diff --git a/changelog.d/1485-cctv-view-thinning.md b/changelog.d/1485-cctv-view-thinning.md
new file mode 100644
index 000000000..307da84ad
--- /dev/null
+++ b/changelog.d/1485-cctv-view-thinning.md
@@ -0,0 +1,4 @@
+- Draw only the CCTV cameras in view, and thin them to one per ground cell when
+  zoomed out. Dragging the globe with the default 4,000-camera catalog no
+  longer stalls, and the camera picker lists the cameras in view with a count
+  of the ones outside it (samehkhamis, #1485).
diff --git a/docs/CURRENT-STATE.md b/docs/CURRENT-STATE.md
index 0492c7c24..26f8178ed 100644
--- a/docs/CURRENT-STATE.md
+++ b/docs/CURRENT-STATE.md
@@ -978,6 +978,15 @@ directory. Every `/api/cctv` route refuses cross-site browser requests but
 accepts forwarding headers, so cameras work behind a reverse proxy; a forwarded
 request skips the key-bearing Street View fallback and gets the placeholder.
 
+The CCTV layer draws only the cameras on or near the screen. Zoomed out, it
+keeps one camera per ground cell sized to the view, preferring a healthy feed,
+and the cells are anchored to the ground so a pan keeps the same cameras. The
+active camera always shows. Ground placement and coverage geometry load only
+for the cameras in view, with at most two attempts per camera and view regime.
+The view pass reruns while the camera moves, paced by its own cost. The camera
+picker lists the cameras in view, active camera first, and closes with the
+count of cameras outside the view.
+
 Traffic owns its rendered-surface observer only while enabled, removes it on disable/destroy, and caches surface identity without allocating on unchanged frames.
 
 Traffic and bikeshare expose factories through `./layers/traffic` and
diff --git a/src/data/cctv.js b/src/data/cctv.js
index da65f5f08..ad95f2794 100644
--- a/src/data/cctv.js
+++ b/src/data/cctv.js
@@ -47,6 +47,8 @@ export const prioritizeActiveCctvGeometryRecord =
 export const processGeometryBatch = layer.processGeometryBatch;
 export const hideCctvRecordVisuals = layer.hideCctvRecordVisuals;
 export const refreshCoverageStyles = layer.refreshCoverageStyles;
+export const refreshHorizonCulling = layer.refreshHorizonCulling;
+export const _viewRecordIdsForTest = layer._viewRecordIdsForTest;
 export const clearProbeClampOnDeactivation =
   layer.clearProbeClampOnDeactivation;
 export const cctvRecordNeedsActivation = layer.cctvRecordNeedsActivation;
diff --git a/src/data/cctv.test.mjs b/src/data/cctv.test.mjs
index dafaeb5f6..0b7cee20f 100644
--- a/src/data/cctv.test.mjs
+++ b/src/data/cctv.test.mjs
@@ -1309,11 +1309,12 @@ test('CCTV null-active coverage, auto-hop, cycling, and panel targets stay hones
     assert.ok(renderer, '_renderCctvState is missing');
     assert.match(
       renderer[0],
-      /else if \(!activeId\)[\s\S]*?selectedIndex = -1/,
+      // Nothing active clears the pick; only the outside-view note may show.
+      /else if \(!activeId\)[\s\S]*?selectedIndex =[^;]*\? 0 : -1;/,
     );
     assert.match(
       renderer[0],
-      /_cctvFocusBtn\.disabled\s*=\s*!enabled\s*\|\|\s*cameras\.length === 0\s*\|\|\s*!activeId/,
+      /_cctvFocusBtn\.disabled\s*=\s*!enabled\s*\|\|\s*catalogCount === 0\s*\|\|\s*!activeId/,
     );
   } finally {
     cctvLayer.setParams({ autoHop: false });
diff --git a/src/data/cctvViewThinning.test.mjs b/src/data/cctvViewThinning.test.mjs
new file mode 100644
index 000000000..1bc94eed8
--- /dev/null
+++ b/src/data/cctvViewThinning.test.mjs
@@ -0,0 +1,251 @@
+// src/data/cctvViewThinning.test.mjs — the CCTV view pass: only cameras on or
+// near the screen show, a zoomed-out view keeps one camera per world-anchored
+// cell, and the layer's UI state never rebuilds every camera for a listener
+// that does not read the list.
+import { test } from 'node:test';
+import assert from 'node:assert/strict';
+import * as Cesium from 'cesium';
+import cctvLayer, {
+  _setCctvCoverageStateForTest,
+  _viewRecordIdsForTest,
+  refreshHorizonCulling,
+} from './cctv.js';
+
+const WIDTH = 800;
+const HEIGHT = 600;
+
+/**
+ * Records at the given [lon, lat] degrees near the origin, each drawn at the
+ * screen point `screenAt` gives it (default: the middle of the screen).
+ */
+function recordsAt(
+  lonLats,
+  screenAt = () => ({ x: WIDTH / 2, y: HEIGHT / 2 }),
+) {
+  const screenByPosition = new Map();
+  const records = lonLats.map(([lon, lat], index) => {
+    const position = Cesium.Cartesian3.fromDegrees(lon, lat, 10);
+    screenByPosition.set(position, screenAt(index));
+    return {
+      camera: {
+        id: `cam-${index}`,
+        name: `Camera ${index}`,
+        city: 'Test',
+        lat,
+        lon,
+        headingDeg: 0,
+        pitchDeg: -18,
+        fovDeg: 44,
+        rangeM: 145,
+        mountHeightM: 8,
+      },
+      position,
+      billboard: { position, show: true },
+      coverageEntities: [],
+      projection: null,
+    };
+  });
+  return { records, screenByPosition };
+}
+
+function viewerAt(heightM, screenByPosition, shiftX = 0) {
+  return {
+    isDestroyed: () => false,
+    camera: {
+      positionWC: Cesium.Cartesian3.fromDegrees(0, 0, heightM),
+      positionCartographic: { latitude: 0, longitude: 0, height: heightM },
+      frustum: { fovy: Math.PI / 3 },
+    },
+    scene: {
+      canvas: { clientWidth: WIDTH, clientHeight: HEIGHT },
+      cartesianToCanvasCoordinates: (position) => {
+        const screen = screenByPosition.get(position);
+        return screen && { x: screen.x + shiftX, y: screen.y };
+      },
+    },
+  };
+}
+
+function shownIds(records) {
+  return records
+    .filter((record) => record.billboard.show)
+    .map((record) => record.camera.id)
+    .sort();
+}
+
+// Three cameras within ~110 m, and one ~22 km away.
+const CLUSTER = [
+  [0, 0],
+  [0.001, 0],
+  [0.0005, 0.0005],
+  [0.2, 0],
+];
+
+test('zoomed out, cameras sharing a ground cell show one icon', () => {
+  const { records, screenByPosition } = recordsAt(CLUSTER);
+  _setCctvCoverageStateForTest({
+    viewer: viewerAt(20_000, screenByPosition),
+    records,
+  });
+  assert.equal(
+    refreshHorizonCulling(),
+    true,
+    'the first pass changes the view',
+  );
+  const shown = shownIds(records);
+  assert.equal(shown.length, 2, 'one for the cluster, one for the far camera');
+  assert.ok(shown.includes('cam-3'), 'a camera alone in its cell shows');
+  assert.deepEqual(_viewRecordIdsForTest().sort(), shown);
+});
+
+test('a pan keeps the same winners, and an unchanged view keeps its list', () => {
+  const { records, screenByPosition } = recordsAt(CLUSTER, (index) => ({
+    x: 100 + index * 7,
+    y: 300,
+  }));
+  _setCctvCoverageStateForTest({
+    viewer: viewerAt(20_000, screenByPosition),
+    records,
+  });
+  refreshHorizonCulling();
+  const before = shownIds(records);
+  const list = _viewRecordIdsForTest();
+  for (const shiftX of [3, 11, 20, 37]) {
+    _setCctvCoverageStateForTest({
+      viewer: viewerAt(20_000, screenByPosition, shiftX),
+      records,
+    });
+    refreshHorizonCulling();
+    assert.deepEqual(shownIds(records), before, `pan by ${shiftX} px`);
+  }
+  assert.equal(refreshHorizonCulling(), false, 'the same view changes nothing');
+  assert.deepEqual(_viewRecordIdsForTest(), list);
+});
+
+test('a small zoom within one cell size keeps the same winners', () => {
+  const { records, screenByPosition } = recordsAt(CLUSTER);
+  const shownAt = (heightM) => {
+    _setCctvCoverageStateForTest({
+      viewer: viewerAt(heightM, screenByPosition),
+      records,
+    });
+    refreshHorizonCulling();
+    return shownIds(records);
+  };
+  // Cells come in power-of-two metre steps: about 1.5 km and 1.6 km cells at
+  // these heights fall in the same step, so the same cameras win.
+  assert.deepEqual(shownAt(20_500), shownAt(20_000));
+});
+
+test('a healthy camera wins its cell over an unknown one', () => {
+  const { records, screenByPosition } = recordsAt(CLUSTER.slice(0, 2));
+  // Whichever camera the id order would pick, the healthy one must win.
+  for (const id of ['cam-0', 'cam-1']) {
+    _setCctvCoverageStateForTest({
+      viewer: viewerAt(20_000, screenByPosition),
+      records,
+      healthById: new Map([[id, { status: 'ok' }]]),
+    });
+    refreshHorizonCulling();
+    assert.deepEqual(shownIds(records), [id]);
+  }
+});
+
+test('the active camera always shows, even in a crowded cell', () => {
+  const { records, screenByPosition } = recordsAt(CLUSTER.slice(0, 3));
+  for (const active of records) {
+    _setCctvCoverageStateForTest({
+      viewer: viewerAt(20_000, screenByPosition),
+      records,
+      activeCameraId: active.camera.id,
+    });
+    refreshHorizonCulling();
+    assert.ok(active.billboard.show, `${active.camera.id} shows while active`);
+  }
+});
+
+test('close up, every camera near the screen shows and far-off ones hide', () => {
+  const { records, screenByPosition } = recordsAt(
+    [
+      [0, 0],
+      [0.0001, 0],
+      [0.0002, 0],
+    ],
+    (index) =>
+      index === 2 ? { x: WIDTH * 3, y: HEIGHT / 2 } : { x: 100, y: 100 },
+  );
+  _setCctvCoverageStateForTest({
+    viewer: viewerAt(300, screenByPosition),
+    records,
+  });
+  refreshHorizonCulling();
+  assert.deepEqual(shownIds(records), ['cam-0', 'cam-1']);
+});
+
+test('a disabled layer skips the view pass', () => {
+  const { records, screenByPosition } = recordsAt(CLUSTER);
+  _setCctvCoverageStateForTest({
+    viewer: viewerAt(20_000, screenByPosition),
+    records,
+    enabled: false,
+  });
+  assert.equal(refreshHorizonCulling(), false);
+  assert.deepEqual(_viewRecordIdsForTest(), []);
+});
+
+test('UI state builds the full camera list only when it is read', () => {
+  const { records, screenByPosition } = recordsAt(CLUSTER.slice(0, 2));
+  _setCctvCoverageStateForTest({
+    viewer: viewerAt(300, screenByPosition),
+    records,
+  });
+  refreshHorizonCulling();
+  let built = 0;
+  for (const record of records) {
+    const camera = record.camera;
+    record.camera = new Proxy(camera, {
+      get(target, key) {
+        if (key === 'calibration') built += 1;
+        return target[key];
+      },
+    });
+  }
+  const first = cctvLayer.getUIState();
+  assert.equal(built, 0, 'listeners that skip `cameras` cost nothing');
+  assert.deepEqual(
+    first.cameraIndex.map((camera) => camera.id),
+    ['cam-0', 'cam-1'],
+  );
+  assert.equal(
+    cctvLayer.getUIState().cameraIndex,
+    first.cameraIndex,
+    'the index keeps its identity while the view does',
+  );
+  assert.equal(first.cameras.length, 2);
+  assert.ok(built > 0, 'reading `cameras` builds it');
+  assert.equal(first.cameras, first.cameras, 'and builds it once');
+});
+
+test('the camera picker lists the view, with the active camera first', () => {
+  const { records, screenByPosition } = recordsAt(
+    [
+      [0, 0],
+      [0.0001, 0],
+      [0.0002, 0],
+    ],
+    (index) => (index === 1 ? { x: WIDTH * 3, y: 0 } : { x: 100, y: 100 }),
+  );
+  _setCctvCoverageStateForTest({
+    viewer: viewerAt(300, screenByPosition),
+    records,
+    activeCameraId: 'cam-1',
+  });
+  refreshHorizonCulling();
+  const state = cctvLayer.getUIState();
+  assert.deepEqual(
+    state.cameraIndex.map((camera) => camera.id),
+    ['cam-1', 'cam-0', 'cam-2'],
+    'the off-screen active camera leads, then what is in view',
+  );
+  assert.equal(state.cameras.length, 3, 'the full list is still every camera');
+});
diff --git a/src/i18n/locales/de.json b/src/i18n/locales/de.json
index 874fd5f31..8e383aded 100644
--- a/src/i18n/locales/de.json
+++ b/src/i18n/locales/de.json
@@ -372,6 +372,10 @@
     "one": "{count} Kamera geladen · CCTV aktivieren",
     "other": "{count} Kameras geladen · CCTV aktivieren"
   },
+  "{count} more cameras outside this view · pan or zoom out": {
+    "one": "{count} weitere Kamera außerhalb dieser Ansicht · verschieben oder herauszoomen",
+    "other": "{count} weitere Kameras außerhalb dieser Ansicht · verschieben oder herauszoomen"
+  },
   "Enable CCTV to load camera intersections": "CCTV aktivieren, um Kreuzungskameras zu laden",
   "Enable CCTV to start camera-linked intelligence summaries.": "CCTV aktivieren, um kameragestützte Lagezusammenfassungen zu starten.",
   "No summary available.": "Keine Zusammenfassung verfügbar.",
diff --git a/src/i18n/locales/es.json b/src/i18n/locales/es.json
index 0ae48f38b..c66eae636 100644
--- a/src/i18n/locales/es.json
+++ b/src/i18n/locales/es.json
@@ -369,6 +369,11 @@
     "many": "{count} de cámaras cargadas · activa CCTV para activarlas",
     "other": "{count} cámaras cargadas · activa CCTV para activarlas"
   },
+  "{count} more cameras outside this view · pan or zoom out": {
+    "one": "{count} cámara más fuera de esta vista · desplaza o aleja",
+    "many": "{count} de cámaras más fuera de esta vista · desplaza o aleja",
+    "other": "{count} cámaras más fuera de esta vista · desplaza o aleja"
+  },
   "No summary available.": "No hay resumen disponible.",
   "Celestial ring — available in Normal style": "Anillo celestial — disponible en el estilo Normal",
   "Detection overlay: {mode}": "Superposición de detección: {mode}",
diff --git a/src/i18n/locales/fr.json b/src/i18n/locales/fr.json
index 005a22417..931e7c3ff 100644
--- a/src/i18n/locales/fr.json
+++ b/src/i18n/locales/fr.json
@@ -1625,6 +1625,11 @@
     "many": "{count} de caméras chargées · activer CCTV pour les utiliser",
     "other": "{count} caméras chargées · activer CCTV pour les utiliser"
   },
+  "{count} more cameras outside this view · pan or zoom out": {
+    "one": "{count} autre caméra hors de cette vue · déplacer ou dézoomer",
+    "many": "{count} d'autres caméras hors de cette vue · déplacer ou dézoomer",
+    "other": "{count} autres caméras hors de cette vue · déplacer ou dézoomer"
+  },
   "Enable CCTV to load camera intersections": "Activez CCTV pour charger les carrefours équipés de caméras",
   "Enable CCTV to start camera-linked intelligence summaries.": "Activez CCTV pour lancer les résumés de renseignement liés aux caméras.",
   "No summary available.": "Aucun résumé disponible.",
diff --git a/src/i18n/locales/it.json b/src/i18n/locales/it.json
index bc9732d67..62c9931c1 100644
--- a/src/i18n/locales/it.json
+++ b/src/i18n/locales/it.json
@@ -1625,6 +1625,11 @@
     "many": "{count} di telecamere caricate · attiva CCTV per attivarle",
     "other": "{count} telecamere caricate · attiva CCTV per attivarle"
   },
+  "{count} more cameras outside this view · pan or zoom out": {
+    "one": "Altra {count} telecamera fuori da questa vista · sposta o riduci lo zoom",
+    "many": "Altre {count} di telecamere fuori da questa vista · sposta o riduci lo zoom",
+    "other": "Altre {count} telecamere fuori da questa vista · sposta o riduci lo zoom"
+  },
   "Enable CCTV to load camera intersections": "Attiva CCTV per caricare le telecamere agli incroci",
   "Enable CCTV to start camera-linked intelligence summaries.": "Attiva CCTV per avviare i riepiloghi intelligence collegati alle telecamere.",
   "No summary available.": "Nessun riepilogo disponibile.",
diff --git a/src/i18n/locales/ja.json b/src/i18n/locales/ja.json
index 7ac2840f6..5a48dc886 100644
--- a/src/i18n/locales/ja.json
+++ b/src/i18n/locales/ja.json
@@ -1579,6 +1579,9 @@
   "{count} cameras loaded · enable CCTV to activate": {
     "other": "カメラ {count}台を読込済み · CCTVを有効にして使用"
   },
+  "{count} more cameras outside this view · pan or zoom out": {
+    "other": "この表示範囲外にさらに {count}台のカメラ · 移動またはズームアウト"
+  },
   "Enable CCTV to load camera intersections": "CCTVを有効にすると、カメラのある交差点を読み込みます",
   "Enable CCTV to start camera-linked intelligence summaries.": "CCTVを有効にすると、カメラ連動のインテリジェンス概要を開始します。",
   "No summary available.": "利用できる概要はありません。",
diff --git a/src/i18n/locales/pt.json b/src/i18n/locales/pt.json
index 5e9a9a3ff..de42b5aab 100644
--- a/src/i18n/locales/pt.json
+++ b/src/i18n/locales/pt.json
@@ -1625,6 +1625,11 @@
     "many": "{count} de câmeras carregadas · ative o CCTV para usar",
     "other": "{count} câmeras carregadas · ative o CCTV para usar"
   },
+  "{count} more cameras outside this view · pan or zoom out": {
+    "one": "Mais {count} câmera fora desta vista · mova ou afaste o zoom",
+    "many": "Mais {count} de câmeras fora desta vista · mova ou afaste o zoom",
+    "other": "Mais {count} câmeras fora desta vista · mova ou afaste o zoom"
+  },
   "Enable CCTV to load camera intersections": "Ative o CCTV para carregar os cruzamentos com câmeras",
   "Enable CCTV to start camera-linked intelligence summaries.": "Ative o CCTV para iniciar resumos de inteligência vinculados às câmeras.",
   "No summary available.": "Nenhum resumo disponível.",
diff --git a/src/i18n/locales/ru.json b/src/i18n/locales/ru.json
index 4839eaaba..3aa2197e9 100644
--- a/src/i18n/locales/ru.json
+++ b/src/i18n/locales/ru.json
@@ -342,6 +342,12 @@
     "many": "Загружено {count} камер · включите CCTV, чтобы активировать",
     "other": "Загружено {count} камеры · включите CCTV, чтобы активировать"
   },
+  "{count} more cameras outside this view · pan or zoom out": {
+    "one": "Ещё {count} камера вне этого вида · сдвиньте или отдалите",
+    "few": "Ещё {count} камеры вне этого вида · сдвиньте или отдалите",
+    "many": "Ещё {count} камер вне этого вида · сдвиньте или отдалите",
+    "other": "Ещё {count} камеры вне этого вида · сдвиньте или отдалите"
+  },
   "Enable CCTV to load camera intersections": "Включите CCTV, чтобы загрузить камеры на перекрёстках",
   "Enable CCTV to start camera-linked intelligence summaries.": "Включите CCTV, чтобы запускать сводки по камерам.",
   "No summary available.": "Сводка недоступна.",
diff --git a/src/i18n/locales/uk.json b/src/i18n/locales/uk.json
index f87454f1a..6041d339f 100644
--- a/src/i18n/locales/uk.json
+++ b/src/i18n/locales/uk.json
@@ -1647,6 +1647,12 @@
     "many": "{count} камер завантажено · увімкніть CCTV, щоб активувати",
     "other": "{count} камери завантажено · увімкніть CCTV, щоб активувати"
   },
+  "{count} more cameras outside this view · pan or zoom out": {
+    "one": "Ще {count} камера поза цим видом · посуньте або віддаліть",
+    "few": "Ще {count} камери поза цим видом · посуньте або віддаліть",
+    "many": "Ще {count} камер поза цим видом · посуньте або віддаліть",
+    "other": "Ще {count} камери поза цим видом · посуньте або віддаліть"
+  },
   "Enable CCTV to load camera intersections": "Увімкніть CCTV, щоб завантажити камери на перехрестях",
   "Enable CCTV to start camera-linked intelligence summaries.": "Увімкніть CCTV, щоб отримувати розвідувальні зведення за даними камер.",
   "No summary available.": "Зведення недоступне.",
diff --git a/src/i18n/locales/zh-CN.json b/src/i18n/locales/zh-CN.json
index 5115ab7c9..5fb455546 100644
--- a/src/i18n/locales/zh-CN.json
+++ b/src/i18n/locales/zh-CN.json
@@ -1553,6 +1553,9 @@
   "{count} cameras loaded · enable CCTV to activate": {
     "other": "已加载 {count} 个摄像头 · 启用 CCTV 以激活"
   },
+  "{count} more cameras outside this view · pan or zoom out": {
+    "other": "此视图外还有 {count} 个摄像头 · 平移或缩小"
+  },
   "Enable CCTV to load camera intersections": "启用 CCTV 以加载路口摄像头",
   "Enable CCTV to start camera-linked intelligence summaries.": "启用 CCTV 以开始生成与摄像头关联的情报摘要。",
   "No summary available.": "暂无摘要。",
diff --git a/src/i18n/locales/zh-Hant.json b/src/i18n/locales/zh-Hant.json
index c3b5ca610..a2bb4ffa0 100644
--- a/src/i18n/locales/zh-Hant.json
+++ b/src/i18n/locales/zh-Hant.json
@@ -1543,6 +1543,9 @@
   "{count} cameras loaded · enable CCTV to activate": {
     "other": "已加載 {count} 個攝像頭 · 啟用 CCTV 以激活"
   },
+  "{count} more cameras outside this view · pan or zoom out": {
+    "other": "此視圖外還有 {count} 個攝像頭 · 平移或縮小"
+  },
   "Enable CCTV to load camera intersections": "啟用 CCTV 以加載路口攝像頭",
   "Enable CCTV to start camera-linked intelligence summaries.": "啟用 CCTV 以開始生成與攝像頭關聯的情報摘要。",
   "No summary available.": "暫無摘要。",
diff --git a/src/layers/cctv/cards.js b/src/layers/cctv/cards.js
index 72f0b88a4..04b8fb86f 100644
--- a/src/layers/cctv/cards.js
+++ b/src/layers/cctv/cards.js
@@ -99,7 +99,8 @@ export function createCards({ state: layerState, services, parts, source }) {
 
     const candidates = [];
     const screenById = new Map();
-    for (const record of layerState._records) {
+    // Cards go only on the icons the view pass shows (refreshHorizonCulling).
+    for (const record of layerState._viewRecords) {
       const id = record.camera.id;
       if (id === activeId || !record.position) continue;
       let inView = false;
@@ -187,7 +188,14 @@ export function createCards({ state: layerState, services, parts, source }) {
       nowMs: Date.now(),
       cardLimit: budgets.cardLimit,
     });
-    layerState._cardIds = new Set(retention.keepIds);
+    // Grace keeps a card briefly after it loses selection, but never on a
+    // camera whose icon the view pass has hidden.
+    const shownIds = new Set(
+      layerState._viewRecords.map((record) => record.camera.id),
+    );
+    layerState._cardIds = new Set(
+      retention.keepIds.filter((id) => shownIds.has(id)),
+    );
     layerState._cardGraceState = retention.graceState;
 
     // Bounded thumbnail LRU: live cards (grace included) never lose their
@@ -371,7 +379,6 @@ export function createCards({ state: layerState, services, parts, source }) {
     layerState._cctvOverlayHost.clearSource(CCTV_OVERLAY_SOURCE_ID);
     layerState._cctvOverlayHost.setVisible(CCTV_OVERLAY_SOURCE_ID, false);
     parts.hover.clearHoverCard();
-    layerState._cameraMoving = false;
     layerState._cardIds = new Set();
     layerState._cardGraceState = new Map();
     layerState._cardFrameSlots = new Map();
diff --git a/src/layers/cctv/controls.js b/src/layers/cctv/controls.js
index af9f1f427..492ff1758 100644
--- a/src/layers/cctv/controls.js
+++ b/src/layers/cctv/controls.js
@@ -205,22 +205,21 @@ export function createControls({ state: layerState, services, parts, source }) {
      * @returns {{ position: Cesium.Cartesian3, id: string, type: string }[]}
      */
     getDetectableObjects(options = {}) {
-      if (!layerState._enabled || layerState._records.length === 0) return [];
+      // The cameras the view pass shows, not icons it has hidden.
+      const records = layerState._viewRecords;
+      if (!layerState._enabled || records.length === 0) return [];
       const maxCount = Number.isFinite(options.maxCount)
         ? Math.max(1, Math.floor(options.maxCount))
-        : layerState._records.length;
+        : records.length;
       const seed = Number.isFinite(options.seed) ? Math.floor(options.seed) : 0;
-      const stride = Math.max(
-        1,
-        Math.ceil(layerState._records.length / maxCount),
-      );
+      const stride = Math.max(1, Math.ceil(records.length / maxCount));
       const start = seed % stride;
 
       const objects = [];
-      for (let i = start; i < layerState._records.length; i += stride) {
-        const camera = layerState._records[i].camera;
+      for (let i = start; i < records.length; i += stride) {
+        const camera = records[i].camera;
         objects.push({
-          position: layerState._records[i].position,
+          position: records[i].position,
           sourceId: camera.id,
           // Short semantic code where the id is opaque (see cameraDisplayCode).
           id: `CAM-${camera.code || camera.id}`,
diff --git a/src/layers/cctv/geometry.js b/src/layers/cctv/geometry.js
index d9db40fe9..362ad4789 100644
--- a/src/layers/cctv/geometry.js
+++ b/src/layers/cctv/geometry.js
@@ -581,6 +581,13 @@ export function createGeometry({ state: layerState, services, parts, source }) {
 
   function coverageNeighborCount(targetRecord) {
     if (!targetRecord) return 0;
+    // The summary asks on every notification; scan the catalog only when the
+    // camera, its pose or the catalog changed.
+    const { id, lat, lon, rangeM } = targetRecord.camera;
+    const key = `${id}|${lat}|${lon}|${rangeM}|${layerState._records.length}`;
+    const cached = layerState._neighborCountCache;
+    if (cached?.key === key && cached.records === layerState._records)
+      return cached.count;
     let count = 0;
     for (const record of layerState._records) {
       if (record === targetRecord) continue;
@@ -594,6 +601,11 @@ export function createGeometry({ state: layerState, services, parts, source }) {
         ((targetRecord.camera.rangeM + record.camera.rangeM) / 1000) * 0.92;
       if (dKm <= overlapKm) count++;
     }
+    layerState._neighborCountCache = {
+      key,
+      records: layerState._records,
+      count,
+    };
     return count;
   }
 
diff --git a/src/layers/cctv/geometryQueue.js b/src/layers/cctv/geometryQueue.js
index 5583e36af..b78950aaa 100644
--- a/src/layers/cctv/geometryQueue.js
+++ b/src/layers/cctv/geometryQueue.js
@@ -6,6 +6,8 @@ import {
   GEO_TRACKING_BATCH_DELAY_MS,
   GEO_LOAD_BATCH_SIZE,
   GEO_LOAD_BATCH_DELAY_MS,
+  VIEW_GEOMETRY_ATTEMPTS,
+  VIEW_GEOMETRY_RETRY_MS,
 } from './policy.js';
 
 export function createGeometryQueue({
@@ -25,6 +27,8 @@ export function createGeometryQueue({
       clearTimeout(layerState._geoQueueTimer);
       layerState._geoQueueTimer = 0;
     }
+    clearTimeout(layerState._viewGeometryRetryTimer);
+    layerState._viewGeometryRetryTimer = 0;
     layerState._geoQueue = [];
     layerState._geoProgressNotifier = null;
     if (clearProgress) {
@@ -223,6 +227,7 @@ export function createGeometryQueue({
             err?.message || err,
           );
         }
+        noteViewAttempt(record);
         if (
           layerState._geoLoading &&
           layerState._geoLoadDone < layerState._geoLoadTotal
@@ -236,14 +241,15 @@ export function createGeometryQueue({
         layerState._geoLoading = false;
         if (wasInitialLoad) {
           layerState._geoLoadDone = layerState._geoLoadTotal;
-          if (layerState._enabled) {
-            parts.rendering.refreshCoverageStyles();
-            // Geometry refinement may have replaced record.position objects — the
-            // one-shot drain completion re-anchors the card entries (event-driven,
-            // not a per-frame or timer pass).
-            parts.cards.refreshAmbientCards();
-          }
+          if (layerState._enabled) parts.rendering.refreshCoverageStyles();
         }
+        // Geometry refinement may have replaced record.position objects — every
+        // drain's completion re-anchors the card entries (event-driven, not a
+        // per-frame or timer pass).
+        // While the camera moves, moveEnd reselects them instead.
+        if (layerState._enabled && !layerState._cameraMoving)
+          parts.cards.refreshAmbientCards();
+        scheduleViewGeometryRetry();
         // Completion is never coalesced: subscribers must observe the final
         // loading state even if the last progress tick just happened.
         layerState._geoProgressNotifier?.finish();
@@ -261,16 +267,17 @@ export function createGeometryQueue({
 
   /**
    * Appends records to the geometry queue (no progress tracking) and starts
-   * the batch timer if idle. Used by update()'s ONE-SHOT tiles-ready completion
-   * pass (records left `!groundResolved` by an enable-time drain that ran while
-   * tiles were still streaming) so it shares the same stagger machinery as the
-   * initial load. Fires at most once per enable — never on a recurring timer.
+   * the batch timer if idle. Used by update()'s one-shot tiles-ready
+   * completion pass and by enqueueViewGeometry, so both share the initial
+   * load's stagger machinery. Never driven by a recurring timer.
    * @param {Object[]} records - Camera records needing geometry refresh.
    */
 
   function enqueueGeometryRefresh(records) {
+    const queued = new Set(layerState._geoQueue);
     for (const record of records) {
-      if (!layerState._geoQueue.includes(record)) {
+      if (!queued.has(record)) {
+        queued.add(record);
         layerState._geoQueue.push(record);
       }
     }
@@ -283,10 +290,10 @@ export function createGeometryQueue({
   }
 
   /**
-   * Starts the initial staggered load: orders all records active-camera-first,
-   * then by distance from the current viewer position (nearest first, so
-   * cameras likely in view refine before off-screen ones), and exposes
-   * loaded/total progress through uiState()/getStats() while running.
+   * Starts the initial staggered load over the geometry working set: active
+   * camera first, then by distance from the current viewer position (nearest
+   * first), and exposes loaded/total progress through uiState()/getStats()
+   * while running. Cameras outside the view refine when they come into it.
    */
 
   function startGeometryLoadQueue() {
@@ -303,7 +310,7 @@ export function createGeometryQueue({
     const refLon = carto
       ? Cesium.Math.toDegrees(carto.longitude)
       : (active?.camera.lon ?? 0);
-    const pending = layerState._records
+    const pending = geometryWorkingSet()
       .filter((record) => record !== active)
       .map((record) => ({
         record,
@@ -325,7 +332,134 @@ export function createGeometryQueue({
     );
     layerState._geoQueueTimer = setTimeout(processGeometryBatch, 0);
   }
+  /**
+   * The records whose geometry the view needs: the shown icons, the active
+   * camera and the coverage cohort drawn around it.
+   * @returns {Object[]}
+   */
+
+  function geometryWorkingSet() {
+    const records = new Set(layerState._viewRecords);
+    const active = parts.selection.getActiveRecord();
+    if (active) records.add(active);
+    for (const id of layerState._coverageVisibleIds) {
+      const record = layerState._recordById.get(id);
+      if (record) records.add(record);
+    }
+    return [...records];
+  }
+
+  /** The regime key a view attempt counts against: interim while tiles stream. */
+  function viewAttemptKey() {
+    const regime = parts.ground.currentSurfaceRegime();
+    const tilesReady =
+      regime !== 'google-3d' || parts.model.projectionTilesReady();
+    return { key: tilesReady ? regime : `${regime}:interim`, tilesReady };
+  }
+
+  /**
+   * Counts a processed geometry attempt against the current regime key; each
+   * key keeps its own count, so tile streaming going on and off does not
+   * reset the ready-tiles budget.
+   */
+  function noteViewAttempt(record) {
+    const { key } = viewAttemptKey();
+    record.viewAttempts ??= {};
+    record.viewAttempts[key] = (record.viewAttempts[key] || 0) + 1;
+    record.viewAttemptAt = performance.now();
+  }
+
+  /**
+   * Whether a record may be attempted again under `key`. After a first
+   * unresolved attempt, the second waits out VIEW_GEOMETRY_RETRY_MS so its
+   * elevation cell can warm, rather than going to the next motion tick.
+   */
+  /** Whether a record still has an attempt left under the current key. */
+  function withinViewBudget(record) {
+    const { key } = viewAttemptKey();
+    return (record.viewAttempts?.[key] || 0) < VIEW_GEOMETRY_ATTEMPTS;
+  }
+
+  function attemptsLeft(record, key) {
+    const attempts = record.viewAttempts?.[key] || 0;
+    if (attempts >= VIEW_GEOMETRY_ATTEMPTS) return false;
+    return (
+      attempts === 0 ||
+      performance.now() - record.viewAttemptAt >= VIEW_GEOMETRY_RETRY_MS
+    );
+  }
+
+  /**
+   * After a drain, gives working-set cameras it left ungrounded one more
+   * attempt once their elevation cells have had time to warm. The timer is
+   * set for the earliest such camera and re-arms for the rest when it fires.
+   */
+  function scheduleViewGeometryRetry() {
+    if (!layerState._enabled || layerState._viewGeometryRetryTimer) return;
+    const { key } = viewAttemptKey();
+    const active = parts.selection.getActiveRecord();
+    let due = Infinity;
+    for (const record of geometryWorkingSet()) {
+      const attempts = record.viewAttempts?.[key] || 0;
+      if (
+        record === active ||
+        attempts === 0 ||
+        attempts >= VIEW_GEOMETRY_ATTEMPTS ||
+        !parts.ground.needsGeometry(record)
+      )
+        continue;
+      due = Math.min(due, record.viewAttemptAt + VIEW_GEOMETRY_RETRY_MS);
+    }
+    if (due === Infinity) return;
+    layerState._viewGeometryRetryTimer = setTimeout(
+      () => {
+        layerState._viewGeometryRetryTimer = 0;
+        enqueueViewGeometry();
+        // A drain re-arms on completion; with none running, the cameras still
+        // cooling down need their own timer.
+        if (!layerState._geoQueueTimer) scheduleViewGeometryRetry();
+      },
+      Math.max(0, due - performance.now()),
+    );
+  }
+
+  /**
+   * Queues working-set cameras the current surface has not grounded yet,
+   * after a camera move brings new ones into view. Attempts are counted when
+   * a record is processed, not queued, so a drain cut short by disable() is
+   * not lost; each record gets VIEW_GEOMETRY_ATTEMPTS per surface regime,
+   * spaced by VIEW_GEOMETRY_RETRY_MS, so a sample the shared floor refuses
+   * (too far, too high) is not retried on every move. While photoreal tiles are still streaming the attempts are
+   * interim, and the update() completion pass is re-armed to retry these
+   * records when they land. The active camera is left to selection, which
+   * owns its re-sampling.
+   */
+
+  function enqueueViewGeometry() {
+    if (!layerState._enabled) return;
+    const { key, tilesReady } = viewAttemptKey();
+    const active = parts.selection.getActiveRecord();
+    const pending = [];
+    let coolingDown = false;
+    for (const record of geometryWorkingSet()) {
+      if (record === active || !parts.ground.needsGeometry(record)) continue;
+      if (!attemptsLeft(record, key)) {
+        coolingDown ||=
+          (record.viewAttempts?.[key] || 0) < VIEW_GEOMETRY_ATTEMPTS;
+        continue;
+      }
+      pending.push(record);
+    }
+    // A camera back in view during its cooldown still needs its retry timer.
+    if (coolingDown && !layerState._geoQueueTimer) scheduleViewGeometryRetry();
+    if (!pending.length) return;
+    if (!tilesReady) layerState._tilesReadyReenqueued = false;
+    enqueueGeometryRefresh(pending);
+  }
   return {
+    enqueueViewGeometry,
+    withinViewBudget,
+    geometryWorkingSet,
     stopGeometryLoadQueue,
     createGeometryProgressNotifier,
     processCctvGeometryQueueBatch,
diff --git a/src/layers/cctv/ground.js b/src/layers/cctv/ground.js
index 5ff41c377..fdf36db14 100644
--- a/src/layers/cctv/ground.js
+++ b/src/layers/cctv/ground.js
@@ -1,3 +1,4 @@
+import * as Cesium from 'cesium';
 import {
   SUPPORT_KEYS,
   planeSupportPoints,
@@ -78,6 +79,14 @@ export function createGround({ state: layerState, services, parts, source }) {
     return record?.groundResolved?.[regime] === true;
   }
 
+  /**
+   * True until the record has geometry grounded for the regime. Cameras are
+   * refined only once they enter the view, so many never have geometry.
+   */
+  function needsGeometry(record, regime = currentSurfaceRegime()) {
+    return !record?.frustumGeometry || !isGroundResolved(record, regime);
+  }
+
   /**
    * Ground altitude used for pure geometry recomputes: the given regime's
    * cached resolution (`record.groundSamples[regime]` — a shared mesh/DEM floor
@@ -184,8 +193,11 @@ export function createGround({ state: layerState, services, parts, source }) {
    * Task 5: applies a LATE-arriving ground-prior batch (init's bounded race
    * lost — cold proxy cache / slow upstream). Pure recomputes only, no scene
    * queries:
+   *  - records without geometry yet: only the icon moves onto the prior;
+   *    the frustum is built when the camera enters the view.
    *  - terrain-globe regime: the prior IS the resolution → re-run the
-   *    resolution (updateRecordGeometry latches it) for every record.
+   *    resolution (updateRecordGeometry latches it) for every record with
+   *    geometry.
    *  - google-3d regime: records still awaiting a shared floor move from the
    *    catalog fallback onto the exact prior; records already holding a shared
    *    mesh/DEM floor keep it untouched.
@@ -198,6 +210,7 @@ export function createGround({ state: layerState, services, parts, source }) {
   function applyLateGroundPriors(records, priors) {
     if (!Array.isArray(records) || !Array.isArray(priors)) return;
     let applied = 0;
+    let moved = 0;
     for (let i = 0; i < records.length && i < priors.length; i++) {
       const record = records[i];
       const prior = priors[i];
@@ -212,6 +225,24 @@ export function createGround({ state: layerState, services, parts, source }) {
         record.camera.absoluteHeightM =
           prior.ellipsoid + record.camera.mountHeightM;
       }
+      // Cameras without geometry only move their icon onto the prior, so the
+      // view pass projects the right point; their frustums are built when
+      // they enter the view, as recomputing the whole catalog here would
+      // stall a large one.
+      if (!record.frustumGeometry) {
+        if (record.position) {
+          Cesium.Cartesian3.fromDegrees(
+            record.camera.lon,
+            record.camera.lat,
+            record.camera.absoluteHeightM,
+            undefined,
+            record.position,
+          );
+          if (record.billboard) record.billboard.position = record.position;
+          moved += 1;
+        }
+        continue;
+      }
       const regime = currentSurfaceRegime();
       if (regime === 'terrain-globe') {
         // Prior IS the resolution — re-latch onto the fresh value.
@@ -224,7 +255,9 @@ export function createGround({ state: layerState, services, parts, source }) {
         applied += 1;
       }
     }
-    if (applied) parts.presentation.notifyListeners();
+    if (moved && parts.rendering.refreshHorizonCulling())
+      parts.geometryQueue.enqueueViewGeometry();
+    if (applied || moved) parts.presentation.notifyListeners();
   }
 
   /**
@@ -234,7 +267,7 @@ export function createGround({ state: layerState, services, parts, source }) {
    * deck/building-top in google-3d vs bare Re:Earth DEM on globe stacks), so
    * on a REGIME change (photoreal ↔ globe; bing→osm stays 'terrain-globe' and
    * no-ops):
-   *  1. every record's geometry recomputes IMMEDIATELY from the new regime's
+   *  1. every record that has geometry recomputes IMMEDIATELY from the new regime's
    *     resolution — cached sample if that regime has one, else the Re:Earth
    *     prior (never blank, zero scene queries);
    *  2. entering google-3d re-arms the one-shot tiles-ready completion latch so
@@ -250,6 +283,9 @@ export function createGround({ state: layerState, services, parts, source }) {
     layerState._lastAppliedRegime = regime;
 
     for (const record of layerState._records) {
+      // Only cameras that already have geometry are rewritten; the rest are
+      // grounded for the new regime when they enter the view.
+      if (!record.frustumGeometry) continue;
       if (regime === 'terrain-globe') {
         // Prior IS the resolution — latch it (zero scene queries).
         record.groundSamples['terrain-globe'] = groundPriorAltFor(record);
@@ -353,6 +389,7 @@ export function createGround({ state: layerState, services, parts, source }) {
     resolveFootprintGround,
     groundPriorAltFor,
     isGroundResolved,
+    needsGeometry,
     groundAltFor,
     rearmGroundResolution,
     resolveCommittedGroundAnchor,
diff --git a/src/layers/cctv/index.js b/src/layers/cctv/index.js
index 0deba580e..6f1cc77c7 100644
--- a/src/layers/cctv/index.js
+++ b/src/layers/cctv/index.js
@@ -91,6 +91,9 @@ export function createCctvLayer({ services, source }) {
       processGeometryBatch: parts.geometryQueue.processGeometryBatch,
       hideCctvRecordVisuals: parts.rendering.hideCctvRecordVisuals,
       refreshCoverageStyles: parts.rendering.refreshCoverageStyles,
+      refreshHorizonCulling: parts.rendering.refreshHorizonCulling,
+      _viewRecordIdsForTest: () =>
+        state._viewRecords.map((record) => record.camera.id),
       clearProbeClampOnDeactivation:
         parts.geometry.clearProbeClampOnDeactivation,
       cctvRecordNeedsActivation: parts.model.cctvRecordNeedsActivation,
diff --git a/src/layers/cctv/ingestion.js b/src/layers/cctv/ingestion.js
index 0947ccdb6..2613de1e1 100644
--- a/src/layers/cctv/ingestion.js
+++ b/src/layers/cctv/ingestion.js
@@ -32,13 +32,22 @@ export function createIngestion({
         // projectionTilesReady() is false while a (hidden) Google tileset
         // exists, so this latch effectively fires for the google-3d regime —
         // terrain-globe records resolve from the prior in their drain pass.
-        const unresolved = layerState._records.filter(
-          (record) => !parts.ground.isGroundResolved(record),
-        );
+        const unresolved = parts.geometryQueue
+          .geometryWorkingSet()
+          .filter(
+            (record) =>
+              !parts.ground.isGroundResolved(record) &&
+              parts.geometryQueue.withinViewBudget(record),
+          );
         if (unresolved.length)
           parts.geometryQueue.enqueueGeometryRefresh(unresolved);
       }
       await parts.health.syncHealthState();
+      // New health can change which camera wins a crowded cell.
+      if (layerState._enabled && parts.rendering.refreshHorizonCulling()) {
+        parts.geometryQueue.enqueueViewGeometry();
+        parts.cards.refreshAmbientCards();
+      }
       parts.navigation.maybeAutoHop(now);
       parts.presentation.notifyListeners();
     },
diff --git a/src/layers/cctv/lifecycle.js b/src/layers/cctv/lifecycle.js
index 6d041aa78..4e4443df8 100644
--- a/src/layers/cctv/lifecycle.js
+++ b/src/layers/cctv/lifecycle.js
@@ -7,8 +7,14 @@ import {
   CAMERA_ICON,
   IDLE_CAMERA_COLOR,
   CALIBRATION_RANGE_FLOOR_M,
+  VIEW_MOTION_REFRESH_MS,
+  VIEW_MOTION_MIN_MOVE_FRACTION,
+  VIEW_MOTION_MIN_TURN_DEG,
+  VIEW_MOTION_COST_FACTOR,
 } from './policy.js';
 
+const MIN_TURN_COS = Math.cos((VIEW_MOTION_MIN_TURN_DEG * Math.PI) / 180);
+
 export function createLifecycle({
   state: layerState,
   services,
@@ -22,15 +28,94 @@ export function createLifecycle({
   const { onFocusTargetAppear } = services.focus;
   const { releaseContinuousRender } = services.render;
 
+  /**
+   * The settled-view pass: thin the icons, queue the newly visible geometry,
+   * reselect cards, and tell subscribers (the camera picker) when the shown
+   * set changed since they were last told, including by motion passes.
+   */
+  function settleView() {
+    const changed = parts.rendering.refreshHorizonCulling();
+    noteViewPose();
+    parts.geometryQueue.enqueueViewGeometry();
+    parts.cards.refreshAmbientCards();
+    if (changed || layerState._viewUnpublished) {
+      layerState._viewUnpublished = false;
+      parts.presentation.notifyListeners();
+    }
+  }
+
+  /** Records the camera pose the last view pass ran at. */
+  function noteViewPose() {
+    const camera = layerState._viewer?.camera;
+    if (!camera?.positionWC) return;
+    layerState._viewPose = {
+      position: Cesium.Cartesian3.clone(camera.positionWC),
+      direction:
+        camera.directionWC && Cesium.Cartesian3.clone(camera.directionWC),
+    };
+  }
+
+  /**
+   * Whether the camera moved or turned enough since the last view pass to
+   * change what is on screen. A camera that tracks or flies without settling
+   * would otherwise rerun the full pass on every tick.
+   */
+  function viewPoseChanged() {
+    const camera = layerState._viewer?.camera;
+    const pose = layerState._viewPose;
+    if (!pose || !camera?.positionWC) return true;
+    const height = Math.max(1, camera.positionCartographic?.height ?? 0);
+    const moved = Cesium.Cartesian3.distance(camera.positionWC, pose.position);
+    if (moved > height * VIEW_MOTION_MIN_MOVE_FRACTION) return true;
+    if (!pose.direction || !camera.directionWC) return false;
+    return (
+      Cesium.Cartesian3.dot(camera.directionWC, pose.direction) < MIN_TURN_COS
+    );
+  }
+
+  function viewMotionTick() {
+    if (!layerState._enabled) return;
+    const now = performance.now();
+    const pass = layerState._viewMotionPass;
+    if (pass && now - pass.at < pass.ms * VIEW_MOTION_COST_FACTOR) return;
+    if (!viewPoseChanged()) return;
+    if (parts.rendering.refreshHorizonCulling())
+      layerState._viewUnpublished = true;
+    noteViewPose();
+    parts.geometryQueue.enqueueViewGeometry();
+    layerState._viewMotionPass = { at: now, ms: performance.now() - now };
+  }
+
+  function startViewMotionTimer() {
+    if (layerState._viewMotionTimer) return;
+    layerState._viewMotionTimer = setInterval(
+      viewMotionTick,
+      VIEW_MOTION_REFRESH_MS,
+    );
+  }
+
+  function stopViewMotionTimer() {
+    clearInterval(layerState._viewMotionTimer);
+    layerState._viewMotionTimer = 0;
+  }
+
   /** Resets all module-scoped runtime state to initial values. */
 
   function clearRuntimeState() {
     parts.geometryQueue.stopGeometryLoadQueue();
+    stopViewMotionTimer();
     // Idempotent — also covers a re-init without a prior destroy().
     parts.cards.teardownAmbientCards();
     parts.projection.clearProjectionOverlay();
     layerState._records = [];
     layerState._recordById = new Map();
+    layerState._viewRecords = [];
+    layerState._cameraIndexCache = null;
+    layerState._coverageVisibleIds = new Set();
+    layerState._neighborCountCache = null;
+    layerState._viewUnpublished = false;
+    layerState._viewPose = null;
+    layerState._viewMotionPass = null;
     layerState._healthById = new Map();
     layerState._count = 0;
     layerState._lastUpdate = null;
@@ -71,6 +156,13 @@ export function createLifecycle({
       layerState._calibrationById = parts.calibration.loadCalibrationStore();
 
       layerState._billboards = new Cesium.BillboardCollection();
+      // Cesium switches a property's buffer-usage hint whenever it starts or
+      // stops changing from one frame to the next, and every switch rewrites
+      // the vertex data of every billboard. The view pass toggles `show` on a
+      // few icons of a large catalog at a time, so on a 33K catalog each pass
+      // cost two full rewrites. With the hint fixed, Cesium uploads only the
+      // icons that changed.
+      layerState._billboards.computeNewBuffersUsage = () => false;
       layerState._viewer.scene.primitives.add(layerState._billboards);
       registerSpriteCollection('cctv', layerState._billboards);
 
@@ -243,28 +335,76 @@ export function createLifecycle({
       // Field-test fix (2026-07-06): horizon-cull on camera settle (pairs with
       // the billboards' always-on-top depth setting) + one initial pass so the
       // first paint is already culled.
+      // A re-init against a new viewer moves the move listeners to its camera.
+      if (
+        layerState._viewListenerCamera &&
+        layerState._viewListenerCamera !== layerState._viewer.camera
+      ) {
+        const previous = layerState._viewListenerCamera;
+        if (layerState._horizonCullListener)
+          previous.moveEnd?.removeEventListener(
+            layerState._horizonCullListener,
+          );
+        if (layerState._moveStartListener)
+          previous.moveStart?.removeEventListener(
+            layerState._moveStartListener,
+          );
+        layerState._horizonCullListener = null;
+        layerState._moveStartListener = null;
+        layerState._cameraMoving = false;
+      }
+      layerState._viewListenerCamera = layerState._viewer.camera;
       if (!layerState._horizonCullListener) {
         // Ambient cards piggyback the same settle event: moveEnd-driven
         // reselection only, never per frame (refreshAmbientCards no-ops while
         // the layer is disabled).
         layerState._horizonCullListener = () => {
           layerState._cameraMoving = false;
-          parts.rendering.refreshHorizonCulling();
-          parts.cards.refreshAmbientCards();
+          stopViewMotionTimer();
+          settleView();
         };
         layerState._viewer.camera.moveEnd.addEventListener(
           layerState._horizonCullListener,
         );
       }
       if (!layerState._moveStartListener) {
-        // Item B: hover picking pauses while the camera is in motion.
+        // Item B: hover picking pauses while the camera is in motion. A long
+        // move (cockpit flight, a drag) never settles, so the view pass also
+        // reruns on an interval until it does.
         layerState._moveStartListener = () => {
           layerState._cameraMoving = true;
+          // enable() starts it for a move already under way.
+          if (layerState._enabled) startViewMotionTimer();
         };
         layerState._viewer.camera.moveStart.addEventListener(
           layerState._moveStartListener,
         );
       }
+      // A resize or rotation changes what is on screen without moving the
+      // camera.
+      const canvas = layerState._viewer.scene?.canvas;
+      if (layerState._viewResizeCanvas !== canvas) {
+        // A re-init against a new viewer observes its canvas instead.
+        layerState._viewResizeObserver?.disconnect();
+        layerState._viewResizeObserver = null;
+        layerState._viewResizeCanvas = null;
+      }
+      if (
+        !layerState._viewResizeObserver &&
+        typeof ResizeObserver === 'function' &&
+        canvas
+      ) {
+        let frame = 0;
+        layerState._viewResizeObserver = new ResizeObserver(() => {
+          if (frame || !layerState._enabled) return;
+          frame = requestAnimationFrame(() => {
+            frame = 0;
+            if (layerState._enabled) settleView();
+          });
+        });
+        layerState._viewResizeObserver.observe(canvas);
+        layerState._viewResizeCanvas = canvas;
+      }
       parts.rendering.refreshHorizonCulling();
 
       layerState._clickHandler = new Cesium.ScreenSpaceEventHandler(
@@ -325,6 +465,12 @@ export function createLifecycle({
       );
 
       await parts.health.syncHealthState(true);
+      // Health ranks crowded cells: thin again now that it has arrived (a
+      // no-op unless enable() ran while health was loading).
+      if (parts.rendering.refreshHorizonCulling()) {
+        parts.geometryQueue.enqueueViewGeometry();
+        parts.cards.refreshAmbientCards();
+      }
       parts.rendering.refreshCoverageStyles();
       parts.presentation.notifyListeners();
       restoreSpriteOrder(layerState._viewer);
@@ -370,6 +516,9 @@ export function createLifecycle({
         parts.projection.ensureProjectionRuntime(activeRecord);
         parts.frames.refreshProjectionImage(activeRecord, true);
       }
+      parts.rendering.refreshHorizonCulling();
+      // Re-enabled mid-flight: moveStart has already fired.
+      if (layerState._cameraMoving) startViewMotionTimer();
       parts.geometryQueue.startGeometryLoadQueue();
       parts.rendering.refreshCoverageStyles();
       parts.projection.startProjectionLoop();
@@ -402,10 +551,14 @@ export function createLifecycle({
       layerState._removeFocusAppearListener = null;
       parts.projection.stopProjectionLoop();
       parts.geometryQueue.stopGeometryLoadQueue();
+      stopViewMotionTimer();
       // Ambient cards tear down COMPLETELY on disable (owner design point 6):
       // source entries, pacer timer, in-flight handlers, and caches.
       parts.cards.teardownAmbientCards();
       parts.rendering.hideCctvVisuals();
+      // Nothing is in view while the layer is off.
+      layerState._viewRecords = [];
+      layerState._cameraIndexCache = null;
       parts.presentation.notifyListeners();
     },
 
@@ -443,6 +596,13 @@ export function createLifecycle({
         );
         layerState._moveStartListener = null;
       }
+      // The move listeners are gone; nothing will report the end of a move.
+      layerState._cameraMoving = false;
+      layerState._viewListenerCamera = null;
+      stopViewMotionTimer();
+      layerState._viewResizeObserver?.disconnect();
+      layerState._viewResizeObserver = null;
+      layerState._viewResizeCanvas = null;
       if (layerState._gizmo) {
         layerState._gizmo.destroy();
         layerState._gizmo = null;
diff --git a/src/layers/cctv/policy.js b/src/layers/cctv/policy.js
index bc369c02a..574f87527 100644
--- a/src/layers/cctv/policy.js
+++ b/src/layers/cctv/policy.js
@@ -60,6 +60,45 @@ export const GEO_PROGRESS_NOTIFY_INTERVAL_MS = 300;
 
 export const GEO_PROGRESS_NOTIFY_BATCH_LIMIT = 10;
 
+// View thinning: camera icons show only on or near the screen, and each
+// camera competes for a world-anchored grid cell sized to about VIEW_CELL_PX
+// at its distance from the viewer (in power-of-two metre steps, so a pan
+// keeps the same winners until a camera's distance crosses a step). Cells
+// smaller than VIEW_MIN_CELL_M do not thin:
+// close up every camera shows, co-located ones included.
+
+export const VIEW_CELL_PX = 40;
+
+export const VIEW_MIN_CELL_M = 60;
+
+// Icons this far past the screen edge (fraction of its size) still show, so
+// a pan does not reveal empty ground before the next view pass.
+
+export const VIEW_ICON_MARGIN = 0.5;
+
+// While the camera moves, the view pass reruns at most this often.
+
+export const VIEW_MOTION_REFRESH_MS = 250;
+// A motion tick is skipped unless the camera moved this fraction of its
+// ellipsoid height, or turned this many degrees, since the last pass.
+
+export const VIEW_MOTION_MIN_MOVE_FRACTION = 0.01;
+
+export const VIEW_MOTION_MIN_TURN_DEG = 0.5;
+
+// Motion passes are spaced at least this many times the last pass's cost, so
+// a large catalog spends at most about a tenth of a long move on them.
+
+export const VIEW_MOTION_COST_FACTOR = 10;
+
+// A view camera's geometry is attempted at most this many times per surface
+// regime (a cold elevation cell can leave the first attempt unresolved); the
+// retry runs this long after the drain that left it unresolved.
+
+export const VIEW_GEOMETRY_ATTEMPTS = 2;
+
+export const VIEW_GEOMETRY_RETRY_MS = 1500;
+
 // Throttle for placeholder repaints — the projection RAF loop must not
 // re-fill a 1080p canvas on every frame while a feed image is still loading.
 
diff --git a/src/layers/cctv/presentation.js b/src/layers/cctv/presentation.js
index 898e4605c..16c67aa4b 100644
--- a/src/layers/cctv/presentation.js
+++ b/src/layers/cctv/presentation.js
@@ -274,14 +274,53 @@ export function createPresentation({
       },
       activeCameraId: activeId,
       activeCamera: active ? getPublicCameraState(active, activeId) : null,
-      cameras: layerState._records.map((record) =>
-        getPublicCameraState(record, activeId),
-      ),
+      cameraIndex: cameraIndex(),
       summary: buildSummaryText(),
     };
+    // The full per-camera state is built only when a subscriber reads it: a
+    // catalog of tens of thousands would otherwise be rebuilt on every
+    // progress tick for listeners that only want the active camera. It reads
+    // the records live at that moment, and spreading or serializing the
+    // payload builds it, so callers that want a count use `count`.
+    let cameras = null;
+    Object.defineProperty(payload, 'cameras', {
+      enumerable: true,
+      get() {
+        cameras ??= layerState._records.map((record) =>
+          getPublicCameraState(record, activeId),
+        );
+        return cameras;
+      },
+    });
     return payload;
   }
 
+  /**
+   * `{ id, city, name }` for the active camera, then every camera the view
+   * pass shows: what the camera picker lists. A whole catalog would put tens
+   * of thousands of options in the DOM and slow every layout. The same array
+   * is returned until the view or the active camera changes, so a consumer
+   * can compare it by identity.
+   * @returns {{id: string, city: string, name: string}[]}
+   */
+
+  function cameraIndex() {
+    const view = layerState._viewRecords;
+    const active = parts.selection.getActiveRecord();
+    const cached = layerState._cameraIndexCache;
+    if (cached?.view === view && cached.active === active) return cached.list;
+    const records = active
+      ? [active, ...view.filter((record) => record !== active)]
+      : view;
+    const list = records.map(({ camera }) => ({
+      id: camera.id,
+      city: camera.city,
+      name: camera.name,
+    }));
+    layerState._cameraIndexCache = { view, active, list };
+    return list;
+  }
+
   /** Dispatches the current UI state to all registered subscriber callbacks. */
 
   function notifyListeners() {
diff --git a/src/layers/cctv/rendering.js b/src/layers/cctv/rendering.js
index d10368d40..4b6f215e4 100644
--- a/src/layers/cctv/rendering.js
+++ b/src/layers/cctv/rendering.js
@@ -1,5 +1,4 @@
 import * as Cesium from 'cesium';
-import { horizonOccluder } from '../../data/iconOrientation.js';
 import { isHeadingEstimated } from './headingConfidence.js';
 import {
   ACTIVE_CAMERA_COLOR,
@@ -11,8 +10,21 @@ import {
   ACTIVE_COVERAGE_EDGE,
   IDLE_COVERAGE_EDGE_MUTED,
   ACTIVE_COVERAGE_EDGE_DEPTHFAIL,
+  VIEW_CELL_PX,
+  VIEW_ICON_MARGIN,
+  VIEW_MIN_CELL_M,
 } from './policy.js';
 
+/** FNV-1a over a camera id: a stable, well-spread order for crowded cells. */
+function idHash(id) {
+  let hash = 0x811c9dc5;
+  for (let i = 0; i < id.length; i++) {
+    hash ^= id.charCodeAt(i);
+    hash = Math.imul(hash, 0x01000193);
+  }
+  return hash >>> 0;
+}
+
 export function createRendering({
   state: layerState,
   services,
@@ -138,24 +150,186 @@ export function createRendering({
    * (Google-3D regime) nothing writes far-side depth, and the billboards are now
    * always-on-top (`disableDepthTestDistance: INFINITY` — the far-zoom submerge
    * fix), so without this pass London's cluster would shine through the planet
-   * from a US viewpoint. Pure math over ≤ catalog-size points; runs on
-   * camera.moveEnd + init only (event-driven — no steady-state work).
+   * from a US viewpoint.
+   *
+   * The same pass thins the view so a large catalog stays readable and cheap:
+   * icons well off screen hide, and each camera competes for a world-anchored
+   * cell sized to VIEW_CELL_PX at its distance (healthiest first, then a stable
+   * id order). The active camera shows whenever it is above the horizon
+   * and near the screen, even in a crowded cell. The shown records become
+   * `_viewRecords`, the working set for geometry, cards and the camera picker;
+   * the array keeps its identity while the shown set is unchanged.
+   *
+   * Runs on camera settle, throttled during motion, on resize, enable and
+   * activation changes — never per frame.
+   * @returns {boolean} Whether the shown set changed.
    */
 
   function refreshHorizonCulling() {
     if (
+      !layerState._enabled ||
       !layerState._viewer ||
       layerState._viewer.isDestroyed() ||
       !layerState._records.length
     )
-      return;
-    const occluder = horizonOccluder(layerState._viewer.camera);
+      return false;
+    const { camera, scene } = layerState._viewer;
+    const width = scene.canvas.clientWidth || scene.canvas.width || 0;
+    const height = scene.canvas.clientHeight || scene.canvas.height || 0;
+    const minX = -width * VIEW_ICON_MARGIN;
+    const maxX = width * (1 + VIEW_ICON_MARGIN);
+    const minY = -height * VIEW_ICON_MARGIN;
+    const maxY = height * (1 + VIEW_ICON_MARGIN);
+    // Ground metres one pixel covers per metre of distance from the viewer.
+    const fovy = Number.isFinite(camera.frustum?.fovy)
+      ? camera.frustum.fovy
+      : Math.PI / 3;
+    const metresPerPixelPerMetre =
+      (2 * Math.tan(fovy / 2)) / Math.max(1, height);
+    const viewer = camera.positionWC;
+    // The horizon test of Cesium's EllipsoidalOccluder, inlined: this loop
+    // visits every camera in the catalog.
+    const inverseRadii = Cesium.Ellipsoid.WGS84.oneOverRadii;
+    const cx = viewer.x * inverseRadii.x;
+    const cy = viewer.y * inverseRadii.y;
+    const cz = viewer.z * inverseRadii.z;
+    const limbSquared = cx * cx + cy * cy + cz * cz - 1;
+    const activeId = layerState._activeCameraId;
+    const project = screenProjector(scene, camera, width, height);
+    const pass = (layerState._viewPassId = (layerState._viewPassId || 0) + 1);
+    const shown = [];
+    const cells = new Map();
+    for (const record of layerState._records) {
+      const position = record.billboard?.position;
+      if (!position) continue;
+      const vx = position.x * inverseRadii.x - cx;
+      const vy = position.y * inverseRadii.y - cy;
+      const vz = position.z * inverseRadii.z - cz;
+      const vtDotVc = -(vx * cx + vy * cy + vz * cz);
+      if (
+        limbSquared < 0
+          ? vtDotVc > 0
+          : vtDotVc > limbSquared &&
+            (vtDotVc * vtDotVc) / (vx * vx + vy * vy + vz * vz) > limbSquared
+      )
+        continue;
+      const screen = project(position);
+      if (
+        !screen ||
+        !(screen.x >= minX && screen.x <= maxX) ||
+        !(screen.y >= minY && screen.y <= maxY)
+      )
+        continue;
+      const dx = position.x - viewer.x;
+      const dy = position.y - viewer.y;
+      const dz = position.z - viewer.z;
+      const cellM =
+        VIEW_CELL_PX *
+        metresPerPixelPerMetre *
+        Math.sqrt(dx * dx + dy * dy + dz * dz);
+      if (record.camera.id === activeId || !(cellM >= VIEW_MIN_CELL_M)) {
+        record.viewPass = pass;
+        shown.push(record);
+        continue;
+      }
+      const key = worldCellKey(record, cellM);
+      const rank = viewRank(record);
+      const holder = cells.get(key);
+      if (!holder || rank < holder.rank) cells.set(key, { record, rank });
+    }
+    for (const { record } of cells.values()) {
+      record.viewPass = pass;
+      shown.push(record);
+    }
     for (const record of layerState._records) {
       const bb = record.billboard;
       if (!bb) continue;
-      const visible = occluder.isPointVisible(bb.position);
+      const visible = record.viewPass === pass;
       if (bb.show !== visible) bb.show = visible;
     }
+    const previous = layerState._viewRecords;
+    const changed =
+      previous.length !== shown.length ||
+      previous.some((record) => record.viewPass !== pass);
+    if (changed) layerState._viewRecords = shown;
+    return changed;
+  }
+
+  /**
+   * Canvas coordinates for world positions. In 3D the pass multiplies by one
+   * view-projection matrix per call instead of Cesium's per-point
+   * cartesianToCanvasCoordinates, which dominated the pass on a large
+   * catalog; other scene modes keep Cesium's projection.
+   * @returns {(position: Cesium.Cartesian3) => ({x: number, y: number}|null)}
+   */
+  function screenProjector(scene, camera, width, height) {
+    const out = layerState._scratchViewScreen;
+    const projection = camera.frustum?.projectionMatrix;
+    if (
+      scene.mode !== Cesium.SceneMode.SCENE3D ||
+      !projection ||
+      !camera.viewMatrix
+    )
+      return (position) => scene.cartesianToCanvasCoordinates(position, out);
+    const m = Cesium.Matrix4.multiply(
+      projection,
+      camera.viewMatrix,
+      layerState._scratchViewProjection,
+    );
+    return (p) => {
+      const w = m[3] * p.x + m[7] * p.y + m[11] * p.z + m[15];
+      const z = m[2] * p.x + m[6] * p.y + m[10] * p.z + m[14];
+      // Behind the eye or before the near plane: Cesium does not draw it.
+      if (!(w > 0) || z < 0) return null;
+      out.x =
+        ((m[0] * p.x + m[4] * p.y + m[8] * p.z + m[12]) / w + 1) * 0.5 * width;
+      out.y =
+        (1 - (m[1] * p.x + m[5] * p.y + m[9] * p.z + m[13]) / w) * 0.5 * height;
+      return out;
+    };
+  }
+
+  /**
+   * A grid cell fixed to the ground: `cellM` rounded up to a power of two
+   * metres, so the cell (and its winner) only changes when the camera's
+   * distance crosses a doubling: a level pan keeps it, a tilted pan or a zoom
+   * can move cameras near the edge of a step to the next one.
+   */
+  function worldCellKey(record, cellM) {
+    const level = Math.ceil(Math.log2(cellM));
+    const { lat, lon } = record.camera;
+    // A camera's key changes only with the level (or an edited position), so
+    // it is kept per level.
+    if (
+      record.viewCellLevel === level &&
+      record.viewCellLat === lat &&
+      record.viewCellLon === lon
+    )
+      return record.viewCellKey;
+    const latStep = 2 ** level / 111_320;
+    const row = Math.floor(lat / latStep);
+    const rowLat = Cesium.Math.toRadians((row + 0.5) * latStep);
+    const lonStep = latStep / Math.max(0.05, Math.cos(rowLat));
+    record.viewCellLevel = level;
+    record.viewCellLat = lat;
+    record.viewCellLon = lon;
+    record.viewCellKey = `${level}:${row}:${Math.floor(lon / lonStep)}`;
+    return record.viewCellKey;
+  }
+
+  /** Lower ranks win a crowded cell: healthy feeds first, then a stable id order. */
+  function viewRank(record) {
+    const status = layerState._healthById.get(record.camera.id)?.status;
+    const tier =
+      status === 'ok'
+        ? 0
+        : status === 'degraded'
+          ? 1
+          : status === 'offline'
+            ? 3
+            : 2;
+    record.viewHash ??= idHash(record.camera.id);
+    return tier * 2 ** 32 + record.viewHash;
   }
 
   /**
@@ -204,6 +378,8 @@ export function createRendering({
     const activeId = activeRecord?.camera.id || null;
     const coverageVisible =
       parts.geometry.buildCoverageVisibleSet(activeRecord);
+    // Drawn coverage needs geometry too (geometryQueue's working set).
+    layerState._coverageVisibleIds = coverageVisible;
     const coverageOn = layerState._coverageMode !== 'off';
     const viewshedOn = layerState._coverageMode === 'viewshed';
     if (coverageOn) {
@@ -215,8 +391,13 @@ export function createRendering({
     for (const record of layerState._records) {
       const isActive = record.camera.id === activeId;
       if (record.billboard) {
-        record.billboard.color = cameraColorFor(record);
-        record.billboard.scale = isActive ? 1.25 : 1.0;
+        // Unchanged writes still dirty the collection, which then rewrites
+        // every icon's vertices: across a large catalog, write only changes.
+        const color = cameraColorFor(record);
+        if (!Cesium.Color.equals(record.billboard.color, color))
+          record.billboard.color = color;
+        const scale = isActive ? 1.25 : 1.0;
+        if (record.billboard.scale !== scale) record.billboard.scale = scale;
         // disableDepthTestDistance stays POSITIVE_INFINITY for every billboard
         // (set at creation) — see the field-test far-zoom submerge fix there.
       }
diff --git a/src/layers/cctv/selection.js b/src/layers/cctv/selection.js
index 317d80c7e..c86425fd0 100644
--- a/src/layers/cctv/selection.js
+++ b/src/layers/cctv/selection.js
@@ -141,6 +141,11 @@ export function createSelection({
     void parts.ground.resolveFootprintGround(record);
     record.activationDone = true;
     parts.rendering.refreshCoverageStyles();
+    // The active camera's icon always shows; the previously active one
+    // competes for its cell again. Its coverage neighbours join the working
+    // set and are queued for grounding.
+    parts.rendering.refreshHorizonCulling();
+    parts.geometryQueue.enqueueViewGeometry();
     // The newly active camera leaves the ambient ring (its monitor plane takes
     // over); the freed slot re-fills on this same pass.
     parts.cards.refreshAmbientCards();
@@ -171,7 +176,9 @@ export function createSelection({
         parts.ground.groundAltFor(previous),
       );
     });
+    parts.rendering.refreshHorizonCulling();
     parts.rendering.refreshCoverageStyles();
+    parts.geometryQueue.enqueueViewGeometry();
     parts.cards.refreshAmbientCards();
     layerState._gizmo?.refresh();
     parts.presentation.notifyListeners();
diff --git a/src/layers/cctv/state.js b/src/layers/cctv/state.js
index b78d18b73..78c2b8fa9 100644
--- a/src/layers/cctv/state.js
+++ b/src/layers/cctv/state.js
@@ -76,6 +76,63 @@ export function createState({ services }) {
 
   layerState._scratchFocusScreen = new Cesium.Cartesian2();
 
+  layerState._scratchViewScreen = new Cesium.Cartesian2();
+
+  layerState._scratchViewProjection = new Cesium.Matrix4();
+
+  // Counts view passes; a record shown by the latest pass carries its number.
+
+  layerState._viewPassId = 0;
+
+  // Records whose icons the last view pass showed (see refreshHorizonCulling):
+  // geometry refinement, the tiles-ready completion pass and ambient cards
+  // work over these, not the whole catalog.
+
+  layerState._viewRecords = [];
+
+  // Coverage ids refreshCoverageStyles last drew: part of the geometry
+  // working set even when their icons are thinned out.
+
+  layerState._coverageVisibleIds = new Set();
+
+  // View pass triggers besides camera settle: an interval while the camera
+  // moves, and a canvas resize observer.
+
+  layerState._viewMotionTimer = 0;
+
+  layerState._viewResizeObserver = null;
+
+  layerState._viewResizeCanvas = null;
+
+  // The camera the moveStart/moveEnd listeners are attached to.
+
+  layerState._viewListenerCamera = null;
+
+  // A motion pass changed the shown set and the settle pass has not yet told
+  // subscribers; the camera pose the last view pass ran at.
+
+  layerState._viewUnpublished = false;
+
+  layerState._viewPose = null;
+
+  // { at, ms } — when the last motion pass ran and what it cost.
+
+  layerState._viewMotionPass = null;
+
+  // Retries the view geometry a drain left unresolved, once cells warm.
+
+  layerState._viewGeometryRetryTimer = 0;
+
+  // { key, count } — the active camera's coverage-overlap count, which the
+  // summary reads on every notification.
+
+  layerState._neighborCountCache = null;
+
+  // { view, active, list } — the camera picker's index uiState publishes,
+  // rebuilt only when the view pass or the active camera changes.
+
+  layerState._cameraIndexCache = null;
+
   // Staggered geometry-load queue state (see startGeometryLoadQueue).
 
   layerState._geoQueue = [];
diff --git a/src/layers/cctv/testing.js b/src/layers/cctv/testing.js
index 5c1e67956..c4aada9bb 100644
--- a/src/layers/cctv/testing.js
+++ b/src/layers/cctv/testing.js
@@ -79,6 +79,7 @@ export function createTesting({ state: layerState, services, parts, source }) {
    * @param {boolean} [options.enabled=true] Layer enabled state.
    * @param {'off'|'on'|'viewshed'} [options.coverageMode='on'] Coverage mode.
    * @param {boolean} [options.showProjection=false] Projection visibility.
+   * @param {Map<string, Object>} [options.healthById] Health rows by camera id.
    * @returns {void}
    */
 
@@ -89,6 +90,7 @@ export function createTesting({ state: layerState, services, parts, source }) {
     enabled = true,
     coverageMode = 'on',
     showProjection = false,
+    healthById = new Map(),
   } = {}) {
     layerState._viewer = viewer;
     layerState._records = Array.isArray(records) ? records : [];
@@ -108,6 +110,11 @@ export function createTesting({ state: layerState, services, parts, source }) {
       'on',
     );
     layerState._showProjection = !!showProjection;
+    layerState._healthById = healthById;
+    layerState._viewRecords = [];
+    layerState._cameraIndexCache = null;
+    layerState._coverageVisibleIds = new Set();
+    layerState._neighborCountCache = null;
   }
   return {
     _pushAmbientCardEntriesForTest,
diff --git a/src/ui/cctvBindings.js b/src/ui/cctvBindings.js
index d79a9e99a..717770989 100644
--- a/src/ui/cctvBindings.js
+++ b/src/ui/cctvBindings.js
@@ -59,6 +59,12 @@ export function _initCctvPanel() {
     );
   });
 
+  // The picker defers rebuilds while focused; catch up once it lets go.
+  this.listen(this._cctvSelect, 'blur', () => {
+    this._cctvSelectSource = null;
+    this._renderCctvState(this._cctvState);
+  });
+
   this.listen(this._cctvSelect, 'change', async () => {
     const generation = ++this._actionGeneration;
     const activeId = this._cctvState?.activeCameraId;
@@ -72,9 +78,9 @@ export function _initCctvPanel() {
       (activeId && activeId !== this._cctvState?.activeCameraId)
     )
       return;
-    // Picking a camera from the dropdown flies to it. The catalog spans
-    // three metros, so a bare selection used to leave the view in the old
-    // city with a camera active thousands of km away.
+    // Picking a camera from the dropdown flies to it. The dropdown lists the
+    // cameras in view (and the active camera, which may be far away), so a
+    // bare selection could leave the view somewhere else.
     this.actions.runExplicitFocus(
       () => (this.cctv.selectCamera(cameraId) ? cameraId : null),
       (selectedId) => this.cctv.focusCamera(selectedId, 2.2),
diff --git a/src/ui/cctvPicker.test.mjs b/src/ui/cctvPicker.test.mjs
new file mode 100644
index 000000000..c96fdeacb
--- /dev/null
+++ b/src/ui/cctvPicker.test.mjs
@@ -0,0 +1,160 @@
+// src/ui/cctvPicker.test.mjs — the camera picker lists the cameras in view and
+// says how many more the view leaves out.
+import test from 'node:test';
+import assert from 'node:assert/strict';
+import { CctvControls } from './cctvControls.js';
+
+/** The slice of a <select> and its document the picker touches. */
+function pickerDocument() {
+  const select = new EventTarget();
+  const option = () => ({
+    value: '',
+    textContent: '',
+    disabled: false,
+    dataset: {},
+    remove() {
+      select.options = select.options.filter((entry) => entry !== this);
+    },
+  });
+  Object.assign(select, {
+    options: [],
+    disabled: false,
+    selectedIndex: -1,
+    get value() {
+      return this.options[this.selectedIndex]?.value ?? '';
+    },
+    set value(id) {
+      this.selectedIndex = this.options.findIndex((o) => o.value === id);
+    },
+    replaceChildren(fragment) {
+      this.options = [...fragment.children];
+      this.selectedIndex = this.options.length ? 0 : -1;
+    },
+    prepend(child) {
+      this.options.unshift(child);
+    },
+  });
+  const document = {
+    activeElement: null,
+    createElement: option,
+    createDocumentFragment: () => ({
+      children: [],
+      appendChild(child) {
+        this.children.push(child);
+      },
+    }),
+  };
+  return { document, select };
+}
+
+function fixture(t) {
+  const { document, select } = pickerDocument();
+  const previous = globalThis.document;
+  globalThis.document = document;
+  const controls = new CctvControls({
+    elements: { _cctvSelect: select },
+    cctv: {},
+    actions: { isEnabled: () => true, setPanelCollapsed() {} },
+  });
+  t.after(() => {
+    controls.destroy();
+    globalThis.document = previous;
+  });
+  return { controls, select };
+}
+
+const camera = (id) => ({ id, city: 'Austin', name: `Cam ${id}` });
+
+test('the picker closes with the count of cameras outside the view', (t) => {
+  const { controls, select } = fixture(t);
+  controls._renderCctvState({
+    enabled: true,
+    count: 40,
+    cameraIndex: [camera('a'), camera('b')],
+  });
+  assert.deepEqual(
+    select.options.map((o) => o.value),
+    ['a', 'b', ''],
+  );
+  const note = select.options[2];
+  assert.equal(note.disabled, true);
+  assert.equal(
+    note.textContent,
+    '38 more cameras outside this view · pan or zoom out',
+  );
+  assert.equal(select.selectedIndex, -1, 'nothing active, nothing picked');
+
+  controls._renderCctvState({
+    enabled: true,
+    count: 2,
+    cameraIndex: [camera('a'), camera('b')],
+  });
+  assert.deepEqual(
+    select.options.map((o) => o.value),
+    ['a', 'b'],
+    'a view holding every camera has no note',
+  );
+});
+
+test('with no camera in view, the closed picker shows the note', (t) => {
+  const { controls, select } = fixture(t);
+  controls._renderCctvState({ enabled: true, count: 1, cameraIndex: [] });
+  assert.equal(select.options.length, 1);
+  assert.equal(
+    select.options[select.selectedIndex].textContent,
+    '1 more camera outside this view · pan or zoom out',
+  );
+});
+
+test('a disabled layer has no note', (t) => {
+  const { controls, select } = fixture(t);
+  controls._renderCctvState({ enabled: false, count: 40, cameraIndex: [] });
+  assert.equal(select.options.length, 0);
+  assert.equal(select.selectedIndex, -1);
+});
+
+test('a camera activated under a focused picker counts against the note', (t) => {
+  const { controls, select } = fixture(t);
+  const a = camera('a');
+  const b = camera('b');
+  controls._renderCctvState({
+    enabled: true,
+    count: 3,
+    cameraIndex: [a],
+    activeCameraId: 'a',
+    activeCamera: a,
+  });
+  assert.equal(
+    select.options.at(-1).textContent,
+    '2 more cameras outside this view · pan or zoom out',
+  );
+  globalThis.document.activeElement = select;
+  controls._renderCctvState({
+    enabled: true,
+    count: 3,
+    cameraIndex: [b, a],
+    activeCameraId: 'b',
+    activeCamera: b,
+  });
+  assert.deepEqual(
+    select.options.map((o) => o.value),
+    ['b', 'a', ''],
+  );
+  assert.equal(
+    select.options.at(-1).textContent,
+    '1 more camera outside this view · pan or zoom out',
+  );
+  assert.equal(select.value, 'b');
+});
+
+test('a focused picker does not pick a stale camera when the view empties', (t) => {
+  const { controls, select } = fixture(t);
+  controls._renderCctvState({
+    enabled: true,
+    count: 3,
+    cameraIndex: [camera('a')],
+  });
+  globalThis.document.activeElement = select;
+  controls._renderCctvState({ enabled: true, count: 3, cameraIndex: [] });
+  assert.equal(select.selectedIndex, -1);
+});
diff --git a/src/ui/cctvPresentation.js b/src/ui/cctvPresentation.js
index 070c47f64..5cf311155 100644
--- a/src/ui/cctvPresentation.js
+++ b/src/ui/cctvPresentation.js
@@ -16,6 +16,15 @@ export function _calBadgeLabel(badge) {
   }
 }
 
+/** The picker's closing line for the cameras the view leaves out. */
+function outsideNoteText(count) {
+  return tn(
+    count,
+    '{count} more camera outside this view · pan or zoom out',
+    '{count} more cameras outside this view · pan or zoom out',
+  );
+}
+
 export function _renderCctvState(state) {
   if (this.destroyed) return;
   if (
@@ -26,7 +35,13 @@ export function _renderCctvState(state) {
     this._calibrationEdit?.(false);
   }
   this._cctvState = state || null;
-  const cameras = state?.cameras || [];
+  // The light index, not `state.cameras`: reading that builds every camera's
+  // full state, which a large catalog cannot afford on each notification.
+  const cameras = state?.cameraIndex || state?.cameras || [];
+  // The picker lists what is in view; the catalog is every camera.
+  const catalogCount = Number.isFinite(state?.count)
+    ? state.count
+    : cameras.length;
   const enabled = !!state?.enabled && !!this.actions.isEnabled();
   const activeId = state?.activeCameraId || '';
   const activeCamera = state?.activeCamera || null;
@@ -59,28 +74,79 @@ export function _renderCctvState(state) {
   }
 
   if (this._cctvSelect) {
+    // The index keeps its identity until the view or the active camera
+    // changes, so an unchanged list skips the per-option comparison. A
+    // focused picker is not rebuilt under the user; blur re-renders it.
+    const picking = document.activeElement === this._cctvSelect;
+    // Cameras the view leaves out get one closing line, so a short list
+    // does not read as the whole catalog. A disabled layer has no view.
+    const outside = enabled ? Math.max(0, catalogCount - cameras.length) : 0;
     const shouldRebuild =
-      this._cctvSelect.options.length !== cameras.length ||
-      cameras.some(
-        (cam, idx) => this._cctvSelect.options[idx]?.value !== cam.id,
-      );
+      !picking &&
+      (this._cctvSelectSource !== cameras ||
+        this._cctvSelectOutside !== outside) &&
+      (this._cctvSelectOutside !== outside ||
+        this._cctvSelect.options.length !==
+          cameras.length + (outside ? 1 : 0) ||
+        cameras.some(
+          (cam, idx) => this._cctvSelect.options[idx]?.value !== cam.id,
+        ));
+    if (!picking) {
+      this._cctvSelectSource = cameras;
+      this._cctvSelectOutside = outside;
+    }
     if (shouldRebuild) {
-      this._cctvSelect.innerHTML = '';
+      const options = document.createDocumentFragment();
       for (const camera of cameras) {
         const option = document.createElement('option');
         option.value = camera.id;
         option.textContent = `${camera.city} · ${camera.name}`;
-        this._cctvSelect.appendChild(option);
+        options.appendChild(option);
       }
+      if (outside) {
+        const note = document.createElement('option');
+        note.value = '';
+        note.disabled = true;
+        note.dataset.cctvOutside = '';
+        note.textContent = outsideNoteText(outside);
+        options.appendChild(note);
+      }
+      this._cctvSelect.replaceChildren(options);
     }
     this._cctvSelect.disabled = !enabled || cameras.length === 0;
+    // A deferred rebuild can miss a camera activated meanwhile (auto-hop,
+    // voice); add it so the picker does not go blank.
     if (
+      picking &&
       activeId &&
-      Array.from(this._cctvSelect.options).some((opt) => opt.value === activeId)
+      activeCamera &&
+      ![...this._cctvSelect.options].some((option) => option.value === activeId)
     ) {
+      const option = document.createElement('option');
+      option.value = activeId;
+      option.textContent = `${activeCamera.city} · ${activeCamera.name}`;
+      this._cctvSelect.prepend(option);
+      // The added camera is one fewer outside the view.
+      const note = [...this._cctvSelect.options].find(
+        (entry) => entry.dataset?.cctvOutside !== undefined,
+      );
+      if (note) {
+        const left = Math.max(
+          0,
+          catalogCount - (this._cctvSelect.options.length - 1),
+        );
+        if (left) note.textContent = outsideNoteText(left);
+        else note.remove();
+      }
+    }
+    // The active camera is always one of the listed records.
+    if (activeId && activeCamera) {
       this._cctvSelect.value = activeId;
     } else if (!activeId) {
-      this._cctvSelect.selectedIndex = -1;
+      // With nothing in view, the closed picker says where the cameras are.
+      // A focused picker keeps its old options, so it only clears.
+      this._cctvSelect.selectedIndex =
+        !picking && cameras.length === 0 && outside ? 0 : -1;
     }
   }
 
@@ -90,10 +156,10 @@ export function _renderCctvState(state) {
     this._cctvNextBtn,
   ]) {
     if (!btn) continue;
-    btn.disabled = !enabled || cameras.length === 0;
+    btn.disabled = !enabled || catalogCount === 0;
   }
   if (this._cctvFocusBtn) {
-    this._cctvFocusBtn.disabled = !enabled || cameras.length === 0 || !activeId;
+    this._cctvFocusBtn.disabled = !enabled || catalogCount === 0 || !activeId;
   }
 
   if (this._cctvCoverageBtn) {
@@ -180,15 +246,15 @@ export function _renderCctvState(state) {
           status: `${credit}${statusMsg}`,
         },
       );
-    } else if (cameras.length > 0) {
+    } else if (catalogCount > 0) {
       this._cctvMeta.textContent = enabled
         ? tn(
-            cameras.length,
+            catalogCount,
             '{count} camera loaded · click a camera to activate',
             '{count} cameras loaded · click a camera to activate',
           )
         : tn(
-            cameras.length,
+            catalogCount,
             '{count} camera loaded · enable CCTV to activate',
             '{count} cameras loaded · enable CCTV to activate',
           );
diff --git a/src/voice/gevActions.js b/src/voice/gevActions.js
index 0fd17dc77..299fbfc61 100644
--- a/src/voice/gevActions.js
+++ b/src/voice/gevActions.js
@@ -1659,9 +1659,8 @@ export async function controlCctv(dataManager, args = {}, styleManager = null) {
     return {
       activeCameraId: ui.activeCameraId || null,
       activeCamera: ui.activeCamera?.name || ui.activeCamera?.id || null,
-      cameraCount: Array.isArray(ui.cameras)
-        ? ui.cameras.length
-        : ui.count || 0,
+      // `count`, not `cameras.length`: reading `cameras` builds every camera.
+      cameraCount: Number.isFinite(ui.count) ? ui.count : 0,
       showCoverage: !!ui.showCoverage,
       coverageMode: ui.coverageMode || (ui.showCoverage ? 'on' : 'off'),
       showProjection: !!ui.showProjection,