feat(perception): split vegetation evidence layers

2026-08-28 15:37:10 +03:00
parent a2c3385062
commit e10b96b546
26 changed files with 868 additions and 233 deletions
@@ -51,6 +51,7 @@ import { M48TRiskQualityResultView } from "./M48TRiskQualityResult";
 import { M49TgsFailClosedResultView } from "./M49TgsFailClosedResult";
 import { M49TgsFullShadowResultView } from "./M49TgsFullShadowResult";
 import { VegetationShadowResultView } from "./VegetationShadowResult";
+import { VegetationBenchmarkResultView } from "./VegetationBenchmarkResult";

 export { isAdvancedLaboratoryWorkId };
 export type { AdvancedLaboratoryWorkId };
@@ -93,6 +94,9 @@ export function AdvancedLaboratoryResult({
  failedSessionId: string | null;
  replayError: string | null;
 }) {
+  if (workId === "lab-v1-vegetation-benchmark" && results.vegetationBenchmark) {
+    return <VegetationBenchmarkResultView rigLabel={rigLabel} result={results.vegetationBenchmark} />;
+  }
  if (workId === "lab-v1-vegetation-shadow" && results.vegetationShadow) {
    return <VegetationShadowResultView rigLabel={rigLabel} result={results.vegetationShadow} />;
  }
@@ -71,7 +71,6 @@ export function M49TgsFullShadowEvidence({
    const controller = new AbortController();
    setSemantic(null);
    setSemanticError(null);
-    if (semanticOverride) return () => controller.abort();
    void fetchE47SemanticSlamResult({
      resultId: result.source.linkedSemanticResultId,
      signal: controller.signal,
@@ -87,7 +86,7 @@ export function M49TgsFullShadowEvidence({
        if (!controller.signal.aborted) setSemanticError(message(caught));
      });
    return () => controller.abort();
-  }, [result.source.linkedSemanticResultId, result.source.linkedVisualResultId, semanticOverride]);
+  }, [result.source.linkedSemanticResultId, result.source.linkedVisualResultId]);

  useEffect(() => {
    const controller = new AbortController();
@@ -201,15 +200,28 @@ export function M49TgsFullShadowEvidence({
  const handleSequenceChange = useCallback((sequence: number | null) => {
    setActiveSequence(sequence);
  }, []);
+  const semanticLayers = useMemo<readonly M4ReplayThreatSemanticLayer[]>(() => [
+    ...(semantic ? [{
+      id: "urban",
+      controlLabel: "ГОРОД · EoMT",
+      resultId: semantic.resultId,
+      taxonomy: semantic.taxonomy,
+      label: "EoMT Cityscapes semantic · recorded video",
+      maskAriaLabel: "EoMT urban semantic prediction",
+    }] : []),
+    ...(semanticOverride ? [{
+      ...semanticOverride,
+      id: semanticOverride.id ?? "vegetation",
+      controlLabel: semanticOverride.controlLabel ?? "ПРИРОДА · DDRNet",
+    }] : []),
+  ], [semantic, semanticOverride]);

  return (
    <>
      <M4ReplayThreatVisual
        resultId={result.source.linkedVisualResultId}
-        semantic={semanticOverride ?? (semantic ? {
-          resultId: semantic.resultId,
-          taxonomy: semantic.taxonomy,
-        } : undefined)}
+        semanticLayers={semanticLayers}
+        initialSemanticLayerId={semanticOverride ? "vegetation" : "urban"}
        showReviewAnchorBoxes={false}
        reviewLabel="4 489 source-paced TGS frames"
        evidenceLabel={evidenceLabel}
@@ -227,7 +239,7 @@ export function M49TgsFullShadowEvidence({
          replacePointCloud: false,
        }}
      />
-      {!semanticOverride && semanticError ? (
+      {semanticError ? (
        <div className="m4-replay-threat-visual__pane-status" role="alert">
          Semantic overlay недоступен: {semanticError}
        </div>
@@ -96,6 +96,8 @@ function SpatialState({ message: text }: { message: string }) {
 }

 export interface M4ReplayThreatSemanticLayer {
+  id?: string;
+  controlLabel?: string;
  resultId: string;
  spatialResultId?: string | null;
  maskUrl?: (sequence: number) => string;
@@ -155,6 +157,8 @@ const EMPTY_REVIEW_ANCHORS: readonly M4ReplayThreatReviewAnchor[] = [];
 export function M4ReplayThreatVisual({
  resultId,
  semantic,
+  semanticLayers,
+  initialSemanticLayerId,
  reviewAnchors = EMPTY_REVIEW_ANCHORS,
  showReviewAnchorBoxes = true,
  reviewLabel = "Контрольные примеры M4.8R1",
@@ -168,6 +172,8 @@ export function M4ReplayThreatVisual({
 }: {
  resultId: string;
  semantic?: M4ReplayThreatSemanticLayer;
+  semanticLayers?: readonly M4ReplayThreatSemanticLayer[];
+  initialSemanticLayerId?: string;
  reviewAnchors?: readonly M4ReplayThreatReviewAnchor[];
  showReviewAnchorBoxes?: boolean;
  reviewLabel?: string;
@@ -199,6 +205,35 @@ export function M4ReplayThreatVisual({
  ));
  const [expanded, setExpanded] = useState(false);
  const [selectedReviewAnchorIndex, setSelectedReviewAnchorIndex] = useState(0);
+  const availableSemanticLayers = useMemo<readonly M4ReplayThreatSemanticLayer[]>(
+    () => semanticLayers?.length ? semanticLayers : semantic ? [semantic] : [],
+    [semantic, semanticLayers],
+  );
+  const semanticLayerIdentity = availableSemanticLayers
+    .map((layer, index) => layer.id ?? `${layer.resultId}:${index}`)
+    .join("|");
+  const [selectedSemanticLayerId, setSelectedSemanticLayerId] = useState(
+    initialSemanticLayerId ?? "",
+  );
+  useEffect(() => {
+    if (!availableSemanticLayers.length) {
+      setSelectedSemanticLayerId("");
+      return;
+    }
+    const selectedStillExists = availableSemanticLayers.some(
+      (layer, index) => (layer.id ?? `${layer.resultId}:${index}`) === selectedSemanticLayerId,
+    );
+    if (selectedStillExists) return;
+    const preferred = initialSemanticLayerId
+      ? availableSemanticLayers.find((layer) => layer.id === initialSemanticLayerId)
+      : null;
+    const next = preferred ?? availableSemanticLayers[0]!;
+    const nextIndex = availableSemanticLayers.indexOf(next);
+    setSelectedSemanticLayerId(next.id ?? `${next.resultId}:${nextIndex}`);
+  }, [availableSemanticLayers, initialSemanticLayerId, semanticLayerIdentity, selectedSemanticLayerId]);
+  const activeSemantic = availableSemanticLayers.find(
+    (layer, index) => (layer.id ?? `${layer.resultId}:${index}`) === selectedSemanticLayerId,
+  ) ?? availableSemanticLayers[0];
  const metricSceneRef = useRef<LaboratoryMetricEvidenceSceneHandle | null>(null);
  const metadata = useM4ThreatTimelineMetadata(resultId, timelineEndpointRoot);
  const playbackRange = useMemo(() => metadata.timeline ? ({
@@ -298,19 +333,21 @@ export function M4ReplayThreatVisual({
    sequence: frame?.sequence ?? null,
    endpointRoot: timelineEndpointRoot,
  });
-  const semanticSpatialResultId = semantic
-    ? semantic.spatialResultId === undefined ? semantic.resultId : semantic.spatialResultId
+  const semanticSpatialResultId = activeSemantic
+    ? activeSemantic.spatialResultId === undefined
+      ? activeSemantic.resultId
+      : activeSemantic.spatialResultId
    : null;
  const spatialSemanticTaxonomy = useMemo<readonly E47SemanticClass[]>(
-    () => semanticSpatialResultId && semantic
-      ? semantic.taxonomy.map((item) => ({
+    () => semanticSpatialResultId && activeSemantic
+      ? activeSemantic.taxonomy.map((item) => ({
          classId: item.classId,
          label: item.label,
          disposition: item.disposition === "ambiguous" ? "ambiguous" : "labeled",
          colorRgb: item.colorRgb,
        }))
      : [],
-    [semantic, semanticSpatialResultId],
+    [activeSemantic, semanticSpatialResultId],
  );
  const semanticTimeline = useE47SemanticTimelineFrame({
    resultId: semanticSpatialResultId,
@@ -386,14 +423,14 @@ export function M4ReplayThreatVisual({
    [frame, reviewAnchorBoxes, showReferenceMediaLayers, staticObstacleBoxes],
  );
  const semanticClasses = useMemo<readonly RecordedEvidenceSemanticClass[]>(
-    () => semantic?.taxonomy.map((item) => ({
+    () => activeSemantic?.taxonomy.map((item) => ({
      id: item.classId,
      label: `semantic: ${item.label}`,
    })) ?? [],
-    [semantic?.taxonomy],
+    [activeSemantic?.taxonomy],
  );
  const semanticPalette = useMemo<readonly RecordedEvidenceSemanticPaletteEntry[]>(
-    () => semantic?.taxonomy.map((item) => ({
+    () => activeSemantic?.taxonomy.map((item) => ({
      classId: item.classId,
      color: item.disposition === "undefined"
        ? { kind: "transparent" as const }
@@ -404,7 +441,7 @@ export function M4ReplayThreatVisual({
        ? 0
        : item.disposition === "ambiguous" ? 0.52 : 0.92,
    })) ?? [],
-    [semantic?.taxonomy],
+    [activeSemantic?.taxonomy],
  );
  const semanticFrame = semanticTimeline.activeFrame?.sequence === frame?.sequence
    ? semanticTimeline.activeFrame
@@ -422,7 +459,7 @@ export function M4ReplayThreatVisual({
      && lastSpatialSemanticFrameRef.current.frame.sequence === spatialFrame?.sequence
      ? lastSpatialSemanticFrameRef.current.frame
      : null;
-  const semanticIntegrityError = semantic && spatialFrame && spatialSemanticFrame && (
+  const semanticIntegrityError = activeSemantic && spatialFrame && spatialSemanticFrame && (
    spatialSemanticFrame.sourcePointCount !== spatialFrame.pointCloudSourceCount
    || spatialFrame.pointCloudSampleCount !== spatialFrame.pointCloudSourceCount
    || spatialFrame.pointCloudBodyXyzM.length !== spatialFrame.pointCloudSourceCount
@@ -431,7 +468,7 @@ export function M4ReplayThreatVisual({
    : null;
  const alignedSemanticPointIds = useMemo<readonly (number | null)[] | undefined>(() => {
    if (
-      !semantic
+      !activeSemantic
      || !showSpatialSemantic
      || !spatialFrame
      || !spatialSemanticFrame
@@ -441,7 +478,7 @@ export function M4ReplayThreatVisual({
      const status = spatialSemanticFrame.statusCodes[index];
      return status === 2 || status === 3 ? classId : null;
    });
-  }, [semantic, semanticIntegrityError, showSpatialSemantic, spatialFrame, spatialSemanticFrame]);
+  }, [activeSemantic, semanticIntegrityError, showSpatialSemantic, spatialFrame, spatialSemanticFrame]);
  const activeSpatialFrame = spatialFrame?.sequence === timelineFrame.activeSequence
    ? spatialFrame
    : null;
@@ -610,19 +647,19 @@ export function M4ReplayThreatVisual({
    },
  ), [metadata.timeline, spatialFrame, timelineFrame.availableFrames]);
  const semanticOverlay: RecordedEvidenceSemanticOverlay | undefined =
-    semantic && showMediaSemantic && frame
+    activeSemantic && showMediaSemantic && frame
      ? {
-          src: semantic.maskUrl?.(frame.sequence)
-            ?? e47SemanticMaskUrl(semantic.resultId, frame.sequence),
+          src: activeSemantic.maskUrl?.(frame.sequence)
+            ?? e47SemanticMaskUrl(activeSemantic.resultId, frame.sequence),
          prefetchSrcs: Array.from({ length: 12 }, (_, index) => index + 1)
            .map((offset) => frame.sequence + offset)
            .filter((sequence) => sequence < (metadata.timeline?.frameCount ?? 0))
-            .map((sequence) => semantic.maskUrl?.(sequence)
-              ?? e47SemanticMaskUrl(semantic.resultId, sequence)),
+            .map((sequence) => activeSemantic.maskUrl?.(sequence)
+              ?? e47SemanticMaskUrl(activeSemantic.resultId, sequence)),
          classes: semanticClasses,
          palette: semanticPalette,
          opacity: 0.9,
-          ariaLabel: `${semantic.maskAriaLabel ?? "Semantic prediction"} frame ${frame.sequence + 1}`,
+          ariaLabel: `${activeSemantic.maskAriaLabel ?? "Semantic prediction"} frame ${frame.sequence + 1}`,
        }
      : undefined;
  const accumulatedCameraPoints = cameraPointOverlay.overlay?.sequence === frame?.sequence
@@ -692,7 +729,7 @@ export function M4ReplayThreatVisual({
    </div>
  );

-  const mediaLayerControls = semantic
+  const mediaLayerControls = activeSemantic
    || (showReferenceMediaLayers && metadata.timeline?.cameraPointDelivery)
    || (showReferenceMediaLayers && metadata.timeline?.cameraObstacleProjectionDelivery) ? (
    <div
@@ -700,7 +737,7 @@ export function M4ReplayThreatVisual({
      role="group"
      aria-label="Слои камеры и видео"
    >
-      {semantic ? (
+      {activeSemantic ? (
        <Button
          size="compact"
          shape="pill"
@@ -711,6 +748,20 @@ export function M4ReplayThreatVisual({
          SEMANTICS
        </Button>
      ) : null}
+      {availableSemanticLayers.length > 1 ? (
+        <SegmentedControl
+          value={selectedSemanticLayerId}
+          items={availableSemanticLayers.map((layer, index) => ({
+            value: layer.id ?? `${layer.resultId}:${index}`,
+            label: layer.controlLabel ?? layer.label ?? `SEMANTIC ${index + 1}`,
+          }))}
+          label="Источник семантики"
+          onChange={(value) => {
+            setSelectedSemanticLayerId(value);
+            setShowMediaSemantic(true);
+          }}
+        />
+      ) : null}
      {showReferenceMediaLayers && metadata.timeline?.cameraPointDelivery ? (
        <Button
          size="compact"
@@ -1022,6 +1073,7 @@ export function M4ReplayThreatVisual({
          <div
            className="m4-replay-threat-visual__pane-toolbar"
            data-pane-toolbar="media"
+            data-multi-semantic={availableSemanticLayers.length > 1 ? "true" : undefined}
          >
            {mediaLayerControls}
            {mediaModeControls}
@@ -1238,8 +1290,8 @@ export function M4ReplayThreatVisual({
  return (
    <div className="l3-visual-audit m4-replay-threat-visual">
      <LaboratoryEvidenceViewer
-        label={semantic
-          ? semantic.label ?? "Semantic diagnostic replay"
+        label={activeSemantic
+          ? activeSemantic.label ?? "Semantic diagnostic replay"
          : `${evidenceLabel} recorded-realtime replay`}
        className="m4-replay-threat-evidence-viewer"
        mode={mediaMode ?? "none"}
@@ -0,0 +1,147 @@
+import {
+  LaboratoryEvidence,
+  LaboratoryResultSummary,
+  LaboratorySummary,
+  LaboratoryWorkTemplate,
+} from "../../components/laboratory/LaboratoryPresentation";
+import type { VegetationShadowResult } from "../../core/laboratory/vegetationShadow";
+import {
+  M48MaskComparisonVisual,
+  type M48MaskComparisonCase,
+} from "./M48FailureAtlasVisual";
+
+function decimal(value: number, digits = 1): string {
+  return value.toLocaleString("ru-RU", { maximumFractionDigits: digits });
+}
+
+const VEGETATION_LABELS: Readonly<Record<string, string>> = {
+  high_grass: "Высокая трава",
+  low_grass: "Низкая трава",
+  bush: "Куст",
+  tree_trunk: "Ствол дерева",
+  tree_crown: "Крона дерева",
+  hedge: "Живая изгородь",
+  forest: "Лесная растительность",
+  crops: "Посевы",
+};
+
+function comparisonCases(result: VegetationShadowResult): readonly M48MaskComparisonCase[] {
+  return result.validationCases.map((item) => {
+    const focus = item.focus!;
+    return {
+      caseId: item.caseId,
+      title: `${VEGETATION_LABELS[focus.className] ?? focus.className} · truth ${decimal(focus.truthFraction * 100, 1)}% кадра`,
+      sourceUrl: item.assets.source,
+      truthUrl: item.assets.truth,
+      predictions: {
+        ddrnet: item.assets.ddrnet,
+        ppliteseg: item.assets.ppliteseg,
+      },
+      errors: {
+        ddrnet: item.assets.ddrnet_error,
+        ppliteseg: item.assets.ppliteseg_error,
+      },
+    };
+  });
+}
+
+export function VegetationBenchmarkResultView({
+  rigLabel,
+  result,
+}: {
+  rigLabel: string;
+  result: VegetationShadowResult;
+}) {
+  const selected = result.candidates.find(
+    (candidate) => candidate.candidate === result.selectedCandidate,
+  )!;
+  const alternative = result.candidates.find(
+    (candidate) => candidate.candidate !== result.selectedCandidate,
+  )!;
+
+  return (
+    <LaboratoryWorkTemplate
+      summary={(
+        <LaboratorySummary
+          title="M4.8 · архивный benchmark растительности"
+          description="Отдельный truth-backed контур GOOSE для сравнения готовых fine-64 весов. Он не является частью RAVNOVES00 realtime LAB и открывается автономно без Worker 006."
+          status="ARCHIVE ANALYSIS · model qualification only · commands OFF"
+          statusTone="warning"
+          facts={[
+            { label: "Источник", value: "GOOSE validation · 962 размеченных кадра · 12 hard cases" },
+            { label: "Сравнение", value: "DDRNet-39 vs PPLiteSeg · official fine-64 weights" },
+            { label: "Кейсы", value: "трава · куст · ствол · крона · изгородь · лес · посевы" },
+            { label: "Authority", value: `${rigLabel} · MODEL QUALIFICATION ONLY · commands OFF` },
+          ]}
+          brief={{
+            question: "Какие готовые веса лучше различают проезжаемую траву, кусты и стволы на размеченных off-road кадрах?",
+            approach: "Обе модели прогнаны на 962 кадрах, а 12 визуальных кейсов выбраны детерминированно по truth-поддержке восьми растительных классов. Viewer показывает source, ручной truth, prediction и error.",
+            principalResult: `${selected.loadedModelName} лидирует по vegetation IoU: ${decimal(selected.vegetationMeanIouPercent, 2)}% против ${decimal(alternative.vegetationMeanIouPercent, 2)}%.`,
+            limitation: "GOOSE — внешний размеченный домен. Результат выбирает стартовые веса, но не доказывает качество на fisheye RAVNOVES00 и не даёт navigation authority.",
+          }}
+          method={{
+            completeness: "complete",
+            executionClass: "ai-inference",
+            pipelineId: "goose-fine64-ready-weights-benchmark-archive/v1",
+            components: result.candidates.map((candidate) => ({
+              kind: "model" as const,
+              name: candidate.loadedModelName,
+              version: candidate.candidate,
+              role: candidate.candidate === result.selectedCandidate
+                ? "selected vegetation candidate"
+                : "comparison candidate",
+              identitySha256: candidate.checkpointSha256,
+            })),
+          }}
+        />
+      )}
+      evidence={(
+        <LaboratoryEvidence
+          eyebrow="M4.8 · GOOSE VEGETATION HARD CASES"
+          title="TRUTH — ручная разметка · PREDICTION — ответ модели · ERROR — расхождение"
+          kind="diagnostic-model"
+          resizable
+        >
+          <M48MaskComparisonVisual
+            cases={comparisonCases(result)}
+            initialCandidate={result.selectedCandidate}
+          />
+        </LaboratoryEvidence>
+      )}
+      result={(
+        <LaboratoryResultSummary
+          title="DDRNet выбран как стартовый vegetation candidate"
+          status={`${selected.loadedModelName} · перенос на ровер не доказан`}
+          statusTone="warning"
+          metrics={[
+            {
+              label: "GOOSE mIoU",
+              value: `${decimal(selected.meanIouPercent, 2)}% / ${decimal(alternative.meanIouPercent, 2)}%`,
+              hint: `${selected.candidate} / ${alternative.candidate} · полный validation split`,
+            },
+            {
+              label: "Vegetation IoU",
+              value: `${decimal(selected.vegetationMeanIouPercent, 2)}% / ${decimal(alternative.vegetationMeanIouPercent, 2)}%`,
+              hint: "grass/vegetation/bush/tree и родственные fine-64 labels",
+            },
+            {
+              label: "Worker shadow p95",
+              value: `${decimal(selected.shadowLatencyP95Ms, 2)} / ${decimal(alternative.shadowLatencyP95Ms, 2)} ms`,
+              hint: "чистый inference · одна тяжёлая модель за раз",
+            },
+            {
+              label: "Peak VRAM",
+              value: `${decimal(selected.peakReservedVramBytes / 1024 ** 3, 2)} / ${decimal(alternative.peakReservedVramBytes / 1024 ** 3, 2)} GiB`,
+              hint: `${selected.candidate} / ${alternative.candidate} · RTX 4090`,
+            },
+          ]}
+          conclusion={{
+            proved: "Обе готовые fine-64 модели воспроизводимо запускаются; DDRNet лучше по aggregate vegetation IoU.",
+            notProved: "Не доказаны accuracy на нашем fisheye, temporal stability, collision safety и физическое поведение ровера.",
+            decision: "Хранить как архив квалификации весов. Проверку на RAVNOVES00 вести только в основной многослойной LAB.",
+          }}
+        />
+      )}
+    />
+  );
+}
@@ -14,10 +14,6 @@ import {
  fetchM49TgsFullShadowResult,
  type M49TgsFullShadowResult,
 } from "../../core/laboratory/m49TgsFullShadow";
-import {
-  M48MaskComparisonVisual,
-  type M48MaskComparisonCase,
-} from "./M48FailureAtlasVisual";
 import { M4ReplayThreatVisual } from "./M4ReplayThreatVisual";
 import { M49TgsFullShadowEvidence } from "./M49TgsFullShadowEvidence";

@@ -25,37 +21,6 @@ function decimal(value: number, digits = 1): string {
  return value.toLocaleString("ru-RU", { maximumFractionDigits: digits });
 }

-const VEGETATION_LABELS: Readonly<Record<string, string>> = {
-  high_grass: "Высокая трава",
-  low_grass: "Низкая трава",
-  bush: "Куст",
-  tree_trunk: "Ствол дерева",
-  tree_crown: "Крона дерева",
-  hedge: "Живая изгородь",
-  forest: "Лесная растительность",
-  crops: "Посевы",
-};
-
-function comparisonCases(result: VegetationShadowResult): readonly M48MaskComparisonCase[] {
-  return result.validationCases.map((item) => {
-    const focus = item.focus!;
-    return {
-      caseId: item.caseId,
-      title: `${VEGETATION_LABELS[focus.className] ?? focus.className} · truth ${decimal(focus.truthFraction * 100, 1)}% кадра`,
-      sourceUrl: item.assets.source,
-      truthUrl: item.assets.truth,
-      predictions: {
-        ddrnet: item.assets.ddrnet,
-        ppliteseg: item.assets.ppliteseg,
-      },
-      errors: {
-        ddrnet: item.assets.ddrnet_error,
-        ppliteseg: item.assets.ppliteseg_error,
-      },
-    };
-  });
-}
-
 function VegetationRouteEvidence({ result }: { result: VegetationShadowResult }) {
  const route = result.routeVideo!;
  const [tgs, setTgs] = useState<M49TgsFullShadowResult | null>(null);
@@ -83,16 +48,14 @@ function VegetationRouteEvidence({ result }: { result: VegetationShadowResult })
  }, [route.baseM4ResultId, route.linkedTgsResultId]);

  const semantic = {
+    id: "vegetation",
+    controlLabel: "ПРИРОДА · DDRNet",
    resultId: route.workerResultId,
    spatialResultId: null,
    taxonomy: route.taxonomy,
    maskUrl: (sequence: number) => vegetationVideoMaskUrl(result.resultId, sequence),
-    label: route.viewKind === "coarse-material-policy-review"
-      ? "Coarse material evidence · recorded video"
-      : "DDRNet vegetation prediction · recorded video",
-    maskAriaLabel: route.viewKind === "coarse-material-policy-review"
-      ? "Coarse material policy evidence"
-      : "DDRNet vegetation prediction",
+    label: "DDRNet coarse vegetation material · recorded video",
+    maskAriaLabel: "DDRNet vegetation material prediction",
  } as const;

  if (route.linkedTgsResultId && tgs) {
@@ -100,19 +63,23 @@ function VegetationRouteEvidence({ result }: { result: VegetationShadowResult })
      <M49TgsFullShadowEvidence
        result={tgs}
        semanticOverride={semantic}
-        evidenceLabel="LAB V1 · MATERIAL + YOLOX + TGS"
+        evidenceLabel="LAB V1 · EoMT + DDRNet + YOLOX + TGS"
      />
    );
  }
  if (route.linkedTgsResultId && !tgsError) {
-    return <div className="m4-replay-threat-visual__pane-status" role="status">Открываем sealed TGS и coarse material timeline…</div>;
+    return (
+      <div className="m4-replay-threat-visual__pane-status" role="status">
+        Открываем sealed EoMT, TGS и coarse vegetation timeline…
+      </div>
+    );
  }
  return (
    <>
      <M4ReplayThreatVisual
        resultId={route.baseM4ResultId}
        evidenceLabel="LAB V1 · DDRNet"
-        showReferenceMediaLayers={route.viewKind === "coarse-material-policy-review"}
+        showReferenceMediaLayers
        showSpatialOverlaySummary={false}
        semantic={semantic}
      />
@@ -132,146 +99,117 @@ export function VegetationShadowResultView({
  rigLabel: string;
  result: VegetationShadowResult;
 }) {
+  const route = result.routeVideo;
  const selected = result.candidates.find(
    (candidate) => candidate.candidate === result.selectedCandidate,
  )!;
-  const alternative = result.candidates.find(
-    (candidate) => candidate.candidate !== result.selectedCandidate,
-  )!;
+
  return (
    <LaboratoryWorkTemplate
      summary={(
        <LaboratorySummary
-          title="LAB V1 · готовые модели растительности"
-          description={result.routeVideo
-            ? result.routeVideo.viewKind === "coarse-material-policy-review"
-              ? "M4.8 сохраняет truth-backed сравнение моделей, а штатный M4.7 синхронно показывает coarse material evidence, frozen YOLOX vetoes и causal TGS на всей записи RAVNOVES00. Все слои запечатаны локально и открываются без Worker 006."
-              : "M4.8 сохраняет truth-backed сравнение моделей, а штатный M4.7 viewer показывает фактический DDRNet prediction на всей записи RAVNOVES00. Все 4489 масок запечатаны локально и открываются без Worker 006."
-            : "Штатный M4.8-инструмент сравнивает две готовые fine-64 модели на полном GOOSE validation split и на 12 truth-backed hard cases, выбранных только по наличию нужной растительности. Sealed evidence открывается локально без Worker 006."}
-          status={result.routeVideo
-            ? result.routeVideo.viewKind === "coarse-material-policy-review"
-              ? "MULTILAYER POLICY REVIEW · commands OFF · route truth отсутствует"
-              : "DDRNet full-video prediction ready · route truth отсутствует"
-            : "Truth-backed model comparison · route transfer не принят"}
+          title="LAB V1 · карта ровера · город + растительность"
+          description="Один recorded-контур RAVNOVES00 синхронно показывает городской EoMT, природный DDRNet, frozen YOLOX detections и causal TGS. Семантические маски переключаются, чтобы их цвета не скрывали друг друга; геометрическое veto остаётся независимым."
+          status={route
+            ? "MULTILAYER RECORDED REVIEW · commands OFF · route truth отсутствует"
+            : "ROUTE EVIDENCE MISSING · commands OFF"}
          statusTone="warning"
          facts={[
-            { label: "Источник", value: "GOOSE validation · 962 размеченных кадра · 12 vegetation hard cases" },
-            { label: "Сравнение", value: "DDRNet-39 vs PPLiteSeg · official fine-64 weights" },
-            { label: "Кейсы", value: "трава · куст · ствол · крона · изгородь · лес · посевы" },
-            ...(result.routeVideo ? [{
-              label: "Видео",
-              value: result.routeVideo.viewKind === "coarse-material-policy-review"
-                ? "RAVNOVES00 · 4489/4489 coarse masks + YOLOX + TGS · exact sequence"
-                : "RAVNOVES00 · 4489/4489 DDRNet masks · exact recorded sequence",
-            }] : []),
-            { label: "Authority", value: `${rigLabel} · MODEL QUALIFICATION ONLY · commands OFF` },
+            { label: "Источник", value: "RAVNOVES00 · sensor.camera.right · 4489 recorded frames" },
+            { label: "Город", value: "EoMT Cityscapes · sealed E47 semantic archive" },
+            { label: "Растительность", value: "DDRNet-39 fine-64 → coarse mission-neutral materials" },
+            { label: "Safety", value: "YOLOX object boxes + causal TGS · semantic masks не снимают veto" },
+            { label: "Authority", value: `${rigLabel} · VISUAL REVIEW ONLY · commands OFF` },
          ]}
          brief={{
-            question: "Какие готовые веса лучше различают проезжаемую траву, кусты и стволы на размеченных off-road кадрах?",
-            approach: "Обе модели последовательно прогнаны в одном изолированном CUDA-runtime на 962 кадрах. 12 визуальных кейсов выбраны детерминированно по truth-поддержке восьми растительных классов; один M4.8 viewer показывает source, truth, prediction и material-error для выбранной модели.",
-            principalResult: `${selected.loadedModelName} лидирует по vegetation IoU: ${decimal(selected.vegetationMeanIouPercent, 2)}% против ${decimal(alternative.vegetationMeanIouPercent, 2)}%. ${result.routeVideo?.viewKind === "coarse-material-policy-review" ? "Fine-64 prediction сведён к mission-neutral материалам; YOLOX и TGS сохраняют независимое veto." : result.routeVideo ? "Его фактическая temporal stability теперь видна на всех 4489 кадрах штатного recorded viewer." : "Ошибки по каждому типу проверяются в одном штатном инструменте."}`,
-            limitation: "GOOSE — внешний размеченный домен; RAVNOVES00 — наш fisheye, но без ручной truth-разметки. Материалы — prediction, а не доказательство проходимости. TGS не проецируется в пиксели без отдельной принятой калибровки.",
+            question: "Можно ли одновременно видеть городской и природный semantic stack, не теряя независимую геометрическую защиту?",
+            approach: "EoMT и DDRNet сохранены как два независимых sealed слоя на одной M4 timeline. В штатном M4.7 viewer пользователь переключает только отображаемую маску; YOLOX и TGS остаются активными слоями evidence.",
+            principalResult: route
+              ? "Оба semantic archive доступны в одном viewer. Это не пиксельный fusion и не единая новая модель: городской и природный ответы остаются раздельными."
+              : "Route archive для этой immutable identity отсутствует.",
+            limitation: "RAVNOVES00 не имеет ручной truth. DDRNet заметно прыгает между HIGH GRASS, WOODY и UNKNOWN; поэтому subtype нельзя подавать напрямую в planner. Отсутствие класса никогда не означает свободный путь.",
          }}
          method={{
-            completeness: "complete",
+            completeness: route ? "complete" : "legacy-partial",
            executionClass: "ai-inference",
-            pipelineId: "goose-fine64-ready-weights-to-ravnoves-policy-shadow/v1",
-            components: result.candidates.map((candidate) => ({
-              kind: "model" as const,
-              name: candidate.loadedModelName,
-              version: candidate.candidate,
-              role: candidate.candidate === result.selectedCandidate ? "selected policy provider" : "comparison candidate",
-              identitySha256: candidate.checkpointSha256,
-            })),
+            pipelineId: "ravnoves-eomt-ddrnet-yolox-causal-tgs-recorded-review/v1",
+            components: [
+              {
+                kind: "model",
+                name: "EoMT Cityscapes semantic",
+                version: "sealed E47 archive",
+                role: "urban semantic review",
+                identitySha256: null,
+              },
+              {
+                kind: "model",
+                name: selected.loadedModelName,
+                version: selected.candidate,
+                role: "vegetation material candidate",
+                identitySha256: selected.checkpointSha256,
+              },
+              {
+                kind: "algorithm",
+                name: "Frozen YOLOX + causal TGS",
+                version: "linked M4/M4.9 archives",
+                role: "independent object and geometry veto",
+                identitySha256: null,
+              },
+            ],
          }}
        />
      )}
-      evidence={(
-        <>
-          <LaboratoryEvidence
-            eyebrow="M4.8 · GOOSE VEGETATION HARD CASES"
-            title="ERROR: красный — пропуск · жёлтый — лишнее · фиолетовый — перепутан тип · зелёный — совпадение"
-            kind="diagnostic-model"
-            resizable
-          >
-            <M48MaskComparisonVisual
-              cases={comparisonCases(result)}
-              initialCandidate={result.selectedCandidate}
-            />
-          </LaboratoryEvidence>
-          {result.routeVideo ? (
-            <LaboratoryEvidence
-              eyebrow="M4.7 · RAVNOVES00 FULL VIDEO"
-              title={result.routeVideo.viewKind === "coarse-material-policy-review"
-                ? "COARSE MATERIAL + YOLOX VETO + CAUSAL TGS · 4489/4489 · TRUTH отсутствует"
-                : "DDRNet PREDICTION · 4489/4489 кадров · TRUTH для этой записи отсутствует"}
-              kind="diagnostic-model"
-              resizable
-            >
-              <VegetationRouteEvidence result={result} />
-            </LaboratoryEvidence>
-          ) : null}
-        </>
+      evidence={route ? (
+        <LaboratoryEvidence
+          eyebrow="M4.7 · RAVNOVES00 FULL VIDEO"
+          title="EoMT CITY / DDRNet VEGETATION + YOLOX + CAUSAL TGS · 4489/4489 · TRUTH отсутствует"
+          kind="diagnostic-model"
+          resizable
+        >
+          <VegetationRouteEvidence result={result} />
+        </LaboratoryEvidence>
+      ) : (
+        <LaboratoryEvidence
+          eyebrow="M4.7 · RAVNOVES00 FULL VIDEO"
+          title="ROUTE ARCHIVE отсутствует"
+          kind="diagnostic-model"
+        >
+          <div className="m4-replay-threat-visual__pane-status" role="alert">
+            Для этой immutable identity нет полного route video evidence.
+          </div>
+        </LaboratoryEvidence>
      )}
      result={(
        <LaboratoryResultSummary
-          title={result.routeVideo?.viewKind === "coarse-material-policy-review"
-            ? "Слои собраны для визуального policy review; управление не авторизовано"
-            : "DDRNet — стартовые веса; перенос на ровер ещё не доказан"}
-          status={result.routeVideo?.viewKind === "coarse-material-policy-review"
-            ? "Materials are advisory · YOLOX/TGS veto cannot be cleared"
-            : `${selected.loadedModelName} выбран только как vegetation candidate`}
+          title="Многослойный visual review собран; управление не авторизовано"
+          status="Semantics advisory · YOLOX/TGS veto cannot be cleared"
          statusTone="warning"
          metrics={[
            {
-              label: "GOOSE mIoU",
-              value: `${decimal(selected.meanIouPercent, 2)}% / ${decimal(alternative.meanIouPercent, 2)}%`,
-              hint: `${selected.candidate} / ${alternative.candidate} · полный validation split`,
+              label: "Route masks",
+              value: route ? `${route.frameCount}/${route.frameCount}` : "0/4489",
+              hint: "sealed local playback · Worker для открытия не нужен",
            },
            {
-              label: "Vegetation IoU",
-              value: `${decimal(selected.vegetationMeanIouPercent, 2)}% / ${decimal(alternative.vegetationMeanIouPercent, 2)}%`,
-              hint: "агрегация классов grass/vegetation/bush/tree и родственных fine-64 labels",
+              label: "Semantic sources",
+              value: route ? "2 independent layers" : "0",
+              hint: "EoMT CITY / DDRNet VEGETATION · display switches, evidence does not fuse",
            },
            {
-              label: "Worker shadow p95",
-              value: `${decimal(selected.shadowLatencyP95Ms, 2)} / ${decimal(alternative.shadowLatencyP95Ms, 2)} ms`,
-              hint: "чистый inference · одна тяжёлая модель за раз",
+              label: "Vegetation worker p95",
+              value: `${decimal(selected.shadowLatencyP95Ms, 2)} ms`,
+              hint: "изолированный DDRNet inference; не совместный realtime stack",
            },
            {
-              label: "Cold prewarm",
-              value: `${decimal(selected.shadowPrewarmLatencyMs, 1)} / ${decimal(alternative.shadowPrewarmLatencyMs, 1)} ms`,
-              hint: "один явный inference до допуска кадров; исключён из steady-state p95",
+              label: "Vegetation peak VRAM",
+              value: `${decimal(selected.peakReservedVramBytes / 1024 ** 3, 2)} GiB`,
+              hint: "DDRNet candidate на Worker 006",
            },
-            {
-              label: "Worker throughput",
-              value: `${decimal(selected.shadowThroughputFps, 1)} / ${decimal(alternative.shadowThroughputFps, 1)} FPS`,
-              hint: "изолированный Worker 006 · не realtime graph целиком",
-            },
-            {
-              label: "Peak VRAM",
-              value: `${decimal(selected.peakReservedVramBytes / 1024 ** 3, 2)} / ${decimal(alternative.peakReservedVramBytes / 1024 ** 3, 2)} GiB`,
-              hint: `${selected.candidate} / ${alternative.candidate} · RTX 4090`,
-            },
-            {
-              label: "Hard-case evidence",
-              value: "12 truth-backed cases",
-              hint: "8 vegetation strata · Worker для открытия не требуется",
-            },
-            ...(result.routeVideo ? [{
-              label: "Route video",
-              value: "4489/4489 masks",
-              hint: result.routeVideo.viewKind === "coarse-material-policy-review"
-                ? "9 coarse states · YOLOX + causal TGS · Worker-independent playback"
-                : "DDRNet prediction · exact sequence · Worker-independent playback",
-            }] : []),
          ]}
          conclusion={{
-            proved: "Обе официальные fine-64 модели воспроизводимо запускаются на Worker 006; DDRNet лучше по aggregate vegetation IoU. Truth-backed hard cases прямо показывают траву, кусты и стволы, а не случайные автомобили и здания.",
-            notProved: "Не доказаны accuracy на нашем fisheye-домене, папоротник как отдельный материал, collision safety и physical-live поведение ровера. Видео позволяет увидеть temporal stability, но без truth не превращает её в метрику качества.",
-            decision: result.routeVideo?.viewKind === "coarse-material-policy-review"
-              ? "На одном M4.7 проверить ложные LOW GRASS/HIGH GRASS кандидаты против YOLOX и TGS. До truth-кейсов и integrated load этот слой не подключать к planner/actuation."
-              : "Смотреть полный prediction на видео и собирать конкретные temporal/domain failure cases. DDRNet остаётся diagnostic candidate; LiDAR/TGS fail-closed геометрию не ослаблять.",
+            proved: "На одной recorded timeline доступны городской EoMT, природный DDRNet, YOLOX detections и causal TGS; LAB автономна от Worker.",
+            notProved: "Не доказаны совместный live-runtime EoMT+DDRNet, truth accuracy на fisheye, стабильные vegetation subtypes и безопасное управление ровером.",
+            decision: "Использовать маски только для диагностики. Следующий qualification gate — motion-aware temporal vegetation fusion и отдельный совместный realtime load test; до него planner/actuation остаются OFF.",
          }}
        />
      )}
@@ -10,6 +10,7 @@ export type LaboratoryProfileId =
  | "rig-camera-local-surface-v1"
  | "rig-track-geometry-temporal-v1"
  | "rig-ravnoves-perception-gate-v1"
+  | "rig-goose-vegetation-benchmark-v1"
  | "rig-pointpillars-transfer-v1"
  | "rig-right-yolox-lidar-range-v1"
  | "rig-nvidia-ready-stack-v1"
@@ -63,12 +64,19 @@ interface KnownWorkDefinition {
 const rig = (rigLabel: string): string => rigLabel.trim() || "Сенсорный риг";

 const KNOWN_WORKS: Readonly<Record<Exclude<LaboratoryWorkId, `session:${string}`>, KnownWorkDefinition>> = {
+  "lab-v1-vegetation-benchmark": {
+    profileId: "rig-goose-vegetation-benchmark-v1",
+    profileName: (rigLabel) => `${rig(rigLabel)} · GOOSE vegetation archive`,
+    experimentId: "lab-v1-vegetation-benchmark-archive",
+    experimentName: "DDRNet vs PPLiteSeg · truth-backed archival comparison",
+    variantName: "M4.8 · GOOSE truth · архивный анализ моделей",
+  },
  "lab-v1-vegetation-shadow": {
    profileId: "rig-ravnoves-perception-gate-v1",
-    profileName: (rigLabel) => `${rig(rigLabel)} · GOOSE vegetation qualification`,
+    profileName: (rigLabel) => `${rig(rigLabel)} · RAVNOVES00 rover perception gate`,
    experimentId: "lab-v1-vegetation-mission-policy",
-    experimentName: "DDRNet vs PPLiteSeg · truth-backed vegetation hard cases",
-    variantName: "LAB V1 · готовые vegetation weights · GOOSE truth",
+    experimentName: "RAVNOVES00 · city + vegetation + TGS review",
+    variantName: "LAB V1 · EoMT + DDRNet + YOLOX + TGS · commands OFF",
  },
  "m48-object-centric-quality": {
    profileId: "rig-dual-evidence-virtual-corridor-v1",
@@ -18,6 +18,7 @@ function mergeResults(
  next: AdvancedLaboratoryResults,
 ): AdvancedLaboratoryResults {
  return {
+    vegetationBenchmark: next.vegetationBenchmark ?? current.vegetationBenchmark,
    vegetationShadow: next.vegetationShadow ?? current.vegetationShadow,
    m47Graph: next.m47Graph ?? current.m47Graph,
    m48: next.m48 ?? current.m48,
@@ -122,6 +123,7 @@ export function useAdvancedLaboratoryCatalog({
    const indexedResultId = index.find((item) => item.workId === selectedWorkId)?.resultId;
    if (
      [
+        "lab-v1-vegetation-benchmark",
        "lab-v1-vegetation-shadow",
        "m47-reference-graph-shadow",
        "m48-object-centric-quality",