95Ms, 2)} мс · gate ≥11,5 FPS / ≤60 мс` },
+            { label: "Geometry", value: `${number(candidate.geometryP95Ms, 2)} / ${number(candidate.geometryP99Ms, 2)} мс`, hint: "p95 / p99 · gates 9 / 16 мс" },
+            { label: "Occupancy delta", value: `+${occupancy.addedCellTotal.toLocaleString("ru-RU")}`, hint: `${percent(occupancy.meanCellGrowthFraction)} cells · ${percent(occupancy.meanComponentGrowthFraction)} components` },
+            { label: "Кадр 1856", value: separated ? `раздельно · ${separatedLabel}` : `не принят · ${separatedLabel}`, hint: "два столбика и две полусферы проверяются отдельными component gates" },
+            { label: "Потери / false free", value: `${occupancy.lostCellTotal} / ${result.metrics.falseFreeCount}`, hint: `capacity drops ${result.metrics.capacityDropCount}` },
+          ]}
+          conclusion={{
+            proved: `Система сама добавила ${result.metrics.provider.additiveObservationCount.toLocaleString("ru-RU")} low-step observations на ${result.metrics.provider.framesWithAdditions.toLocaleString("ru-RU")} кадрах, сохранила baseline без потерь и выдержала полный realtime shadow.`,
+            notProved: "Не доказаны физический clearance, planner-authoritative free space, независимые precision/recall и безопасность движения на реальном шасси.",
+            decision: result.accepted
+              ? "Worker-кандидат принят в ограниченном replay-shadow контуре. Следующий шаг — отдельное решение о cutover и регрессия на новых сценах; ручная разметка не становится runtime-зависимостью."
+              : "Cutover запрещён. Исправить провалившийся gate и повторить полный immutable shadow без ослабления порогов.",
+          }}
+        />
+      )}
+    />
+  );
+}

apps/control-station/src/workspaces/laboratory/M4ReplayThreatVisual.tsx

@@ -110,12 +110,16 @@ export function M4ReplayThreatVisual({
   resultId,
   semantic,
   reviewAnchors = EMPTY_REVIEW_ANCHORS,
+  showReviewAnchorBoxes = true,
+  reviewLabel = "Контрольные примеры M4.8R1",
   timelineEndpointRoot,
   evidenceLabel = "M4.6",
 }: {
   resultId: string;
   semantic?: M4ReplayThreatSemanticLayer;
   reviewAnchors?: readonly M4ReplayThreatReviewAnchor[];
+  showReviewAnchorBoxes?: boolean;
+  reviewLabel?: string;
   timelineEndpointRoot?: string;
   evidenceLabel?: string;
 }) {
@@ -124,6 +128,7 @@ export function M4ReplayThreatVisual({
   const [showCurrentIncrement, setShowCurrentIncrement] = useState(true);
   const [showLocalSurface, setShowLocalSurface] = useState(true);
   const [showRollingMap, setShowRollingMap] = useState(true);
+  const [showLowStep, setShowLowStep] = useState(true);
   const [showMediaSemantic, setShowMediaSemantic] = useState(true);
   const [showSpatialSemantic, setShowSpatialSemantic] = useState(true);
   const [showMediaPoints, setShowMediaPoints] = useState(false);
@@ -262,7 +267,7 @@ export function M4ReplayThreatVisual({
   }, [metadata.timeline, resultId, reviewAnchorIdentity, reviewAnchors, seekPlayback, setPlaybackPlaying]);
   const reviewAnchorBoxes = useMemo<readonly RecordedEvidenceBox[]>(() => {
     const timeline = metadata.timeline;
-    if (!frame || !timeline) return [];
+    if (!frame || !timeline || !showReviewAnchorBoxes) return [];
     return reviewAnchors
       .filter((anchor) => anchor.sourceSequence === frame.sequence)
       .map((anchor) => {
@@ -281,7 +286,7 @@ export function M4ReplayThreatVisual({
           dashed: true,
         };
       });
-  }, [frame, metadata.timeline, reviewAnchors]);
+  }, [frame, metadata.timeline, reviewAnchors, showReviewAnchorBoxes]);
   const activeBoxes = useMemo(
     () => [...boxes(frame?.cameraProposals ?? []), ...reviewAnchorBoxes],
     [frame, reviewAnchorBoxes],
@@ -345,6 +350,7 @@ export function M4ReplayThreatVisual({
     state: obstacle.state,
     centroidBodyXyzM: obstacle.centroidBodyXyzM,
     cellCentersBodyXyzM: obstacle.cellCentersBodyXyzM,
+    occupancySource: obstacle.occupancySource,
   })) ?? [], [spatialFrame]);
   const currentIncrementObstacles = spatialFrame?.metricObstacles.filter(
     (item) => item.state === "current",
@@ -352,6 +358,9 @@ export function M4ReplayThreatVisual({
   const rollingMapObstacles = spatialFrame?.metricObstacles.filter(
     (item) => item.state === "retained",
   ) ?? [];
+  const lowStepObstacles = spatialFrame?.metricObstacles.filter(
+    (item) => item.occupancySource !== "baseline",
+  ) ?? [];
   const nearest = spatialFrame?.metricObstacles
     .map((item) => item.assessment.closestApproachM)
     .filter((value): value is number => value !== null)
@@ -513,6 +522,18 @@ export function M4ReplayThreatVisual({
       >
         ROLLING
       </Button>
+      {metadata.timeline?.occupancyProvenanceDelivery ? (
+        <Button
+          size="compact"
+          shape="pill"
+          variant={showLowStep ? "primary" : "secondary"}
+          aria-pressed={showLowStep}
+          title="Добавочные occupied-only компоненты low-step; без ручных рамок"
+          onClick={() => setShowLowStep((visible) => !visible)}
+        >
+          LOW-STEP
+        </Button>
+      ) : null}
       {semantic ? (
         <Button
           size="compact"
@@ -567,7 +588,7 @@ export function M4ReplayThreatVisual({
         <Icon name="chevron-right" size={16} />
       </IconButton>
       <Select
-        label="Контрольные примеры M4.8R1"
+        label={reviewLabel}
         value={String(selectedReviewAnchorIndex)}
         options={reviewAnchors.map((anchor, index) => ({
           value: String(index),
@@ -618,6 +639,9 @@ export function M4ReplayThreatVisual({
         <span>Spatial evidence</span>
         <strong>
           {currentIncrementObstacles.length} current · {rollingMapObstacles.length} rolling
+          {metadata.timeline.occupancyProvenanceDelivery
+            ? ` · ${lowStepObstacles.length} low-step`
+            : ""}
         </strong>
         <small>
           {spatialFrame
@@ -758,6 +782,7 @@ export function M4ReplayThreatVisual({
             showCurrentIncrement={showCurrentIncrement}
             showLocalSurface={showLocalSurface}
             showRollingMap={showRollingMap}
+            showLowStep={showLowStep}
             pointSemanticClassIds={alignedSemanticPointIds}
             semanticClasses={semanticClasses}
             semanticPalette={semanticPalette}

apps/control-station/src/workspaces/laboratory/laboratoryArchiveProfiles.ts

@@ -84,6 +84,13 @@ const KNOWN_WORKS: Readonly<Record<Exclude<LaboratoryWorkId, `session:${string}`
     experimentName: "M4.8 · conservative static occupancy qualification",
     variantName: "M4.8R2 · current/rolling + low-step occupied-only candidate",
   },
+  "m48r3-static-occupancy-shadow": {
+    profileId: "rig-dual-evidence-virtual-corridor-v1",
+    profileName: (rigLabel) => `${rig(rigLabel)} RIGHT · Canonical Reference Graph`,
+    experimentId: "m48r3-static-occupancy-shadow",
+    experimentName: "M4.8R3 · full Worker static occupancy shadow",
+    variantName: "M4.8R3 · 4 489 frames · additive low-step occupied-only",
+  },
   "m48s-fixed-class-detector": {
     profileId: "rig-ravnoves-perception-gate-v1",
     profileName: (rigLabel) => `${rig(rigLabel)} RIGHT · RAVNOVES00 perception gate`,

apps/control-station/src/workspaces/laboratory/useAdvancedLaboratoryCatalog.ts

@@ -22,6 +22,7 @@ function mergeResults(
     m48: next.m48 ?? current.m48,
     m48SmallStatic: next.m48SmallStatic ?? current.m48SmallStatic,
     m48StaticOccupancy: next.m48StaticOccupancy ?? current.m48StaticOccupancy,
+    m48r3StaticOccupancy: next.m48r3StaticOccupancy ?? current.m48r3StaticOccupancy,
     m48s: next.m48s ?? current.m48s,
     m48t: next.m48t ?? current.m48t,
     m4Threat: next.m4Threat ?? current.m4Threat,
@@ -122,6 +123,7 @@ export function useAdvancedLaboratoryCatalog({
         "m48-object-centric-quality",
         "m48-small-static-passage-regression",
         "m48-static-occupancy-qualification",
+        "m48r3-static-occupancy-shadow",
       ].includes(selectedWorkId)
       && !indexedResultId
     ) return;

apps/control-station/test/simulationWorkspace.test.mjs

@@ -63,11 +63,29 @@ test("PlayCanvas owns the realtime scene graph without an iframe or React entity
   assert.match(packageDocument, /"playcanvas": "2\.21\.4"/);
   assert.doesNotMatch(packageDocument, /@playcanvas\/react/);
   assert.match(runtime, /new Application\(canvas/);
+  assert.match(runtime, /setCanvasFillMode\(FILLMODE_NONE/);
+  assert.match(runtime, /setCanvasResolution\(RESOLUTION_AUTO\)/);
   assert.match(runtime, /new Asset\([^,]+, "gsplat"/);
   assert.match(runtime, /camera\.script\?\.create\(CameraControls/);
+  assert.match(runtime, /const HOME_POSITION = new Vec3\(0, 1, 0\)/);
+  assert.match(runtime, /const HOME_FOCUS = new Vec3\(1, 1, 0\)/);
+  assert.match(runtime, /maximum: \{[^}]*lodRangeMin: 0, lodRangeMax: 0, pixelRatio: 2/);
+  assert.match(runtime, /frame\.mouse\[0\] \*= -1/);
+  assert.match(runtime, /new Asset\([\s\S]*"container"/);
+  assert.match(runtime, /instantiateRenderEntity/);
+  assert.match(runtime, /applyWorldTransform/);
+  assert.match(runtime, /setLayerWorldTransform/);
+  assert.match(runtime, /PLAYCANVAS_X_180_TRANSFORM/);
+  assert.match(runtime, /ensureCollision/);
   assert.match(runtime, /app\.root\.addChild/);
   assert.match(runtime, /dispose\(\)/);
   assert.doesNotMatch(`${runtime}\n${viewport}`, /iframe|<GSplat/);
   assert.match(viewport, /Collision недоступен/);
   assert.match(viewport, /Для этой сборки collision GLB не был запрошен/);
+  assert.match(viewport, /<Select[\s\S]*Качество Gaussian-сцены/);
+  assert.match(viewport, /Максимум/);
+  assert.match(viewport, /runtimeRef\.current\?\.home\(\)/);
+  assert.match(viewport, /Настройки системы координат/);
+  assert.match(viewport, /Инверсия визуального слоя/);
+  assert.match(viewport, /Инверсия collision-слоя/);
 });

config/laboratories/m48r3-static-occupancy-shadow.json

@@ -0,0 +1,10 @@
+{
+  "schema_version": "missioncore.laboratory-evidence-definition/v1",
+  "work_id": "m48r3-static-occupancy-shadow",
+  "evidence": {
+    "runtime_relative_root": "m48/static-occupancy-shadow-results",
+    "result_id_prefix": "m48r3-static-occupancy-shadow",
+    "document_name": "manifest.json",
+    "schema_version": "missioncore.m48r3-static-occupancy-shadow-result/v1"
+  }
+}

config/laboratory-execution.json

@@ -165,6 +165,7 @@
     }
   ],
   "legacy_work_ids": [
+    "m48r3-static-occupancy-shadow",
     "m47-reference-graph-shadow",
     "e31-source-binding",
     "e32-track-geometry",

config/perception/m48r3-additive-low-step-occupancy-v1.json

@@ -20,6 +20,10 @@
     "voxel_size_m": 0.45,
     "neighbor_radius_cells": 1,
     "minimum_points": 5,
+    "sparse_persistence_minimum_points": 2,
+    "sparse_persistence_window_frames": 6,
+    "sparse_persistence_minimum_hits": 6,
+    "sparse_persistence_maximum_range_m": 8.0,
     "minimum_voxels": 1,
     "local_radius_m": 10.0,
     "maximum_candidate_points_per_frame": 768,

config/perception/m48r3-native-low-step-reference-graph-shadow-v1.json

@@ -22,7 +22,7 @@
       "provider_id": "ravnoves00-additive-low-step-geometry/v1",
       "version": "0.1.0",
       "revision": "m48r3-ravnoves00-additive-low-step/v1",
-      "sha256": "0d46dcb28542902849ee31de4c4594ef90407834d9e5551474fc4191cadbf1a9"
+      "sha256": "00fc197ee7200e2f0447b6bec7cc2f21b3c2970d7dd9414cc41e6787d712fee5"
     },
     {
       "role": "temporal",

docs/22_MILESTONE_4_OBJECT_CENTRIC_RECORDED_REALTIME_CV.md

@@ -1504,6 +1504,42 @@ anchors onto its immutable timeline; no third perception instrument exists.
 Navigation, commands, actuation, physical clearance and collision-safety
 authority remain false.
 
+### 2026-08-26 — M4.8R3 bounded static-occupancy Worker shadow accepted
+
+M4.8R3 closes the bounded Worker integration requested by M4.8R2. The sealed
+result is
+`m48r3-static-occupancy-shadow-d1577870b098bcd0b67cc51798234f224b348ad3954ead72f3ce6df414875a29`.
+It composes the accepted camera/geometry graph with an occupied-only low-step
+provider; RF-DETR, the native `800×600` fisheye raster and the semantic risk
+policy remain unchanged. Weak two-to-four-point geometry is published only
+when the exact `0.45 m` voxel is present in all six causal frames and its
+current range is at most `8 m`. Strong five-point components remain immediate.
+The provider never uses a semantic class, clears a cell, infers free space or
+runs another neural-network pass.
+
+The isolated Worker 006 replay delivered all `4,489/4,489` frames at
+`11.79902 FPS`. World-state completion p95 was `56.74324 ms`; geometry p95/p99
+were `8.420913/13.732692 ms`; GPU utilization p95/maximum were `54%/58%` and
+maximum used GPU memory was `9,743 MiB`. The additive provider contributed a
+mean `0.587893 ms` per frame. Relative to the frozen native baseline, FPS
+regression was `0.3745552%` and world-state p95 increased by `8.802461 ms`,
+both inside the predeclared envelope.
+
+All `9/9` critical-near assisted anchors are occupied in the delivered graph.
+At display frame `1856`, the two thin posts remain independent (`5` observed
+components in their review extent) and the two concrete hemispheres remain
+independent (`6` observed components); the system does not inherit the
+operator's old paired boxes. The ledger adds `613,146` occupied cells, loses
+zero baseline cells, makes zero false-free claims and records zero capacity
+drops. The two `8–12 m` approach anchors intentionally remain unknown because
+the sparse-persistence policy is bounded to the critical `0–8 m` band.
+
+This accepts the reproducible bounded Worker shadow and its LAB evidence, not a
+production navigation cutover. The result remains replay-simulated;
+physical-live authority, commands, actuation, planner-authoritative free space
+and collision-safety acceptance are false. The next milestone boundary is
+M4.9 recorded-realtime release-candidate validation using this frozen graph.
+
 ## Implementation order
 
 The implementation sequence is intentionally strict:

experiments/perception/seal_m48r3_static_occupancy_shadow.py

@@ -0,0 +1,57 @@
+#!/usr/bin/env python3
+"""Seal a complete M4.8R3 Worker shadow against the accepted native baseline."""
+
+from __future__ import annotations
+
+import argparse
+import json
+from pathlib import Path
+
+from k1link.laboratory.m48r3_static_occupancy_shadow import (
+    M48R3StaticOccupancyShadowError,
+    build_m48r3_static_occupancy_shadow,
+)
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--repository-root", type=Path, required=True)
+    parser.add_argument("--profile", type=Path, required=True)
+    parser.add_argument("--baseline-result", type=Path, required=True)
+    parser.add_argument("--baseline-frames", type=Path, required=True)
+    parser.add_argument("--candidate-result", type=Path, required=True)
+    parser.add_argument("--candidate-frames", type=Path, required=True)
+    parser.add_argument("--m48r2-result-root", type=Path, required=True)
+    parser.add_argument("--output-root", type=Path, required=True)
+    arguments = parser.parse_args()
+    try:
+        result = build_m48r3_static_occupancy_shadow(
+            repository_root=arguments.repository_root,
+            profile_path=arguments.profile,
+            baseline_result_path=arguments.baseline_result,
+            baseline_frames_path=arguments.baseline_frames,
+            candidate_result_path=arguments.candidate_result,
+            candidate_frames_path=arguments.candidate_frames,
+            m48r2_result_root=arguments.m48r2_result_root,
+            output_root=arguments.output_root,
+        )
+    except (M48R3StaticOccupancyShadowError, OSError, ValueError) as exc:
+        parser.error(str(exc))
+    print(
+        json.dumps(
+            {
+                "result_id": result.result_id,
+                "result_root": str(result.result_root),
+                "accepted": result.manifest["accepted"],
+                "gates": result.report["gates"],
+            },
+            ensure_ascii=False,
+            indent=2,
+            sort_keys=True,
+        )
+    )
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

experiments/perception/worker/Invoke-M48NNativeReferenceGraph.ps1

@@ -113,10 +113,10 @@ $expectedConfigs = [ordered]@{
 if ($AdditiveLowStep) {
     $expectedConfigs.Remove("m48n-rf-detr-native-reference-graph-shadow-v0.json")
     $expectedConfigs["m48r3-native-low-step-reference-graph-shadow-v1.json"] = (
-        "5f5832a0a0c1879374b166071efed02de14aeda815046da5711d6f4053eb9af4"
+        "e2e307347265076908c4ec1b7da75028998490dfa74ba64f12939e233406d9d0"
     )
     $expectedConfigs["m48r3-additive-low-step-occupancy-v1.json"] = (
-        "0d46dcb28542902849ee31de4c4594ef90407834d9e5551474fc4191cadbf1a9"
+        "00fc197ee7200e2f0447b6bec7cc2f21b3c2970d7dd9414cc41e6787d712fee5"
     )
 }
 foreach ($entry in $expectedConfigs.GetEnumerator()) {

src/k1link/laboratory/m48r3_static_occupancy_shadow.py

@@ -0,0 +1,845 @@
+"""Seal the full M4.8R3 occupied-only Worker shadow and its bounded diff."""
+
+from __future__ import annotations
+
+import hashlib
+import json
+import math
+import os
+import shutil
+import uuid
+from collections.abc import Iterable
+from dataclasses import dataclass
+from datetime import UTC, datetime
+from pathlib import Path
+from typing import Any, Final
+
+import numpy as np
+
+from k1link.laboratory.m48_static_occupancy_qualification import (
+    M48StaticOccupancyQualificationError,
+    read_m48_static_occupancy_qualification,
+)
+from k1link.perception.geometry import RecordedGeometryStore
+from k1link.perception.geometry_math import project_map_points_kb4
+from k1link.perception.m48_low_step_occupancy import M48_LOW_STEP_PROFILE_SCHEMA
+
+M48R3_SHADOW_RESULT_SCHEMA: Final = (
+    "missioncore.m48r3-static-occupancy-shadow-result/v1"
+)
+M48R3_SHADOW_REPORT_SCHEMA: Final = (
+    "missioncore.m48r3-static-occupancy-shadow-report/v1"
+)
+M48R3_SHADOW_CASE_SCHEMA: Final = (
+    "missioncore.m48r3-static-occupancy-shadow-case/v1"
+)
+M48R3_SHADOW_DIFF_SCHEMA: Final = (
+    "missioncore.m48r3-static-occupancy-frame-diff/v1"
+)
+M48R3_SHADOW_PREFIX: Final = "m48r3-static-occupancy-shadow-"
+FRAME_EVIDENCE_SCHEMA: Final = "missioncore.m48s-reference-graph-frame-evidence/v1"
+WORKER_RESULT_SCHEMA: Final = "missioncore.m48s-reference-graph-shadow-load/v5"
+EXPECTED_FRAMES: Final = 4_489
+VOXEL_SIZE_M: Final = 0.45
+_AUTHORITY: Final = {
+    "mode": "replay-simulated",
+    "physical_live": False,
+    "commands_enabled": False,
+    "actuation_allowed": False,
+    "navigation_or_safety_accepted": False,
+}
+
+Cell = tuple[int, int, int]
+
+
+class M48R3StaticOccupancyShadowError(RuntimeError):
+    """The M4.8R3 Worker evidence or immutable result is invalid."""
+
+
+@dataclass(frozen=True, slots=True)
+class M48R3StaticOccupancyShadowResult:
+    result_id: str
+    result_root: Path
+    manifest: dict[str, Any]
+    report: dict[str, Any]
+    cases: tuple[dict[str, Any], ...]
+
+
+@dataclass(frozen=True, slots=True)
+class LedgerComparison:
+    frame_count: int
+    baseline_cell_total: int
+    candidate_cell_total: int
+    added_cell_total: int
+    lost_cell_total: int
+    baseline_component_total: int
+    candidate_component_total: int
+    false_free_count: int
+    maximum_added_cells_per_frame: int
+    maximum_candidate_components_per_frame: int
+    mean_cell_growth_fraction: float
+    mean_component_growth_fraction: float
+    diff_rows: tuple[dict[str, Any], ...]
+    selected_candidate_rows: dict[int, dict[str, Any]]
+
+
+def build_m48r3_static_occupancy_shadow(
+    *,
+    repository_root: Path,
+    profile_path: Path,
+    baseline_result_path: Path,
+    baseline_frames_path: Path,
+    candidate_result_path: Path,
+    candidate_frames_path: Path,
+    m48r2_result_root: Path,
+    output_root: Path,
+) -> M48R3StaticOccupancyShadowResult:
+    """Compare full ledgers and publish one append-only M4.8R3 result."""
+
+    repository = repository_root.resolve(strict=True)
+    profile_bytes = profile_path.resolve(strict=True).read_bytes()
+    profile = _object(json.loads(profile_bytes), "M4.8R3 profile")
+    if profile.get("schema_version") != M48_LOW_STEP_PROFILE_SCHEMA:
+        raise M48R3StaticOccupancyShadowError("M4.8R3 profile changed")
+    profile_sha256 = hashlib.sha256(profile_bytes).hexdigest()
+    acceptance = _object(profile.get("acceptance"), "M4.8R3 acceptance")
+    source = _object(profile.get("source"), "M4.8R3 source")
+    if source.get("frame_count") != EXPECTED_FRAMES:
+        raise M48R3StaticOccupancyShadowError("M4.8R3 frame contract changed")
+
+    baseline = _read_worker_result(baseline_result_path)
+    candidate = _read_worker_result(candidate_result_path)
+    _validate_worker_binding(
+        baseline,
+        baseline_frames_path,
+        expected_frames=EXPECTED_FRAMES,
+        additive_profile_sha256=None,
+    )
+    _validate_worker_binding(
+        candidate,
+        candidate_frames_path,
+        expected_frames=EXPECTED_FRAMES,
+        additive_profile_sha256=profile_sha256,
+    )
+    try:
+        m48r2 = read_m48_static_occupancy_qualification(m48r2_result_root)
+    except M48StaticOccupancyQualificationError as exc:
+        raise M48R3StaticOccupancyShadowError("M4.8R2 binding changed") from exc
+    if (
+        m48r2.result_id != source.get("m48r2_result_id")
+        or _file_sha256(m48r2.result_root / "cases.jsonl")
+        != source.get("m48r2_cases_sha256")
+    ):
+        raise M48R3StaticOccupancyShadowError("M4.8R2 source changed")
+
+    selected_sequences = {int(row["sequence"]) - 1 for row in m48r2.cases}
+    comparison = compare_m48r3_frame_ledgers(
+        baseline_frames_path,
+        candidate_frames_path,
+        selected_sequences=selected_sequences,
+        expected_frames=EXPECTED_FRAMES,
+    )
+    store = RecordedGeometryStore.from_repository(repository)
+    cases = _evaluate_cases(
+        m48r2.cases,
+        comparison.selected_candidate_rows,
+        store,
+    )
+    separation = _evaluate_separation(profile, cases)
+    baseline_metrics = _worker_metrics(baseline)
+    candidate_metrics = _worker_metrics(candidate)
+    provider = _provider_metrics(candidate)
+    critical = [row for row in cases if row["distance_band"] == "critical-near"]
+    near_recall = _rate(critical, "matched")
+    canonical_recall = (
+        1.0
+        if comparison.lost_cell_total == 0
+        and m48r2.report["metrics"]["canonical_engineering_recall"] == 1.0
+        else 0.0
+    )
+    capacity_drop_count = (
+        int(provider["geometry_failed_frames"])
+        + int(provider["temporal_failed_frames"])
+        + int(provider["rolling_capacity_evicted_cells"])
+    )
+    fps_regression = max(
+        0.0,
+        (baseline_metrics["fps"] - candidate_metrics["fps"])
+        / baseline_metrics["fps"],
+    )
+    world_p95_delta = (
+        candidate_metrics["world_p95_ms"] - baseline_metrics["world_p95_ms"]
+    )
+    gates = {
+        "complete_frame_accounting": candidate_metrics["delivered_frames"]
+        == EXPECTED_FRAMES,
+        "minimum_effective_world_state_fps": candidate_metrics["fps"]
+        >= float(acceptance["minimum_effective_world_state_fps"]),
+        "maximum_world_state_completion_p95_ms": candidate_metrics[
+            "world_p95_ms"
+        ]
+        <= float(acceptance["maximum_world_state_completion_p95_ms"]),
+        "maximum_geometry_stage_p95_ms": candidate_metrics["geometry_p95_ms"]
+        <= float(acceptance["maximum_geometry_stage_p95_ms"]),
+        "maximum_geometry_stage_p99_ms": candidate_metrics["geometry_p99_ms"]
+        <= float(acceptance["maximum_geometry_stage_p99_ms"]),
+        "maximum_fps_regression_fraction": fps_regression
+        <= float(acceptance["maximum_fps_regression_fraction_vs_native_baseline"]),
+        "maximum_world_state_p95_delta_ms": world_p95_delta
+        <= float(acceptance["maximum_world_state_p95_delta_ms_vs_native_baseline"]),
+        "maximum_component_mean_growth_fraction": comparison.mean_component_growth_fraction
+        <= float(acceptance["maximum_additive_component_mean_growth_fraction"]),
+        "maximum_cell_mean_growth_fraction": comparison.mean_cell_growth_fraction
+        <= float(acceptance["maximum_additive_cell_mean_growth_fraction"]),
+        "zero_capacity_drops": capacity_drop_count
+        <= int(acceptance["maximum_capacity_drop_count"]),
+        "zero_baseline_cell_loss": comparison.lost_cell_total == 0,
+        "critical_near_recall": near_recall
+        >= float(acceptance["minimum_critical_near_recall"]),
+        "canonical_engineering_recall": canonical_recall
+        >= float(acceptance["minimum_canonical_engineering_recall"]),
+        "zero_false_free": comparison.false_free_count
+        <= int(acceptance["maximum_false_free_count"]),
+        "separation_expectations": all(row["passed"] for row in separation),
+    }
+    accepted = all(gates.values())
+    producer_sha256 = _file_sha256(Path(__file__).resolve())
+    candidate_frame_sha256 = _file_sha256(candidate_frames_path)
+    baseline_frame_sha256 = _file_sha256(baseline_frames_path)
+    created_at = _worker_completed_at(candidate)
+    identity = {
+        "schema_version": M48R3_SHADOW_RESULT_SCHEMA,
+        "created_at_utc": created_at,
+        "profile_id": profile["profile_id"],
+        "profile_sha256": profile_sha256,
+        "producer_sha256": producer_sha256,
+        "baseline_result_sha256": _file_sha256(baseline_result_path),
+        "baseline_frames_sha256": baseline_frame_sha256,
+        "candidate_result_sha256": _file_sha256(candidate_result_path),
+        "candidate_frames_sha256": candidate_frame_sha256,
+        "m48r2_result_id": m48r2.result_id,
+        "authority": dict(_AUTHORITY),
+    }
+    result_id = M48R3_SHADOW_PREFIX + _canonical_sha256(identity)
+    metrics = {
+        "frames": {
+            "expected": EXPECTED_FRAMES,
+            "baseline_delivered": baseline_metrics["delivered_frames"],
+            "candidate_delivered": candidate_metrics["delivered_frames"],
+        },
+        "performance": {
+            "baseline": baseline_metrics,
+            "candidate": candidate_metrics,
+            "fps_regression_fraction": round(fps_regression, 9),
+            "world_state_p95_delta_ms": round(world_p95_delta, 6),
+        },
+        "occupancy": {
+            "baseline_cell_total": comparison.baseline_cell_total,
+            "candidate_cell_total": comparison.candidate_cell_total,
+            "added_cell_total": comparison.added_cell_total,
+            "lost_cell_total": comparison.lost_cell_total,
+            "baseline_component_total": comparison.baseline_component_total,
+            "candidate_component_total": comparison.candidate_component_total,
+            "mean_cell_growth_fraction": round(
+                comparison.mean_cell_growth_fraction, 9
+            ),
+            "mean_component_growth_fraction": round(
+                comparison.mean_component_growth_fraction, 9
+            ),
+            "maximum_added_cells_per_frame": comparison.maximum_added_cells_per_frame,
+            "maximum_candidate_components_per_frame": (
+                comparison.maximum_candidate_components_per_frame
+            ),
+        },
+        "provider": provider,
+        "assisted_anchors": {
+            "count": len(cases),
+            "critical_near_count": len(critical),
+            "critical_near_recall": near_recall,
+            "matched_count": sum(bool(row["matched"]) for row in cases),
+            "canonical_engineering_recall": canonical_recall,
+            "separation": separation,
+        },
+        "capacity_drop_count": capacity_drop_count,
+        "false_free_count": comparison.false_free_count,
+    }
+    report = {
+        "schema_version": M48R3_SHADOW_REPORT_SCHEMA,
+        "result_id": result_id,
+        "accepted": accepted,
+        "profile": {
+            "id": profile["profile_id"],
+            "sha256": profile_sha256,
+            "componentization": profile["componentization"],
+            "acceptance": acceptance,
+        },
+        "metrics": metrics,
+        "gates": gates,
+        "decision": {
+            "state": "accepted-bounded-worker-shadow"
+            if accepted
+            else "rejected-bounded-worker-shadow",
+            "candidate_accepted": accepted,
+            "production_accepted": False,
+            "next_action": "product-cutover-decision" if accepted else "reduce-load-or-coverage",
+        },
+        "limitations": [
+            "Operator-assisted rectangles are development anchors, not independent truth.",
+            "Separated occupied components do not assert passability for an unknown chassis.",
+            (
+                "No free-space, navigation, command, actuation or collision-safety "
+                "authority is granted."
+            ),
+        ],
+        "authority": dict(_AUTHORITY),
+    }
+    destination = output_root.resolve(strict=False) / result_id
+    _publish(
+        destination,
+        identity=identity,
+        report=report,
+        cases=cases,
+        diff_rows=comparison.diff_rows,
+        candidate_result_path=candidate_result_path,
+        candidate_frames_path=candidate_frames_path,
+    )
+    return read_m48r3_static_occupancy_shadow(destination)
+
+
+def compare_m48r3_frame_ledgers(
+    baseline_frames_path: Path,
+    candidate_frames_path: Path,
+    *,
+    selected_sequences: set[int],
+    expected_frames: int,
+) -> LedgerComparison:
+    """Stream two aligned ledgers and retain only a bounded cell-diff."""
+
+    diffs: list[dict[str, Any]] = []
+    selected: dict[int, dict[str, Any]] = {}
+    baseline_cells_total = 0
+    candidate_cells_total = 0
+    baseline_components_total = 0
+    candidate_components_total = 0
+    added_total = 0
+    lost_total = 0
+    false_free = 0
+    maximum_added = 0
+    maximum_components = 0
+    count = 0
+    with baseline_frames_path.open("r", encoding="utf-8") as baseline_stream, (
+        candidate_frames_path.open("r", encoding="utf-8")
+    ) as candidate_stream:
+        for baseline_line, candidate_line in zip(
+            baseline_stream,
+            candidate_stream,
+            strict=True,
+        ):
+            baseline = _frame_row(json.loads(baseline_line))
+            candidate = _frame_row(json.loads(candidate_line))
+            baseline_sequence = _sequence(baseline)
+            candidate_sequence = _sequence(candidate)
+            if baseline_sequence != candidate_sequence or baseline_sequence != count:
+                raise M48R3StaticOccupancyShadowError("frame ledgers are not aligned")
+            baseline_map = _obstacle_map(baseline)
+            candidate_map = _obstacle_map(candidate)
+            baseline_components = _active_components(baseline_map)
+            candidate_components = _active_components(candidate_map)
+            baseline_cells = _cell_union(baseline_components)
+            candidate_cells = _cell_union(candidate_components)
+            added = candidate_cells - baseline_cells
+            lost = baseline_cells - candidate_cells
+            provenance: dict[str, str] = {}
+            for component_id, cells in candidate_components:
+                if not cells.intersection(added):
+                    continue
+                provenance[component_id] = (
+                    "mixed" if cells.intersection(baseline_cells) else "additive-low-step"
+                )
+            diffs.append(
+                {
+                    "schema_version": M48R3_SHADOW_DIFF_SCHEMA,
+                    "sequence": baseline_sequence,
+                    "baseline_cell_count": len(baseline_cells),
+                    "candidate_cell_count": len(candidate_cells),
+                    "added_cells": [list(cell) for cell in sorted(added)],
+                    "lost_cell_count": len(lost),
+                    "component_provenance": provenance,
+                }
+            )
+            if baseline_sequence in selected_sequences:
+                selected[baseline_sequence] = candidate
+            baseline_cells_total += len(baseline_cells)
+            candidate_cells_total += len(candidate_cells)
+            baseline_components_total += len(baseline_components)
+            candidate_components_total += len(candidate_components)
+            added_total += len(added)
+            lost_total += len(lost)
+            false_free += candidate_map.get("free_space_claimed") is True
+            maximum_added = max(maximum_added, len(added))
+            maximum_components = max(maximum_components, len(candidate_components))
+            count += 1
+    if count != expected_frames or set(selected) != selected_sequences:
+        raise M48R3StaticOccupancyShadowError("frame ledgers are incomplete")
+    return LedgerComparison(
+        frame_count=count,
+        baseline_cell_total=baseline_cells_total,
+        candidate_cell_total=candidate_cells_total,
+        added_cell_total=added_total,
+        lost_cell_total=lost_total,
+        baseline_component_total=baseline_components_total,
+        candidate_component_total=candidate_components_total,
+        false_free_count=false_free,
+        maximum_added_cells_per_frame=maximum_added,
+        maximum_candidate_components_per_frame=maximum_components,
+        mean_cell_growth_fraction=_growth(candidate_cells_total, baseline_cells_total),
+        mean_component_growth_fraction=_growth(
+            candidate_components_total,
+            baseline_components_total,
+        ),
+        diff_rows=tuple(diffs),
+        selected_candidate_rows=selected,
+    )
+
+
+def read_m48r3_static_occupancy_shadow(
+    result_root: Path,
+) -> M48R3StaticOccupancyShadowResult:
+    root = result_root.resolve(strict=True)
+    if result_root.is_symlink() or not root.name.startswith(M48R3_SHADOW_PREFIX):
+        raise M48R3StaticOccupancyShadowError("M4.8R3 result root is invalid")
+    manifest = _read_json(root / "manifest.json", maximum=4 * 1024 * 1024)
+    report = _read_json(root / "report.json", maximum=8 * 1024 * 1024)
+    cases = tuple(_read_jsonl(root / "cases.jsonl", maximum=8 * 1024 * 1024))
+    if (
+        manifest.get("schema_version") != M48R3_SHADOW_RESULT_SCHEMA
+        or manifest.get("result_id") != root.name
+        or report.get("schema_version") != M48R3_SHADOW_REPORT_SCHEMA
+        or report.get("result_id") != root.name
+        or manifest.get("authority") != _AUTHORITY
+    ):
+        raise M48R3StaticOccupancyShadowError("M4.8R3 result identity changed")
+    identity = _object(manifest.get("identity"), "M4.8R3 identity")
+    if (
+        root.name != M48R3_SHADOW_PREFIX + _canonical_sha256(identity)
+        or manifest.get("identity_sha256") != _canonical_sha256(identity)
+    ):
+        raise M48R3StaticOccupancyShadowError("M4.8R3 digest changed")
+    artifacts = manifest.get("artifacts")
+    if not isinstance(artifacts, list):
+        raise M48R3StaticOccupancyShadowError("M4.8R3 artifacts changed")
+    for artifact in artifacts:
+        item = _object(artifact, "M4.8R3 artifact")
+        path = (root / str(item.get("path"))).resolve(strict=True)
+        if path.parent != root or path.is_symlink() or _file_sha256(path) != item.get("sha256"):
+            raise M48R3StaticOccupancyShadowError("M4.8R3 artifact changed")
+    return M48R3StaticOccupancyShadowResult(root.name, root, manifest, report, cases)
+
+
+def _evaluate_cases(
+    source_cases: Iterable[dict[str, Any]],
+    candidate_rows: dict[int, dict[str, Any]],
+    store: RecordedGeometryStore,
+) -> tuple[dict[str, Any], ...]:
+    result: list[dict[str, Any]] = []
+    for source in source_cases:
+        display_frame = int(source["sequence"])
+        sequence = display_frame - 1
+        frame = store.frame_for_index(sequence)
+        if frame is None:
+            raise M48R3StaticOccupancyShadowError("anchor frame is unavailable")
+        extent = source.get("extent_xyxy")
+        if not isinstance(extent, list) or len(extent) != 4:
+            raise M48R3StaticOccupancyShadowError("anchor extent changed")
+        bbox = (
+            float(extent[0]) * frame.projection.width,
+            float(extent[1]) * frame.projection.height,
+            float(extent[2]) * frame.projection.width,
+            float(extent[3]) * frame.projection.height,
+        )
+        matches = _component_matches(candidate_rows[sequence], frame, bbox)
+        result.append(
+            {
+                "schema_version": M48R3_SHADOW_CASE_SCHEMA,
+                "anchor_id": source["anchor_id"],
+                "display_frame": display_frame,
+                "source_sequence": sequence,
+                "extent_xyxy": extent,
+                "distance_band": source["distance_band"],
+                "component_count": len(matches),
+                "components": matches,
+                "matched": bool(matches),
+                "authority": "operator-assisted-development-anchor-not-truth",
+            }
+        )
+    return tuple(result)
+
+
+def _evaluate_separation(
+    profile: dict[str, Any],
+    cases: tuple[dict[str, Any], ...],
+) -> list[dict[str, Any]]:
+    by_anchor = {str(row["anchor_id"]): row for row in cases}
+    result: list[dict[str, Any]] = []
+    for value in profile.get("separation_expectations", []):
+        item = _object(value, "separation expectation")
+        anchor_id = str(item["anchor_id"])
+        case = by_anchor.get(anchor_id)
+        if case is None or case["display_frame"] != item["sequence"]:
+            raise M48R3StaticOccupancyShadowError("separation anchor changed")
+        expected = int(item["expected_minimum_components"])
+        observed = int(case["component_count"])
+        result.append(
+            {
+                "anchor_id": anchor_id,
+                "display_frame": case["display_frame"],
+                "source_sequence": case["source_sequence"],
+                "expected_minimum_components": expected,
+                "observed_components": observed,
+                "passed": observed >= expected,
+                "interpretation": item["interpretation"],
+            }
+        )
+    return result
+
+
+def _component_matches(
+    graph_row: dict[str, Any],
+    frame: Any,
+    bbox: tuple[float, float, float, float],
+) -> list[dict[str, Any]]:
+    projected_rows: list[dict[str, Any]] = []
+    for component_id, cells in _active_components(_obstacle_map(graph_row)):
+        if not cells:
+            continue
+        points = np.asarray(
+            [
+                (
+                    (cell[0] + 0.5) * VOXEL_SIZE_M,
+                    (cell[1] + 0.5) * VOXEL_SIZE_M,
+                    (cell[2] + 0.5) * VOXEL_SIZE_M,
+                )
+                for cell in cells
+            ],
+            dtype=np.float64,
+        )
+        projected = project_map_points_kb4(
+            points,
+            position_map_xyz=frame.sensor_position_map,
+            orientation_map_from_lidar_xyzw=frame.sensor_orientation_xyzw,
+            profile=frame.projection,
+        )
+        inside = (
+            (projected.pixels_xy[:, 0] >= bbox[0])
+            & (projected.pixels_xy[:, 0] <= bbox[2])
+            & (projected.pixels_xy[:, 1] >= bbox[1])
+            & (projected.pixels_xy[:, 1] <= bbox[3])
+        )
+        if not np.any(inside):
+            continue
+        depths = projected.depths_m[inside]
+        projected_rows.append(
+            {
+                "component_id": component_id,
+                "projected_cell_count": int(depths.size),
+                "nearest_depth_m": round(float(np.min(depths)), 6),
+            }
+        )
+    return sorted(projected_rows, key=lambda row: (row["nearest_depth_m"], row["component_id"]))
+
+
+def _worker_metrics(result: dict[str, Any]) -> dict[str, Any]:
+    execution = _object(result.get("execution"), "worker execution")
+    timing = _object(
+        _object(result.get("metrics"), "worker metrics").get("pipeline_timing"),
+        "pipeline timing",
+    )
+    geometry = _object(
+        _object(timing.get("provider_ms"), "provider timing").get("geometry"),
+        "geometry timing",
+    )
+    world = _object(
+        _object(result.get("metrics"), "worker metrics").get(
+            "world_state_completion_age_ms"
+        ),
+        "world timing",
+    )
+    return {
+        "admitted_frames": int(execution["admitted_frames"]),
+        "delivered_frames": int(execution["delivered_world_states"]),
+        "fps": float(execution["effective_world_state_fps"]),
+        "world_p95_ms": float(world["p95"]),
+        "world_p99_ms": float(world["p99"]),
+        "geometry_p95_ms": float(geometry["p95"]),
+        "geometry_p99_ms": float(geometry["p99"]),
+    }
+
+
+def _provider_metrics(result: dict[str, Any]) -> dict[str, Any]:
+    loops = _object(result["execution"]["loops"][0], "worker loop")
+    providers = _object(loops.get("providers"), "providers")
+    geometry = _object(providers.get("geometry"), "geometry provider")
+    temporal = _object(providers.get("temporal"), "temporal provider")
+    rolling = _object(providers.get("rolling"), "rolling provider")
+    additive_duration_ns = int(geometry["additive_core_duration_ns"])
+    completed = int(geometry["completed_frames"])
+    return {
+        "additive_observation_count": int(geometry["additive_observation_count"]),
+        "additive_voxel_count": int(geometry["additive_voxel_count"]),
+        "frames_with_additions": int(geometry["frames_with_additions"]),
+        "peak_additive_observations_per_frame": int(
+            geometry["peak_additive_observations_per_frame"]
+        ),
+        "peak_candidate_points_per_frame": int(geometry["peak_candidate_points_per_frame"]),
+        "peak_voxels_per_component": int(geometry["peak_voxels_per_component"]),
+        "additive_mean_ms_per_frame": round(
+            additive_duration_ns / max(1, completed) / 1_000_000,
+            6,
+        ),
+        "geometry_failed_frames": int(geometry["failed_frames"]),
+        "temporal_failed_frames": int(temporal["failed_frames"]),
+        "rolling_capacity_evicted_cells": int(rolling["capacity_evicted_cells"]),
+        "peak_temporal_components": int(temporal["peak_active_components"]),
+        "peak_rolling_cells": int(rolling["peak_active_cells"]),
+    }
+
+
+def _validate_worker_binding(
+    result: dict[str, Any],
+    frames_path: Path,
+    *,
+    expected_frames: int,
+    additive_profile_sha256: str | None,
+) -> None:
+    execution = _object(result.get("execution"), "worker execution")
+    evidence = _object(execution.get("frame_evidence"), "frame evidence")
+    identity = _object(result.get("identity"), "worker identity")
+    inputs = _object(identity.get("inputs"), "worker inputs")
+    if (
+        result.get("schema_version") != WORKER_RESULT_SCHEMA
+        or result.get("completed") is not True
+        or execution.get("admitted_frames") != expected_frames
+        or evidence.get("schema_version") != FRAME_EVIDENCE_SCHEMA
+        or evidence.get("sha256") != _file_sha256(frames_path)
+        or evidence.get("row_count") != execution.get("delivered_world_states")
+        or identity.get("worker_id") != "worker-006"
+        or result.get("authority")
+        != {
+            "actuation_allowed": False,
+            "candidate_accepted": False,
+            "commands_enabled": False,
+            "ground_truth": False,
+            "navigation_or_safety_accepted": False,
+        }
+    ):
+        raise M48R3StaticOccupancyShadowError("worker result binding changed")
+    if additive_profile_sha256 is None:
+        if "additive_low_step_profile" in inputs:
+            raise M48R3StaticOccupancyShadowError("baseline is not native-only")
+    elif inputs.get("additive_low_step_profile") != additive_profile_sha256:
+        raise M48R3StaticOccupancyShadowError("candidate profile binding changed")
+
+
+def _read_worker_result(path: Path) -> dict[str, Any]:
+    return _read_json(path, maximum=128 * 1024 * 1024)
+
+
+def _frame_row(value: object) -> dict[str, Any]:
+    row = _object(value, "frame row")
+    if row.get("schema_version") != FRAME_EVIDENCE_SCHEMA:
+        raise M48R3StaticOccupancyShadowError("frame schema changed")
+    return row
+
+
+def _sequence(row: dict[str, Any]) -> int:
+    envelope = _object(row.get("source_envelope"), "source envelope")
+    value = envelope.get("sequence")
+    if not isinstance(value, int) or isinstance(value, bool):
+        raise M48R3StaticOccupancyShadowError("frame sequence changed")
+    return value
+
+
+def _obstacle_map(row: dict[str, Any]) -> dict[str, Any]:
+    delivery = _object(row.get("delivery"), "delivery")
+    return _object(delivery.get("obstacle_map"), "obstacle map")
+
+
+def _active_components(obstacle_map: dict[str, Any]) -> list[tuple[str, set[Cell]]]:
+    result: list[tuple[str, set[Cell]]] = []
+    for collection in (obstacle_map.get("occupied"), obstacle_map.get("unknown")):
+        if not isinstance(collection, list):
+            raise M48R3StaticOccupancyShadowError("obstacle collection changed")
+        for value in collection:
+            item = _object(value, "obstacle")
+            component_id = item.get("component_id")
+            cells = item.get("cells")
+            if not isinstance(component_id, str) or not isinstance(cells, list):
+                raise M48R3StaticOccupancyShadowError("obstacle component changed")
+            parsed: set[Cell] = set()
+            for cell in cells:
+                row = _object(cell, "obstacle cell")
+                x, y, z = row.get("x"), row.get("y"), row.get("z")
+                if any(
+                    not isinstance(item, int) or isinstance(item, bool)
+                    for item in (x, y, z)
+                ):
+                    raise M48R3StaticOccupancyShadowError("obstacle cell changed")
+                assert isinstance(x, int) and isinstance(y, int) and isinstance(z, int)
+                parsed.add((x, y, z))
+            if parsed:
+                result.append((component_id, parsed))
+    return result
+
+
+def _cell_union(components: Iterable[tuple[str, set[Cell]]]) -> set[Cell]:
+    result: set[Cell] = set()
+    for _, cells in components:
+        result.update(cells)
+    return result
+
+
+def _growth(candidate: int, baseline: int) -> float:
+    if baseline <= 0:
+        return math.inf if candidate > 0 else 0.0
+    return max(0.0, (candidate - baseline) / baseline)
+
+
+def _rate(rows: list[dict[str, Any]], key: str) -> float:
+    return sum(bool(row[key]) for row in rows) / len(rows) if rows else 0.0
+
+
+def _worker_completed_at(result: dict[str, Any]) -> str:
+    value = result.get("completed_utc_ns")
+    if not isinstance(value, int) or isinstance(value, bool) or value <= 0:
+        raise M48R3StaticOccupancyShadowError("worker completion time changed")
+    return datetime.fromtimestamp(value / 1_000_000_000, UTC).isoformat().replace(
+        "+00:00", "Z"
+    )
+
+
+def _publish(
+    destination: Path,
+    *,
+    identity: dict[str, Any],
+    report: dict[str, Any],
+    cases: tuple[dict[str, Any], ...],
+    diff_rows: tuple[dict[str, Any], ...],
+    candidate_result_path: Path,
+    candidate_frames_path: Path,
+) -> None:
+    destination.parent.mkdir(mode=0o700, parents=True, exist_ok=True)
+    staging = destination.parent / f".{destination.name}.{uuid.uuid4().hex}.tmp"
+    staging.mkdir(mode=0o700)
+    try:
+        _write_json(staging / "report.json", report)
+        _write_jsonl(staging / "cases.jsonl", cases)
+        _write_jsonl(staging / "frame-diff.jsonl", diff_rows)
+        shutil.copyfile(candidate_result_path, staging / "worker-result.json")
+        shutil.copyfile(candidate_frames_path, staging / "frames.jsonl")
+        artifacts = [
+            _artifact(staging / name, role)
+            for name, role in (
+                ("report.json", "m48r3-shadow-report"),
+                ("cases.jsonl", "operator-assisted-anchor-projection"),
+                ("frame-diff.jsonl", "baseline-versus-candidate-occupied-diff"),
+                ("worker-result.json", "worker-load-result"),
+                ("frames.jsonl", "candidate-frame-ledger"),
+            )
+        ]
+        manifest = {
+            "schema_version": M48R3_SHADOW_RESULT_SCHEMA,
+            "result_id": destination.name,
+            "identity_sha256": _canonical_sha256(identity),
+            "identity": identity,
+            "created_at_utc": identity["created_at_utc"],
+            "accepted": report["accepted"],
+            "ground_truth": False,
+            "authority": dict(_AUTHORITY),
+            "artifacts": artifacts,
+        }
+        _write_json(staging / "manifest.json", manifest)
+        if destination.exists():
+            raise M48R3StaticOccupancyShadowError("immutable result already exists")
+        os.replace(staging, destination)
+    except BaseException:
+        shutil.rmtree(staging, ignore_errors=True)
+        raise
+
+
+def _artifact(path: Path, role: str) -> dict[str, Any]:
+    return {
+        "path": path.name,
+        "role": role,
+        "byte_length": path.stat().st_size,
+        "sha256": _file_sha256(path),
+        "media_type": "application/x-ndjson" if path.suffix == ".jsonl" else "application/json",
+    }
+
+
+def _read_json(path: Path, *, maximum: int) -> dict[str, Any]:
+    if path.is_symlink() or not path.is_file() or path.stat().st_size > maximum:
+        raise M48R3StaticOccupancyShadowError(f"{path.name} is unavailable")
+    return _object(json.loads(path.read_text("utf-8")), path.name)
+
+
+def _read_jsonl(path: Path, *, maximum: int) -> list[dict[str, Any]]:
+    if path.is_symlink() or not path.is_file() or path.stat().st_size > maximum:
+        raise M48R3StaticOccupancyShadowError(f"{path.name} is unavailable")
+    return [
+        _object(json.loads(line), path.name)
+        for line in path.read_text("utf-8").splitlines()
+        if line.strip()
+    ]
+
+
+def _write_json(path: Path, value: object) -> None:
+    path.write_text(
+        json.dumps(value, ensure_ascii=False, sort_keys=True, indent=2) + "\n",
+        "utf-8",
+    )
+
+
+def _write_jsonl(path: Path, rows: Iterable[dict[str, Any]]) -> None:
+    with path.open("w", encoding="utf-8") as stream:
+        for row in rows:
+            stream.write(
+                json.dumps(
+                    row,
+                    ensure_ascii=False,
+                    sort_keys=True,
+                    separators=(",", ":"),
+                )
+                + "\n"
+            )
+
+
+def _file_sha256(path: Path) -> str:
+    digest = hashlib.sha256()
+    with path.open("rb") as stream:
+        for chunk in iter(lambda: stream.read(1024 * 1024), b""):
+            digest.update(chunk)
+    return digest.hexdigest()
+
+
+def _canonical_sha256(value: object) -> str:
+    return hashlib.sha256(
+        json.dumps(value, ensure_ascii=False, sort_keys=True, separators=(",", ":")).encode(
+            "utf-8"
+        )
+    ).hexdigest()
+
+
+def _object(value: object, label: str) -> dict[str, Any]:
+    if not isinstance(value, dict):
+        raise M48R3StaticOccupancyShadowError(f"{label} is invalid")
+    return value
+
+
+__all__ = [
+    "LedgerComparison",
+    "M48R3StaticOccupancyShadowError",
+    "M48R3StaticOccupancyShadowResult",
+    "build_m48r3_static_occupancy_shadow",
+    "compare_m48r3_frame_ledgers",
+    "read_m48r3_static_occupancy_shadow",
+]

src/k1link/perception/m48_low_step_occupancy.py

@@ -36,9 +36,7 @@ from .geometry import (
 from .geometry_math import POINT_OCCUPIED
 from .providers import SourcePacket
 
-M48_LOW_STEP_PROFILE_SCHEMA: Final = (
+M48_LOW_STEP_PROFILE_SCHEMA: Final = "missioncore.m48-additive-low-step-occupancy-profile/v1"
-    "missioncore.m48-additive-low-step-occupancy-profile/v1"
-)
 M48_LOW_STEP_PROVIDER_ID: Final = "ravnoves00-additive-low-step-geometry/v1"
 
 IntArray = npt.NDArray[np.int64]
@@ -53,6 +51,10 @@ class LowStepComponentProfile:
     voxel_size_m: float
     neighbor_radius_cells: int
     minimum_points: int
+    sparse_persistence_minimum_points: int
+    sparse_persistence_window_frames: int
+    sparse_persistence_minimum_hits: int
+    sparse_persistence_maximum_range_m: float
     minimum_voxels: int
     local_radius_m: float
     maximum_candidate_points_per_frame: int
@@ -65,6 +67,13 @@ class LowStepComponentProfile:
             or not 0.05 <= self.voxel_size_m <= 2.0
             or self.neighbor_radius_cells != 1
             or not 1 <= self.minimum_points <= 256
+            or not 1 <= self.sparse_persistence_minimum_points < self.minimum_points
+            or not 2 <= self.sparse_persistence_window_frames <= 32
+            or not 2
+            <= self.sparse_persistence_minimum_hits
+            <= self.sparse_persistence_window_frames
+            or not math.isfinite(self.sparse_persistence_maximum_range_m)
+            or not 1.0 <= self.sparse_persistence_maximum_range_m <= self.local_radius_m
             or not 1 <= self.minimum_voxels <= 128
             or not math.isfinite(self.local_radius_m)
             or not 1.0 <= self.local_radius_m <= 100.0
@@ -156,6 +165,11 @@ class M48AdditiveLowStepGeometryProvider:
         self._peak_additive_observations = 0
         self._peak_component_voxels = 0
         self._additive_core_duration_ns = 0
+        self._sparse_lock = Lock()
+        self._sparse_history: deque[tuple[int, tuple[frozenset[tuple[int, int, int]], ...]]] = (
+            deque()
+        )
+        self._last_sparse_sequence: int | None = None
 
     def associate(
         self,
@@ -175,9 +189,7 @@ class M48AdditiveLowStepGeometryProvider:
         except Exception:
             with self._lock:
                 self._failed_frames += 1
-                self._additive_core_duration_ns += max(
+                self._additive_core_duration_ns += max(0, time.perf_counter_ns() - started)
-                    0, time.perf_counter_ns() - started
-                )
             raise
         with self._lock:
             self._completed_frames += 1
@@ -185,18 +197,10 @@ class M48AdditiveLowStepGeometryProvider:
             self._candidate_points += candidate_points
             self._additive_observation_count += len(additive)
             self._additive_voxels += voxel_count
-            self._peak_candidate_points = max(
+            self._peak_candidate_points = max(self._peak_candidate_points, candidate_points)
-                self._peak_candidate_points, candidate_points
+            self._peak_additive_observations = max(self._peak_additive_observations, len(additive))
-            )
+            self._peak_component_voxels = max(self._peak_component_voxels, peak_component_voxels)
-            self._peak_additive_observations = max(
+            self._additive_core_duration_ns += max(0, time.perf_counter_ns() - started)
-                self._peak_additive_observations, len(additive)
-            )
-            self._peak_component_voxels = max(
-                self._peak_component_voxels, peak_component_voxels
-            )
-            self._additive_core_duration_ns += max(
-                0, time.perf_counter_ns() - started
-            )
         return tuple(result)
 
     def _build_additive_observations(
@@ -213,9 +217,9 @@ class M48AdditiveLowStepGeometryProvider:
         claimed = {
             point_id for observation in baseline for point_id in observation.source_point_ids
         }
-        candidate = np.flatnonzero(
+        candidate = np.flatnonzero((step > 0) & (frame.point_class != POINT_OCCUPIED)).astype(
-            (step > 0) & (frame.point_class != POINT_OCCUPIED)
+            np.int64
-        ).astype(np.int64)
+        )
         if claimed and candidate.size:
             candidate = candidate[
                 np.fromiter(
@@ -240,12 +244,19 @@ class M48AdditiveLowStepGeometryProvider:
             candidate,
             self.profile.component,
         )
-        qualified = tuple(
+        strong = tuple(
             item
             for item in components
             if item[0].size >= self.profile.component.minimum_points
             and item[1] >= self.profile.component.minimum_voxels
         )
+        persistent_sparse = self._persistent_sparse_components(
+            sequence=packet.envelope.sequence,
+            components=components,
+            points_map=frame.points_map,
+            sensor_position_map=frame.sensor_position_map,
+        )
+        qualified = (*strong, *persistent_sparse)
         qualified = tuple(
             sorted(
                 qualified,
@@ -253,8 +264,7 @@ class M48AdditiveLowStepGeometryProvider:
                     float(
                         np.min(
                             np.linalg.norm(
-                                frame.points_map[item[0]]
+                                frame.points_map[item[0]] - frame.sensor_position_map,
-                                - frame.sensor_position_map,
                                 axis=1,
                             )
                         )
@@ -279,17 +289,11 @@ class M48AdditiveLowStepGeometryProvider:
             points = frame.points_map[indices]
             centroid = np.median(points, axis=0)
             covariance = points.var(axis=0)
-            nearest = float(
+            nearest = float(np.min(np.linalg.norm(points - frame.sensor_position_map, axis=1)))
-                np.min(np.linalg.norm(points - frame.sensor_position_map, axis=1))
-            )
             observations.append(
                 ObstacleObservation(
-                    observation_id=(
+                    observation_id=(f"{packet.envelope.frame_id}:low-step:{component_index}"),
-                        f"{packet.envelope.frame_id}:low-step:{component_index}"
+                    occupancy_key=(f"{packet.envelope.frame_id}:low-step:{component_index}"),
-                    ),
-                    occupancy_key=(
-                        f"{packet.envelope.frame_id}:low-step:{component_index}"
-                    ),
                     source_id=packet.envelope.source_id,
                     frame_id=packet.envelope.frame_id,
                     evidence_time_ns=packet.envelope.timestamps.source_ns,
@@ -323,6 +327,65 @@ class M48AdditiveLowStepGeometryProvider:
             peak_voxels = max(peak_voxels, cells)
         return tuple(observations), candidate_count, voxel_count, peak_voxels
 
+    def _persistent_sparse_components(
+        self,
+        *,
+        sequence: int,
+        components: tuple[tuple[IntArray, int], ...],
+        points_map: npt.NDArray[np.float64],
+        sensor_position_map: npt.NDArray[np.float64],
+    ) -> tuple[tuple[IntArray, int], ...]:
+        """Promote only weak geometry repeated in a bounded causal window."""
+
+        component_profile = self.profile.component
+        current = tuple(
+            (
+                item,
+                _component_cells(
+                    points_map,
+                    item[0],
+                    voxel_size_m=component_profile.voxel_size_m,
+                ),
+            )
+            for item in components
+            if item[0].size >= component_profile.sparse_persistence_minimum_points
+            and item[1] >= component_profile.minimum_voxels
+        )
+        with self._sparse_lock:
+            if self._last_sparse_sequence is not None and (
+                sequence <= self._last_sparse_sequence
+                or sequence - self._last_sparse_sequence
+                > component_profile.sparse_persistence_window_frames
+            ):
+                self._sparse_history.clear()
+            first_allowed = sequence - component_profile.sparse_persistence_window_frames + 1
+            while self._sparse_history and self._sparse_history[0][0] < first_allowed:
+                self._sparse_history.popleft()
+            promoted: list[tuple[IntArray, int]] = []
+            for item, cells in current:
+                if item[0].size >= component_profile.minimum_points:
+                    continue
+                hit_count = 1 + sum(
+                    any(not cells.isdisjoint(previous) for previous in previous_components)
+                    for _, previous_components in self._sparse_history
+                )
+                if (
+                    hit_count >= component_profile.sparse_persistence_minimum_hits
+                    and float(
+                        np.min(
+                            np.linalg.norm(
+                                points_map[item[0]] - sensor_position_map,
+                                axis=1,
+                            )
+                        )
+                    )
+                    <= component_profile.sparse_persistence_maximum_range_m
+                ):
+                    promoted.append(item)
+            self._sparse_history.append((sequence, tuple(cells for _, cells in current)))
+            self._last_sparse_sequence = sequence
+        return tuple(promoted)
+
     def snapshot(self) -> M48LowStepOccupancySnapshot:
         with self._lock:
             return M48LowStepOccupancySnapshot(
@@ -335,9 +398,7 @@ class M48AdditiveLowStepGeometryProvider:
                 additive_observation_count=self._additive_observation_count,
                 additive_voxel_count=self._additive_voxels,
                 peak_candidate_points_per_frame=self._peak_candidate_points,
-                peak_additive_observations_per_frame=(
+                peak_additive_observations_per_frame=(self._peak_additive_observations),
-                    self._peak_additive_observations
-                ),
                 peak_voxels_per_component=self._peak_component_voxels,
                 additive_core_duration_ns=self._additive_core_duration_ns,
             )
@@ -401,6 +462,10 @@ def load_m48_low_step_occupancy_profile(
             "voxel_size_m",
             "neighbor_radius_cells",
             "minimum_points",
+            "sparse_persistence_minimum_points",
+            "sparse_persistence_window_frames",
+            "sparse_persistence_minimum_hits",
+            "sparse_persistence_maximum_range_m",
             "minimum_voxels",
             "local_radius_m",
             "maximum_candidate_points_per_frame",
@@ -460,9 +525,7 @@ def load_m48_low_step_occupancy_profile(
         if _number(acceptance, key) < 0.0:
             raise M48LowStepOccupancyError("low-step acceptance bounds are invalid")
     if (
-        not _string(source, "m48r2_result_id").startswith(
+        not _string(source, "m48r2_result_id").startswith("m48-static-occupancy-qualification-")
-            "m48-static-occupancy-qualification-"
-        )
         or len(_string(source, "m48r2_result_id"))
         != len("m48-static-occupancy-qualification-") + 64
     ):
@@ -500,9 +563,7 @@ def load_m48_low_step_occupancy_profile(
             LowStepSeparationExpectation(
                 anchor_id=_string(item, "anchor_id"),
                 sequence=_positive_integer(item, "sequence"),
-                expected_minimum_components=_positive_integer(
+                expected_minimum_components=_positive_integer(item, "expected_minimum_components"),
-                    item, "expected_minimum_components"
-                ),
                 interpretation=_string(item, "interpretation"),
             )
         )
@@ -519,18 +580,30 @@ def load_m48_low_step_occupancy_profile(
         base_geometry_profile_sha256=_digest(base, "sha256"),
         component=LowStepComponentProfile(
             voxel_size_m=_number(component, "voxel_size_m"),
-            neighbor_radius_cells=_positive_integer(
+            neighbor_radius_cells=_positive_integer(component, "neighbor_radius_cells"),
-                component, "neighbor_radius_cells"
-            ),
             minimum_points=_positive_integer(component, "minimum_points"),
+            sparse_persistence_minimum_points=_positive_integer(
+                component,
+                "sparse_persistence_minimum_points",
+            ),
+            sparse_persistence_window_frames=_positive_integer(
+                component,
+                "sparse_persistence_window_frames",
+            ),
+            sparse_persistence_minimum_hits=_positive_integer(
+                component,
+                "sparse_persistence_minimum_hits",
+            ),
+            sparse_persistence_maximum_range_m=_number(
+                component,
+                "sparse_persistence_maximum_range_m",
+            ),
             minimum_voxels=_positive_integer(component, "minimum_voxels"),
             local_radius_m=_number(component, "local_radius_m"),
             maximum_candidate_points_per_frame=_positive_integer(
                 component, "maximum_candidate_points_per_frame"
             ),
-            maximum_cells_per_component=_positive_integer(
+            maximum_cells_per_component=_positive_integer(component, "maximum_cells_per_component"),
-                component, "maximum_cells_per_component"
-            ),
             maximum_components_per_frame=_positive_integer(
                 component, "maximum_components_per_frame"
             ),
@@ -547,9 +620,7 @@ def _voxel_components(
 ) -> tuple[tuple[IntArray, int], ...]:
     if source_indices.size == 0:
         return ()
-    cells = np.floor(
+    cells = np.floor(points_map[source_indices] / profile.voxel_size_m).astype(np.int64)
-        points_map[source_indices] / profile.voxel_size_m
-    ).astype(np.int64)
     cell_points: dict[tuple[int, int, int], list[int]] = {}
     for local_index, row in enumerate(cells):
         key = (int(row[0]), int(row[1]), int(row[2]))
@@ -591,6 +662,16 @@ def _voxel_components(
     return tuple(components)
 
 
+def _component_cells(
+    points_map: npt.NDArray[np.float64],
+    source_indices: IntArray,
+    *,
+    voxel_size_m: float,
+) -> frozenset[tuple[int, int, int]]:
+    rows = np.floor(points_map[source_indices] / voxel_size_m).astype(np.int64)
+    return frozenset((int(row[0]), int(row[1]), int(row[2])) for row in rows)
+
+
 def _object(value: object, label: str) -> dict[str, object]:
     if not isinstance(value, dict) or not all(isinstance(key, str) for key in value):
         raise M48LowStepOccupancyError(f"{label} must be an object")

src/k1link/perception/m48s_replay_timeline.py

@@ -70,6 +70,7 @@ class M48sReplayTimeline:
         frames_name: str = "reference-graph-replay-frames.jsonl",
         worker_result_name: str = "reference-graph-replay-worker-result.json",
         frame_evidence_schema: str = FRAME_EVIDENCE_SCHEMA,
+        frame_diff_name: str | None = None,
         camera_endpoint_root: str = (
             "/api/v1/laboratory/m48s/fixed-class-detector"
         ),
@@ -80,6 +81,10 @@ class M48sReplayTimeline:
         if (
             Path(frames_name).name != frames_name
             or Path(worker_result_name).name != worker_result_name
+            or (
+                frame_diff_name is not None
+                and Path(frame_diff_name).name != frame_diff_name
+            )
             or frame_evidence_schema not in {
                 FRAME_EVIDENCE_SCHEMA,
                 M48R3_FRAME_EVIDENCE_SCHEMA,
@@ -92,11 +97,23 @@ class M48sReplayTimeline:
         self.camera_endpoint_root = camera_endpoint_root
         self.frames_path = (self.result_root / frames_name).resolve(strict=True)
         self.worker_path = (self.result_root / worker_result_name).resolve(strict=True)
+        self.frame_diff_path = (
+            None
+            if frame_diff_name is None
+            else (self.result_root / frame_diff_name).resolve(strict=True)
+        )
         if (
             self.frames_path.parent != self.result_root
             or self.worker_path.parent != self.result_root
             or self.frames_path.is_symlink()
             or self.worker_path.is_symlink()
+            or (
+                self.frame_diff_path is not None
+                and (
+                    self.frame_diff_path.parent != self.result_root
+                    or self.frame_diff_path.is_symlink()
+                )
+            )
         ):
             raise M48sReplayTimelineError("M4.8S replay artifacts are invalid")
         self.profile = load_replay_threat_profile(
@@ -123,6 +140,11 @@ class M48sReplayTimeline:
             self.outcomes,
             frame_evidence_schema=self.frame_evidence_schema,
         )
+        self.frame_diff_offsets = (
+            {}
+            if self.frame_diff_path is None
+            else _index_frame_diff(self.frame_diff_path)
+        )
         self._cache_lock = Lock()
         self._chunk_json_cache: OrderedDict[tuple[int, int], bytes] = OrderedDict()
         self._camera_point_json_cache: OrderedDict[int, bytes] = OrderedDict()
@@ -156,6 +178,11 @@ class M48sReplayTimeline:
             "camera_point_window_seconds": CAMERA_ACCUMULATION_WINDOW_SECONDS,
             "camera_point_sample_limit": CAMERA_ACCUMULATION_POINT_LIMIT,
             "world_state_delivery": "source-paced-latest-wins",
+            "occupancy_provenance_delivery": (
+                "baseline-versus-additive-component-diff"
+                if self.frame_diff_path is not None
+                else None
+            ),
             "world_state_frame_count": len(self.index.offsets_by_sequence),
             "superseded_frame_count": sum(
                 value == "superseded" for value in self.outcomes.values()
@@ -399,6 +426,12 @@ class M48sReplayTimeline:
                     body_frame,
                     occupied_voxel_size_m=self.profile.corridor.occupied_voxel_size_m,
                 )
+                provenance = self._component_provenance(sequence)
+                for visual in metric_visuals:
+                    visual["occupancy_source"] = provenance.get(
+                        str(visual["component_id"]),
+                        "baseline",
+                    )
             associated = set(_strings(row.get("associated_proposal_ids"), "associated ids"))
             for proposal in _objects(row.get("detector_proposals"), "detector proposals"):
                 proposal_id = _text(proposal.get("proposal_id"), "proposal id")
@@ -480,6 +513,27 @@ class M48sReplayTimeline:
             raise M48sReplayTimelineError("M4.8S frame row is invalid")
         return value
 
+    def _component_provenance(self, sequence: int) -> dict[str, str]:
+        if self.frame_diff_path is None:
+            return {}
+        offset = self.frame_diff_offsets.get(sequence)
+        if offset is None:
+            raise M48sReplayTimelineError("M4.8R3 frame diff is incomplete")
+        with self.frame_diff_path.open("rb") as stream:
+            stream.seek(offset)
+            line = stream.readline()
+        value = _object(json.loads(line), "frame diff")
+        if value.get("sequence") != sequence:
+            raise M48sReplayTimelineError("M4.8R3 frame diff binding changed")
+        provenance = value.get("component_provenance")
+        if not isinstance(provenance, dict) or any(
+            not isinstance(key, str)
+            or item not in {"mixed", "additive-low-step"}
+            for key, item in provenance.items()
+        ):
+            raise M48sReplayTimelineError("M4.8R3 component provenance changed")
+        return provenance
+
 
 def _index_ledger(
     path: Path,
@@ -517,6 +571,33 @@ def _index_ledger(
     return _LedgerIndex(offsets)
 
 
+def _index_frame_diff(path: Path) -> dict[int, int]:
+    offsets: dict[int, int] = {}
+    with path.open("rb") as stream:
+        while True:
+            offset = stream.tell()
+            line = stream.readline()
+            if not line:
+                break
+            try:
+                value = json.loads(line)
+            except json.JSONDecodeError:
+                raise M48sReplayTimelineError("M4.8R3 frame diff is invalid") from None
+            if not isinstance(value, dict):
+                raise M48sReplayTimelineError("M4.8R3 frame diff is invalid")
+            sequence = value.get("sequence")
+            if (
+                not isinstance(sequence, int)
+                or isinstance(sequence, bool)
+                or sequence != len(offsets)
+            ):
+                raise M48sReplayTimelineError("M4.8R3 frame diff sequence changed")
+            offsets[sequence] = offset
+    if len(offsets) != EXPECTED_FRAME_COUNT:
+        raise M48sReplayTimelineError("M4.8R3 frame diff is incomplete")
+    return offsets
+
+
 def _ledger_source_envelope(
     line: bytes,
     *,

src/k1link/simulation/projects.py

@@ -297,9 +297,17 @@ class SimulationProjectStore:
         result: dict[str, Any],
         artifacts: list[dict[str, Any]],
         world_manifest: dict[str, Any],
+        provider_job_id: str | None = None,
+        provider_progress: object = None,
     ) -> dict[str, Any]:
         with self._lock:
             document = self._read(project_id)
+            if provider_job_id is not None:
+                if PROVIDER_JOB_ID_PATTERN.fullmatch(provider_job_id) is None:
+                    raise SimulationProjectError("simulation provider job id is invalid")
+                document["provider"]["job_id"] = provider_job_id
+            if provider_progress is not None:
+                document["provider"]["progress"] = provider_progress
             document["status"] = "ready"
             document["provider"]["state"] = "ready"
             document["provider"]["runtime"] = result.get("runtime")
@@ -464,10 +472,10 @@ class SimulationProjectService:
                     "outputs": {
                         "preview_sog": True,
                         "streamed_sog": True,
-                        "collision": False,
+                        "collision": True,
                     },
                     "preview_lod": "coarsest",
-                    "collision_profile": None,
+                    "collision_profile": _collision_profile(str(project["scene_type"])),
                 }
                 submitted = provider.submit_build(request)
                 job_id = submitted.get("job_id")
@@ -607,12 +615,24 @@ def _world_manifest(project_id: str, artifacts: list[dict[str, Any]]) -> dict[st
             "available": url_for("collision-mesh") is not None,
         },
         "transforms": {
-            "world_from_visual": [1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1],
+            "world_from_visual": [1, 0, 0, 0, 0, -1, 0, 0, 0, 0, -1, 0, 0, 0, 0, 1],
             "world_from_collision": [1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1],
         },
     }
 
 
+def _collision_profile(scene_type: str) -> dict[str, Any]:
+    """Build the portable walkable-volume contract around the scanner origin."""
+    return {
+        "scene_type": scene_type,
+        "seed_position": [0, 1, 0],
+        "capsule_height": 1.6,
+        "capsule_radius": 0.2,
+        "voxel_size": 0.05,
+        "mesh_shape": "smooth",
+    }
+
+
 def _project_name(value: str) -> str:
     normalized = " ".join(value.split())
     if not 1 <= len(normalized) <= 120:

src/k1link/web/app.py

@@ -126,6 +126,9 @@ from k1link.web.lidar_api import build_lidar_router
 from k1link.web.lidar_local_surface_service import K1LocalSurfaceReadService
 from k1link.web.m4_threat_replay_api import build_m4_threat_replay_router
 from k1link.web.m48_object_quality_api import build_m48_object_quality_router
+from k1link.web.m48r3_static_occupancy_api import (
+    build_m48r3_static_occupancy_router,
+)
 from k1link.web.m48s_fixed_class_detector_lab_api import (
     build_m48s_fixed_class_detector_lab_router,
 )
@@ -961,6 +964,23 @@ app.include_router(
         ),
     )
 )
+app.include_router(
+    build_m48r3_static_occupancy_router(
+        root_provider=lambda: (
+            REPOSITORY_ROOT
+            / ".runtime"
+            / "compute-experiments"
+            / "m48"
+            / "static-occupancy-shadow-results"
+        ),
+        repository_root_provider=lambda: REPOSITORY_ROOT,
+        camera_frame_provider=(
+            session_recorded_camera_frame_service.extract
+            if session_recorded_camera_frame_service is not None
+            else None
+        ),
+    )
+)
 app.include_router(
     build_m48s_fixed_class_detector_lab_router(
         root_provider=lambda: (

src/k1link/web/m48r3_static_occupancy_api.py

@@ -0,0 +1,290 @@
+"""Read-only API for sealed M4.8R3 static-occupancy Worker shadows."""
+
+from __future__ import annotations
+
+import copy
+import re
+from collections.abc import Callable
+from functools import lru_cache
+from pathlib import Path
+from typing import Final
+
+from fastapi import APIRouter, HTTPException, Query, Response
+
+from k1link.laboratory.m48r3_static_occupancy_shadow import (
+    M48R3_SHADOW_PREFIX,
+    M48R3StaticOccupancyShadowError,
+    M48R3StaticOccupancyShadowResult,
+    read_m48r3_static_occupancy_shadow,
+)
+from k1link.perception.m48s_replay_timeline import (
+    M48R3_FRAME_EVIDENCE_SCHEMA,
+    M48sReplayTimeline,
+    M48sReplayTimelineError,
+)
+from k1link.perception.threat_timeline import RECORDED_SPATIAL_MAX_CHUNK_FRAMES
+from k1link.sessions import RecordedCameraFrame, SessionIntegrityError
+
+RootProvider = Callable[[], Path | None]
+CameraFrameProvider = Callable[[str, int], RecordedCameraFrame]
+
+RESULT_ID: Final = re.compile(rf"^{re.escape(M48R3_SHADOW_PREFIX)}[a-f0-9]{{64}}$")
+RESULT_VIEW_SCHEMA: Final = "missioncore.m48r3-static-occupancy-shadow-view/v1"
+RESULT_CATALOG_SCHEMA: Final = "missioncore.m48r3-static-occupancy-shadow-catalog/v1"
+CASE_CATALOG_SCHEMA: Final = "missioncore.m48r3-static-occupancy-shadow-cases/v1"
+ENDPOINT_ROOT: Final = "/api/v1/laboratory/m48r3/static-occupancy"
+
+
+def build_m48r3_static_occupancy_router(
+    *,
+    root_provider: RootProvider = lambda: None,
+    repository_root_provider: RootProvider = lambda: None,
+    camera_frame_provider: CameraFrameProvider | None = None,
+) -> APIRouter:
+    router = APIRouter(prefix=ENDPOINT_ROOT, tags=["laboratory"])
+
+    def result(result_id: str) -> M48R3StaticOccupancyShadowResult:
+        candidate = _resolve_candidate(root_provider, result_id)
+        try:
+            return _read_result_cached(str(candidate), _result_signature(candidate))
+        except (M48R3StaticOccupancyShadowError, OSError, ValueError):
+            raise HTTPException(status_code=404, detail="M4.8R3 result not found") from None
+
+    def timeline(result_id: str) -> M48sReplayTimeline:
+        candidate = _resolve_candidate(root_provider, result_id)
+        repository = _configured_root(repository_root_provider)
+        if repository is None:
+            raise HTTPException(status_code=503, detail="M4.8R3 timeline source unavailable")
+        result(result_id)
+        try:
+            return _read_timeline_cached(
+                str(repository),
+                str(candidate),
+                result_id,
+                _timeline_signature(candidate),
+            )
+        except (M48sReplayTimelineError, OSError, ValueError):
+            raise HTTPException(
+                status_code=503,
+                detail="M4.8R3 bounded timeline failed verification",
+            ) from None
+
+    @router.get("/results")
+    def list_results(limit: int = Query(default=1, ge=1, le=10)) -> dict[str, object]:
+        root = _configured_root(root_provider)
+        if root is None:
+            return _empty_catalog(configured=False)
+        items: list[dict[str, object]] = []
+        invalid_total = 0
+        for candidate in sorted(root.iterdir()):
+            if not candidate.is_dir() or RESULT_ID.fullmatch(candidate.name) is None:
+                continue
+            try:
+                sealed = _read_result_cached(str(candidate), _result_signature(candidate))
+                items.append(_project_result(sealed))
+            except (M48R3StaticOccupancyShadowError, OSError, ValueError):
+                invalid_total += 1
+        items.sort(
+            key=lambda item: (str(item["created_at_utc"]), str(item["result_id"])),
+            reverse=True,
+        )
+        return {
+            "schema_version": RESULT_CATALOG_SCHEMA,
+            "configured": True,
+            "items": items[:limit],
+            "candidate_total": len(items) + invalid_total,
+            "invalid_total": invalid_total,
+            "access": "read-only",
+        }
+
+    @router.get("/{result_id}")
+    def get_result(result_id: str) -> dict[str, object]:
+        return _project_result(result(result_id))
+
+    @router.get("/{result_id}/cases")
+    def get_cases(result_id: str) -> dict[str, object]:
+        sealed = result(result_id)
+        return {
+            "schema_version": CASE_CATALOG_SCHEMA,
+            "result_id": result_id,
+            "cases": copy.deepcopy(sealed.cases),
+            "case_count": len(sealed.cases),
+            "ground_truth": False,
+            "authority": "replay-simulated",
+            "access": "read-only",
+        }
+
+    @router.get("/{result_id}/timeline")
+    def get_timeline(result_id: str) -> dict[str, object]:
+        return copy.deepcopy(timeline(result_id).metadata())
+
+    @router.get("/{result_id}/timeline/chunk")
+    def get_timeline_chunk(
+        result_id: str,
+        start: int = Query(default=0, ge=0),
+        count: int = Query(default=12, ge=1, le=RECORDED_SPATIAL_MAX_CHUNK_FRAMES),
+    ) -> Response:
+        try:
+            content = timeline(result_id).chunk_json(
+                start_sequence=start,
+                frame_count=count,
+            )
+        except M48sReplayTimelineError:
+            raise HTTPException(status_code=404, detail="M4.8R3 chunk not found") from None
+        return _immutable_json(content)
+
+    @router.get("/{result_id}/timeline/frames/{sequence}/camera-points")
+    def get_camera_points(result_id: str, sequence: int) -> Response:
+        try:
+            content = timeline(result_id).camera_point_overlay_json(sequence=sequence)
+        except M48sReplayTimelineError:
+            raise HTTPException(
+                status_code=404,
+                detail="M4.8R3 camera points not found",
+            ) from None
+        return _immutable_json(content)
+
+    @router.get("/{result_id}/timeline/frames/{sequence}/camera")
+    def get_camera(result_id: str, sequence: int) -> Response:
+        if camera_frame_provider is None:
+            raise HTTPException(status_code=503, detail="M4.8R3 camera decoder unavailable")
+        projected = timeline(result_id)
+        if not 0 <= sequence < len(projected.source_times_ns):
+            raise HTTPException(status_code=404, detail="M4.8R3 frame not found")
+        try:
+            camera = camera_frame_provider(projected.profile.session_id, sequence)
+        except (OSError, SessionIntegrityError, ValueError):
+            raise HTTPException(status_code=503, detail="M4.8R3 camera unavailable") from None
+        if camera.width != 800 or camera.height != 600:
+            raise HTTPException(status_code=503, detail="M4.8R3 camera size changed")
+        return Response(
+            content=camera.payload,
+            media_type=camera.media_type,
+            headers={
+                "Cache-Control": "private, max-age=31536000, immutable",
+                "ETag": f'"{camera.sha256}"',
+                "X-Content-Type-Options": "nosniff",
+            },
+        )
+
+    return router
+
+
+@lru_cache(maxsize=2)
+def _read_result_cached(
+    result_root: str,
+    signature: tuple[int, ...],
+) -> M48R3StaticOccupancyShadowResult:
+    del signature
+    return read_m48r3_static_occupancy_shadow(Path(result_root))
+
+
+@lru_cache(maxsize=2)
+def _read_timeline_cached(
+    repository_root: str,
+    result_root: str,
+    result_id: str,
+    signature: tuple[int, ...],
+) -> M48sReplayTimeline:
+    del signature
+    return M48sReplayTimeline(
+        repository_root=Path(repository_root),
+        result_root=Path(result_root),
+        result_id=result_id,
+        frames_name="frames.jsonl",
+        worker_result_name="worker-result.json",
+        frame_evidence_schema=M48R3_FRAME_EVIDENCE_SCHEMA,
+        frame_diff_name="frame-diff.jsonl",
+        camera_endpoint_root=ENDPOINT_ROOT,
+    )
+
+
+def _project_result(result: M48R3StaticOccupancyShadowResult) -> dict[str, object]:
+    return {
+        "schema_version": RESULT_VIEW_SCHEMA,
+        "result_id": result.result_id,
+        "created_at_utc": result.manifest["created_at_utc"],
+        "accepted": result.manifest["accepted"],
+        "profile": copy.deepcopy(result.report["profile"]),
+        "metrics": copy.deepcopy(result.report["metrics"]),
+        "gates": copy.deepcopy(result.report["gates"]),
+        "decision": copy.deepcopy(result.report["decision"]),
+        "limitations": copy.deepcopy(result.report["limitations"]),
+        "ground_truth": False,
+        "authority": copy.deepcopy(result.report["authority"]),
+        "access": "read-only",
+    }
+
+
+def _resolve_candidate(provider: RootProvider, result_id: str) -> Path:
+    if RESULT_ID.fullmatch(result_id) is None:
+        raise HTTPException(status_code=404, detail="M4.8R3 result not found")
+    root = _configured_root(provider)
+    if root is None:
+        raise HTTPException(status_code=404, detail="M4.8R3 result not found")
+    candidate = root / result_id
+    if candidate.is_symlink() or not candidate.is_dir():
+        raise HTTPException(status_code=404, detail="M4.8R3 result not found")
+    resolved = candidate.resolve(strict=True)
+    if resolved.parent != root:
+        raise HTTPException(status_code=404, detail="M4.8R3 result not found")
+    return resolved
+
+
+def _configured_root(provider: RootProvider) -> Path | None:
+    value = provider()
+    if value is None:
+        return None
+    if value.is_symlink() or not value.is_dir():
+        return None
+    return value.resolve(strict=True)
+
+
+def _result_signature(candidate: Path) -> tuple[int, ...]:
+    return _signature(
+        candidate,
+        ("manifest.json", "report.json", "cases.jsonl", "worker-result.json", "frames.jsonl"),
+    )
+
+
+def _timeline_signature(candidate: Path) -> tuple[int, ...]:
+    return _signature(
+        candidate,
+        ("worker-result.json", "frames.jsonl", "frame-diff.jsonl"),
+    )
+
+
+def _signature(candidate: Path, names: tuple[str, ...]) -> tuple[int, ...]:
+    result: list[int] = []
+    for name in names:
+        path = candidate / name
+        if path.is_symlink() or not path.is_file():
+            raise ValueError("M4.8R3 artifact unavailable")
+        stat = path.stat()
+        result.extend((stat.st_size, stat.st_mtime_ns))
+    return tuple(result)
+
+
+def _immutable_json(content: bytes) -> Response:
+    return Response(
+        content=content,
+        media_type="application/json",
+        headers={
+            "Cache-Control": "private, max-age=31536000, immutable",
+            "X-Content-Type-Options": "nosniff",
+        },
+    )
+
+
+def _empty_catalog(*, configured: bool) -> dict[str, object]:
+    return {
+        "schema_version": RESULT_CATALOG_SCHEMA,
+        "configured": configured,
+        "items": [],
+        "candidate_total": 0,
+        "invalid_total": 0,
+        "access": "read-only",
+    }
+
+
+__all__ = ["build_m48r3_static_occupancy_router"]

tests/test_m48_low_step_occupancy.py

@@ -23,10 +23,7 @@ from k1link.perception.providers import SourcePacket
 from k1link.perception.recorded_source import RecordedFrameReference
 
 REPOSITORY_ROOT = Path(__file__).resolve().parents[1]
-PROFILE_PATH = (
+PROFILE_PATH = REPOSITORY_ROOT / "config/perception/m48r3-additive-low-step-occupancy-v1.json"
-    REPOSITORY_ROOT
-    / "config/perception/m48r3-additive-low-step-occupancy-v1.json"
-)
 R2_CASES_PATH = (
     REPOSITORY_ROOT
     / ".runtime/compute-experiments/m48/static-occupancy-qualification-results"
@@ -126,9 +123,7 @@ def test_wide_operator_region_cannot_bridge_two_spatial_components() -> None:
 
     observations = provider.associate(_packet(), ())
     additive = tuple(
-        item
+        item for item in observations if "additive-low-step-current-component" in item.reason_codes
-        for item in observations
-        if "additive-low-step-current-component" in item.reason_codes
     )
 
     assert len(additive) == 2
@@ -144,6 +139,42 @@ def test_wide_operator_region_cannot_bridge_two_spatial_components() -> None:
     assert snapshot.failed_frames == 0
 
 
+def test_sparse_component_requires_bounded_causal_persistence() -> None:
+    points = np.asarray(
+        ((0.00, 0.0, 5.00), (0.04, 0.0, 5.00)),
+        dtype=np.float64,
+    )
+    store = _Store(_frame(points), np.ones(2, dtype=np.uint8))
+    provider = M48AdditiveLowStepGeometryProvider(  # type: ignore[arg-type]
+        store=store,
+        profile=load_m48_low_step_occupancy_profile(PROFILE_PATH),
+    )
+
+    first_five = tuple(provider.associate(_packet(sequence), ()) for sequence in range(5))
+    sixth = provider.associate(_packet(5), ())
+
+    assert first_five == ((), (), (), (), ())
+    assert len(sixth) == 1
+    assert sixth[0].source_point_ids == (0, 1)
+    assert "additive-low-step-current-component" in sixth[0].reason_codes
+
+
+def test_persistent_sparse_component_is_bounded_to_critical_range() -> None:
+    points = np.asarray(
+        ((0.00, 0.0, 8.20), (0.04, 0.0, 8.20)),
+        dtype=np.float64,
+    )
+    store = _Store(_frame(points), np.ones(2, dtype=np.uint8))
+    provider = M48AdditiveLowStepGeometryProvider(  # type: ignore[arg-type]
+        store=store,
+        profile=load_m48_low_step_occupancy_profile(PROFILE_PATH),
+    )
+
+    observations = tuple(provider.associate(_packet(sequence), ()) for sequence in range(6))
+
+    assert observations == ((), (), (), (), (), ())
+
+
 def test_frame_1856_preserves_baseline_posts_and_splits_low_hemisphere_support() -> None:
     store = RecordedGeometryStore.from_repository(REPOSITORY_ROOT)
     provider = M48AdditiveLowStepGeometryProvider(
@@ -163,8 +194,7 @@ def test_frame_1856_preserves_baseline_posts_and_splits_low_hemisphere_support()
         profile=frame.projection,
     )
     source_rows = {
-        int(source_index): row
+        int(source_index): row for row, source_index in enumerate(projected.source_indices)
-        for row, source_index in enumerate(projected.source_indices)
     }
     cases = [
         json.loads(line)
@@ -176,14 +206,62 @@ def test_frame_1856_preserves_baseline_posts_and_splits_low_hemisphere_support()
     hemispheres = by_anchor["anchor-924a4623077fe5df18816b47"]
 
     assert posts["accepted_graph"]["component_count"] >= 2
-    assert _component_hits(
+    assert (
+        _component_hits(
             observations,
             hemispheres["extent_xyxy"],
             projected.pixels_xy,
             source_rows,
             width=frame.projection.width,
             height=frame.projection.height,
-    ) >= 2
+        )
+        >= 2
+    )
+
+
+def test_exact_six_frame_persistence_recovers_critical_near_anchor() -> None:
+    store = RecordedGeometryStore.from_repository(REPOSITORY_ROOT)
+    provider = M48AdditiveLowStepGeometryProvider(
+        store=store,
+        profile=load_m48_low_step_occupancy_profile(PROFILE_PATH),
+    )
+    observations: tuple[ObstacleObservation, ...] = ()
+    for source_sequence in range(1084, 1093):
+        observations = provider.associate(_packet(source_sequence), ())
+
+    evidence_frame = store.frame_for_index(1092)
+    target_frame = store.frame_for_index(1093)
+    assert evidence_frame is not None
+    assert target_frame is not None
+    projected = project_map_points_kb4(
+        evidence_frame.points_map,
+        position_map_xyz=target_frame.sensor_position_map,
+        orientation_map_from_lidar_xyzw=target_frame.sensor_orientation_xyzw,
+        profile=target_frame.projection,
+    )
+    source_rows = {
+        int(source_index): row for row, source_index in enumerate(projected.source_indices)
+    }
+    critical = next(
+        json.loads(line)
+        for line in R2_CASES_PATH.read_text("utf-8").splitlines()
+        if "anchor-0df056d9c565b74a25d3cca3" in line
+    )
+    additive = tuple(
+        item for item in observations if "additive-low-step-current-component" in item.reason_codes
+    )
+
+    assert (
+        _component_hits(
+            additive,
+            critical["extent_xyxy"],
+            projected.pixels_xy,
+            source_rows,
+            width=target_frame.projection.width,
+            height=target_frame.projection.height,
+        )
+        >= 1
+    )
 
 
 def _component_hits(
@@ -207,9 +285,6 @@ def _component_hits(
         if not indices:
             continue
         rows = [source_rows[index] for index in indices if index in source_rows]
-        if any(
+        if any(x1 <= pixels[row, 0] <= x2 and y1 <= pixels[row, 1] <= y2 for row in rows):
-            x1 <= pixels[row, 0] <= x2 and y1 <= pixels[row, 1] <= y2
-            for row in rows
-        ):
             count += 1
     return count

tests/test_m48r3_static_occupancy_api.py

@@ -0,0 +1,105 @@
+from __future__ import annotations
+
+import json
+from pathlib import Path
+from types import SimpleNamespace
+
+from fastapi import FastAPI
+from fastapi.testclient import TestClient
+
+from k1link.laboratory.m48r3_static_occupancy_shadow import (
+    M48R3StaticOccupancyShadowResult,
+)
+from k1link.web import m48r3_static_occupancy_api as api
+
+
+def test_m48r3_api_projects_result_cases_and_provenance_timeline(
+    tmp_path: Path,
+    monkeypatch,
+) -> None:
+    result_id = "m48r3-static-occupancy-shadow-" + "0" * 64
+    root = tmp_path / "results"
+    result_root = root / result_id
+    result_root.mkdir(parents=True)
+    for name in (
+        "manifest.json",
+        "report.json",
+        "cases.jsonl",
+        "worker-result.json",
+        "frames.jsonl",
+        "frame-diff.jsonl",
+    ):
+        (result_root / name).write_text("{}\n", encoding="utf-8")
+    sealed = M48R3StaticOccupancyShadowResult(
+        result_id=result_id,
+        result_root=result_root,
+        manifest={"created_at_utc": "2026-08-26T00:00:00Z", "accepted": True},
+        report={
+            "profile": {"id": "profile"},
+            "metrics": {"frames": {"candidate_delivered": 4489}},
+            "gates": {"complete_frame_accounting": True},
+            "decision": {"state": "accepted-bounded-worker-shadow"},
+            "limitations": ["replay only"],
+            "authority": {"mode": "replay-simulated"},
+        },
+        cases=({"anchor_id": "anchor", "source_sequence": 1855},),
+    )
+
+    class Timeline:
+        source_times_ns = tuple(range(4489))
+        profile = SimpleNamespace(session_id="20260720T065719Z_viewer_live")
+
+        def metadata(self) -> dict[str, object]:
+            return {
+                "schema_version": "missioncore.recorded-spatial-evidence-timeline/v1",
+                "result_id": result_id,
+                "occupancy_provenance_delivery": (
+                    "baseline-versus-additive-component-diff"
+                ),
+            }
+
+        def chunk_json(self, *, start_sequence: int, frame_count: int) -> bytes:
+            return json.dumps(
+                {"start_sequence": start_sequence, "frame_count": frame_count}
+            ).encode()
+
+        def camera_point_overlay_json(self, *, sequence: int) -> bytes:
+            return json.dumps({"sequence": sequence}).encode()
+
+    monkeypatch.setattr(api, "_read_result_cached", lambda *_args: sealed)
+    monkeypatch.setattr(api, "_read_timeline_cached", lambda *_args: Timeline())
+    app = FastAPI()
+    app.include_router(
+        api.build_m48r3_static_occupancy_router(
+            root_provider=lambda: root,
+            repository_root_provider=lambda: tmp_path,
+        )
+    )
+    client = TestClient(app)
+
+    catalog = client.get("/api/v1/laboratory/m48r3/static-occupancy/results")
+    assert catalog.status_code == 200
+    assert catalog.json()["items"][0]["result_id"] == result_id
+    result = client.get(f"/api/v1/laboratory/m48r3/static-occupancy/{result_id}")
+    assert result.status_code == 200
+    assert result.json()["accepted"] is True
+    cases = client.get(f"/api/v1/laboratory/m48r3/static-occupancy/{result_id}/cases")
+    assert cases.status_code == 200
+    assert cases.json()["cases"][0]["source_sequence"] == 1855
+    timeline = client.get(
+        f"/api/v1/laboratory/m48r3/static-occupancy/{result_id}/timeline"
+    )
+    assert timeline.status_code == 200
+    assert timeline.json()["occupancy_provenance_delivery"] == (
+        "baseline-versus-additive-component-diff"
+    )
+    chunk = client.get(
+        f"/api/v1/laboratory/m48r3/static-occupancy/{result_id}/timeline/chunk",
+        params={"start": 1855, "count": 1},
+    )
+    assert chunk.status_code == 200
+    assert chunk.json() == {"start_sequence": 1855, "frame_count": 1}
+    assert (
+        client.get("/api/v1/laboratory/m48r3/static-occupancy/not-a-result").status_code
+        == 404
+    )

tests/test_m48r3_static_occupancy_shadow.py

@@ -0,0 +1,90 @@
+from __future__ import annotations
+
+import json
+from pathlib import Path
+
+from k1link.laboratory.m48r3_static_occupancy_shadow import (
+    FRAME_EVIDENCE_SCHEMA,
+    compare_m48r3_frame_ledgers,
+)
+
+
+def _component(component_id: str, cells: list[tuple[int, int, int]]) -> dict[str, object]:
+    return {
+        "component_id": component_id,
+        "cells": [{"x": x, "y": y, "z": z} for x, y, z in cells],
+    }
+
+
+def _row(
+    sequence: int,
+    occupied: list[dict[str, object]],
+    *,
+    unknown: list[dict[str, object]] | None = None,
+) -> dict[str, object]:
+    return {
+        "schema_version": FRAME_EVIDENCE_SCHEMA,
+        "source_envelope": {"sequence": sequence},
+        "delivery": {
+            "obstacle_map": {
+                "occupied": occupied,
+                "unknown": unknown or [],
+                "free_space_claimed": False,
+            }
+        },
+    }
+
+
+def _write(path: Path, rows: list[dict[str, object]]) -> None:
+    path.write_text(
+        "".join(
+            json.dumps(row, sort_keys=True, separators=(",", ":")) + "\n"
+            for row in rows
+        ),
+        "utf-8",
+    )
+
+
+def test_streaming_diff_preserves_gaps_and_marks_additive_components(tmp_path: Path) -> None:
+    baseline = tmp_path / "baseline.jsonl"
+    candidate = tmp_path / "candidate.jsonl"
+    _write(
+        baseline,
+        [
+            _row(0, [_component("base-0", [(0, 0, 0)])]),
+            _row(1, [_component("base-1", [(10, 0, 0)])]),
+        ],
+    )
+    _write(
+        candidate,
+        [
+            _row(0, [_component("mixed-0", [(0, 0, 0), (1, 0, 0)])]),
+            _row(
+                1,
+                [_component("base-1", [(10, 0, 0)])],
+                unknown=[_component("step-1", [(20, 0, 0)])],
+            ),
+        ],
+    )
+
+    result = compare_m48r3_frame_ledgers(
+        baseline,
+        candidate,
+        selected_sequences={1},
+        expected_frames=2,
+    )
+
+    assert result.frame_count == 2
+    assert result.baseline_cell_total == 2
+    assert result.candidate_cell_total == 4
+    assert result.added_cell_total == 2
+    assert result.lost_cell_total == 0
+    assert result.mean_cell_growth_fraction == 1.0
+    assert result.mean_component_growth_fraction == 0.5
+    assert result.diff_rows[0]["component_provenance"] == {"mixed-0": "mixed"}
+    assert result.diff_rows[1]["component_provenance"] == {
+        "step-1": "additive-low-step"
+    }
+    assert result.diff_rows[0]["added_cells"] == [[1, 0, 0]]
+    assert set(result.selected_candidate_rows) == {1}
+

tests/test_simulation_projects.py

@@ -93,6 +93,7 @@ class _ReadyProvider:
         self.deleted: list[str] = []
         self.upload_calls = 0
         self.submit_calls = 0
+        self.submitted_document: dict[str, object] | None = None
 
     def capabilities(self) -> dict[str, object]:
         return {"outputs": ["preview.sog", "streamed-sog"]}
@@ -118,8 +119,9 @@ class _ReadyProvider:
             members=members,
         )
 
-    def submit_build(self, _document: dict[str, object]) -> dict[str, object]:
+    def submit_build(self, document: dict[str, object]) -> dict[str, object]:
         self.submit_calls += 1
+        self.submitted_document = document
         return {
             "schema_version": "gaussian-pipeline.job/v1",
             "job_id": "gsp-20260826000000-deadbeef",
@@ -195,8 +197,28 @@ def test_service_materializes_world_manifest_and_deletes_both_copies(tmp_path: P
     assert ready["status"] == "ready"
     assert ready["world_manifest"]["visual"]["preview_sog_url"].endswith("/preview.sog")
     assert ready["world_manifest"]["collision"]["available"] is False
+    assert ready["world_manifest"]["transforms"]["world_from_visual"] == [
+        1, 0, 0, 0, 0, -1, 0, 0, 0, 0, -1, 0, 0, 0, 0, 1,
+    ]
+    assert ready["world_manifest"]["transforms"]["world_from_collision"] == [
+        1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1,
+    ]
     assert provider.upload_calls == 1
     assert provider.submit_calls == 1
+    assert provider.submitted_document is not None
+    assert provider.submitted_document["outputs"] == {
+        "preview_sog": True,
+        "streamed_sog": True,
+        "collision": True,
+    }
+    assert provider.submitted_document["collision_profile"] == {
+        "scene_type": "interior",
+        "seed_position": [0, 1, 0],
+        "capsule_height": 1.6,
+        "capsule_radius": 0.2,
+        "voxel_size": 0.05,
+        "mesh_shape": "smooth",
+    }
     service.delete(project["project_id"])
     assert provider.deleted == ["gsp-20260826000000-deadbeef"]