Merge pull request 'fix(web): remove prompt registry shortcut' (#57)

fix(web): remove prompt registry shortcut

.forgejo/workflows/dimensionlab-website.yml

@@ -0,0 +1,85 @@
+name: Dimension Lab website
+
+on:
+  pull_request:
+    types:
+      - opened
+      - synchronize
+      - reopened
+  push:
+    branches:
+      - main
+  workflow_dispatch:
+
+concurrency:
+  group: dimensionlab-website-${{ github.ref }}
+  cancel-in-progress: ${{ github.ref != 'refs/heads/main' }}
+
+jobs:
+  ci:
+    runs-on: docker
+    timeout-minutes: 30
+    steps:
+      - name: Checkout
+        uses: https://data.forgejo.org/actions/checkout@v4
+        with:
+          fetch-depth: 0
+          submodules: false
+
+      - name: Initialize submodules
+        run: |
+          git config --global url."https://git.dimensionlab.net/".insteadOf "ssh://git@git.dimensionlab.net/"
+          git submodule update --init --recursive
+
+      - name: Install Bun
+        run: |
+          curl -fsSL https://bun.sh/install | bash -s "bun-v1.3.14"
+          "$HOME/.bun/bin/bun" --version
+
+      - name: Check, test, and build
+        run: |
+          export BUN_INSTALL="$HOME/.bun"
+          export PATH="$BUN_INSTALL/bin:$PATH"
+          bun install --frozen-lockfile
+          bun run check
+          bun run test
+          bun run build
+
+  deploy:
+    needs: ci
+    if: github.event_name == 'push' && github.ref == 'refs/heads/main'
+    runs-on: deploy
+    timeout-minutes: 30
+    steps:
+      - name: Checkout
+        run: |
+          if [ -d .git ]; then
+            git remote set-url origin git@git.dimensionlab.net:vince/dimensionlab-website.git
+          else
+            git init
+            git remote add origin git@git.dimensionlab.net:vince/dimensionlab-website.git
+          fi
+          git fetch --force --prune --depth=1 origin "$GITHUB_SHA"
+          git checkout --force --detach "$GITHUB_SHA"
+          git clean -ffdx
+
+      - name: Initialize submodules
+        run: |
+          git config --global url."https://git.dimensionlab.net/".insteadOf "ssh://git@git.dimensionlab.net/"
+          git submodule update --init --recursive
+
+      - name: Verify Podman deployment socket
+        run: |
+          command -v podman
+          command -v systemctl
+          unit="$(timeout 15s podman inspect dimensionlab-website --format '{{ index .Config.Labels "PODMAN_SYSTEMD_UNIT" }}')"
+          test "$unit" = "dimensionlab-website.service"
+
+      - name: Deploy production website
+        env:
+          DEPLOY_CONTAINER_CLI: podman
+          DEPLOY_EVENT_NAME: ${{ github.event_name }}
+          DEPLOY_REF: ${{ github.ref }}
+          DEPLOY_RESTART_STRATEGY: quadlet-container
+          DEPLOY_SHA: ${{ github.sha }}
+        run: scripts/deploy-dimensionlab-website.sh

README.md

@@ -161,3 +161,27 @@ The env file must provide `AGENT_CONFIG_TOKEN`. Runtime defaults inside the
 image set `HOST=0.0.0.0`, `PORT=3000`,
 `DATABASE_URL=file:/data/dimensionlab.sqlite`, and
 `DASHBOARD_MIGRATIONS_DIR=/repo/apps/web/drizzle`.
+
+### Forgejo Actions Deployment
+
+Merges to `main` run `.forgejo/workflows/dimensionlab-website.yml`. Pull
+requests run check, test, and build only; the deploy job is guarded to run only
+for `push` events on `refs/heads/main`.
+
+The workflow uses two runner classes. Pull request CI runs on the containerized
+`docker` runner. Production deployment runs on a separate host runner with the
+`deploy:host` label so the guarded deploy script can use the user's rootless
+`podman` and `systemctl --user` commands directly. The deploy job uses a
+shell-only `git fetch` checkout so the host runner does not need a Node runtime
+for checkout actions.
+
+```yaml
+runner:
+  labels:
+    - deploy:host
+```
+
+The deploy job also performs a host preflight against the
+`dimensionlab-website.service` Podman label before it builds or restarts the
+production container. Do not give the general pull request runner deployment
+socket access; keep deploy privileges on the dedicated `deploy` runner.

apps/web/package.json

@@ -8,8 +8,8 @@
     "build": "rm -rf build && vite build && bun build src/server/index.ts --target bun --outdir build",
     "preview": "HOST=0.0.0.0 PORT=4173 bun build/index.js",
     "check": "tsc --noEmit",
-    "test": "vitest run",
+    "test": "bun --bun vitest run",
-    "test:unit": "vitest run",
+    "test:unit": "bun --bun vitest run",
     "test:e2e": "env -u NO_COLOR playwright test",
     "db:generate": "drizzle-kit generate",
     "db:check": "drizzle-kit check"

apps/web/src/App.test.tsx

@@ -2,7 +2,12 @@ import { renderToString } from "react-dom/server";
 import { describe, expect, test } from "vitest";
 import { genericDashboardFixture } from "@dimensionlab/dashboard-model/fixtures";
 import { dimensionLabDashboardFixture } from "$lib/dashboard-seed/dimensionlab";
-import { AppStateView, dashboardHydrationTiles } from "./App";
+import {
+  AppStateView,
+  dashboardTileMatchKey,
+  dashboardHydrationTiles,
+  restoreDashboardTileSnapshots,
+} from "./App";
 
 describe("React app dashboard state view", () => {
   test("renders loading dashboard state", () => {
@@ -82,4 +87,152 @@ describe("React app dashboard state view", () => {
       id: "auto-refresh",
     });
   });
+
+  test("prioritizes status, telemetry, modules, then services for hydration", () => {
+    const tiles = dashboardHydrationTiles(dimensionLabDashboardFixture);
+    const kinds = tiles.map((tile) => tile.kind);
+    const firstServiceIndex = kinds.indexOf("service");
+
+    expect(kinds.slice(0, 5)).toEqual([
+      "status",
+      "status",
+      "status",
+      "status",
+      "status",
+    ]);
+    expect(kinds.lastIndexOf("telemetry")).toBeLessThan(kinds.indexOf("module"));
+    expect(kinds.lastIndexOf("module")).toBeLessThan(firstServiceIndex);
+    expect(
+      tiles.filter((tile) => tile.kind === "telemetry").map((tile) => tile.id),
+    ).toEqual(
+      dimensionLabDashboardFixture.telemetry
+        .filter((card) => card.datasource?.type === "external")
+        .map((card) => card.id),
+    );
+    expect(
+      tiles.filter((tile) => tile.kind === "service").map((tile) => ({
+        groupId: tile.groupId,
+        id: tile.id,
+      })),
+    ).toEqual(
+      dimensionLabDashboardFixture.serviceGroups.flatMap((group) =>
+        group.services
+          .filter((service) => service.datasource?.type === "external")
+          .map((service) => ({
+            groupId: group.id,
+            id: service.id,
+          }))
+      ),
+    );
+  });
+
+  test("applies restored tile snapshots without marking them as loading", () => {
+    const restored = restoreDashboardTileSnapshots(
+      {
+        state: "ready",
+        document: dimensionLabDashboardFixture,
+        schemaVersion: "dashboard.v1",
+        currentRevisionId: "revision-a",
+      },
+      [
+        {
+          ageMs: 60_000,
+          response: {
+            state: "ready",
+            tile: { kind: "telemetry", id: "infra-ram" },
+            item: {
+              ...dimensionLabDashboardFixture.telemetry[0],
+              detail: "cached - stale 60s",
+              severity: "stale",
+            },
+          },
+        },
+      ],
+    );
+
+    expect(restored.restoredItemIds).toEqual(new Set(["telemetry:infra-ram"]));
+    if (restored.dashboard.state !== "ready") {
+      throw new Error("Expected dashboard to be ready");
+    }
+    expect(restored.dashboard.document.telemetry[0]).toMatchObject({
+      id: "infra-ram",
+      detail: "cached - stale 60s",
+      severity: "stale",
+    });
+  });
+
+  test("does not restore aggregate health snapshots over the fresh shell", () => {
+    const restored = restoreDashboardTileSnapshots(
+      {
+        state: "ready",
+        document: dimensionLabDashboardFixture,
+        schemaVersion: "dashboard.v1",
+        currentRevisionId: "revision-a",
+      },
+      [
+        {
+          ageMs: 60_000,
+          response: {
+            state: "ready",
+            tile: { kind: "status", stripId: "footer-status", id: "system-status" },
+            item: {
+              id: "system-status",
+              label: "System Status",
+              value: "20 services down",
+              severity: "stale",
+            },
+          },
+        },
+        {
+          ageMs: 60_000,
+          response: {
+            state: "ready",
+            tile: { kind: "module", id: "runtime-health-summary" },
+            item: {
+              id: "runtime-health-summary",
+              kind: "summary",
+              title: "Runtime Health",
+              value: "20 services down",
+              detail: "0 warnings - 8 services ok - stale 60s",
+              severity: "stale",
+            },
+          },
+        },
+      ],
+    );
+
+    expect(restored.restoredItemIds).toEqual(new Set());
+    if (restored.dashboard.state !== "ready") {
+      throw new Error("Expected dashboard to be ready");
+    }
+    expect(
+      restored.dashboard.document.statusStrips[0].items.find((item) =>
+        item.id === "system-status"
+      ),
+    ).toMatchObject({
+      id: "system-status",
+      value: "Fallback operational",
+    });
+    expect(
+      restored.dashboard.document.modules?.find((module) =>
+        module.id === "runtime-health-summary"
+      ),
+    ).toMatchObject({
+      id: "runtime-health-summary",
+      value: "fallback",
+    });
+  });
+
+  test("uses structured tile match keys for delimiter-bearing ids", () => {
+    expect(
+      dashboardTileMatchKey({ kind: "service", groupId: "a:b", id: "c" }),
+    ).not.toBe(
+      dashboardTileMatchKey({ kind: "service", groupId: "a", id: "b:c" }),
+    );
+    expect(
+      dashboardTileMatchKey({ kind: "status", stripId: "a:b", id: "c" }),
+    ).not.toBe(
+      dashboardTileMatchKey({ kind: "status", stripId: "a", id: "b:c" }),
+    );
+  });
 });

apps/web/src/App.tsx

@@ -9,8 +9,21 @@ import type {
 import type { DashboardRuntimeState } from "$lib/server/dashboard";
 import {
   attachDashboardRefreshLifecycle,
+  collectVisibleDashboardModelIds,
+  createDashboardPerformanceMarks,
+  createDashboardRefreshDelay,
+  createDashboardTileSnapshotStore,
+  createDashboardTileBackoff,
   createDashboardRequestAborter,
+  isPersistableDashboardTileSnapshot,
+  runViewportAwareDashboardHydrationQueue,
   shouldPauseDashboardRefresh,
+  subscribeToDashboardTileEvents,
+  waitForDashboardHydrationIdle,
+  type DashboardIntersectionObserverFactory,
+  type DashboardTileReference,
+  type DashboardTileSnapshotStoreContext,
+  type RestoredDashboardTileSnapshot,
 } from "$lib/client/dashboard-refresh";
 import { dashboardDocumentToUiDashboard } from "$lib/ui-adapter/model-renderer";
 import {
@@ -24,12 +37,6 @@ import {
   type UiDashboardPreview,
 } from "@dimensionlab/ui";
 
-type DashboardTileReference =
-  | { kind: "telemetry"; id: string }
-  | { kind: "service"; groupId: string; id: string }
-  | { kind: "module"; id: string }
-  | { kind: "status"; stripId: string; id: string };
-
 type DashboardTileResponse =
   | {
       state: "ready";
@@ -40,8 +47,25 @@ type DashboardTileResponse =
       state: "not_found";
       tile: DashboardTileReference;
       message: string;
+    }
+  | {
+      state: "disabled";
+      tile: DashboardTileReference;
+      message: string;
     };
 
+type DashboardTilesBatchResponse = {
+  state: "ready";
+  tiles: DashboardTileResponse[];
+};
+
+type DashboardTileHydrationResult = "aborted" | "failed" | "ready";
+
+const dashboardTileHydrationConcurrency = 6;
+const dashboardFallbackRefreshIntervalMs = 30_000;
+const dashboardViewportObservationTimeoutMs = 80;
+type DashboardTileSnapshotStore = ReturnType<typeof createDashboardTileSnapshotStore>;
+
 export function AppStateView({
   dashboard,
   onThemeChange,
@@ -155,8 +179,16 @@ export default function App() {
   useEffect(() => {
     let cancelled = false;
     let refreshTimer: number | undefined;
+    let unsubscribeTileEvents: (() => void) | undefined;
+    let lastRefreshIntervalMs = dashboardFallbackRefreshIntervalMs;
     let hydrationRun = 0;
     const requestAborter = createDashboardRequestAborter();
+    const performanceMarks = createDashboardPerformanceMarks();
+    const refreshDelay = createDashboardRefreshDelay();
+    const tileBackoff = createDashboardTileBackoff();
+    const tileSnapshotStore = createDashboardTileSnapshotStore(
+      getTileSnapshotStorage(),
+    );
 
     function refreshPaused() {
       return shouldPauseDashboardRefresh({
@@ -166,14 +198,26 @@ export default function App() {
     }
 
     function clearRefreshTimer() {
-      if (refreshTimer) {
+      if (refreshTimer !== undefined) {
-        window.clearInterval(refreshTimer);
+        window.clearTimeout(refreshTimer);
         refreshTimer = undefined;
       }
     }
 
+    function scheduleNextRefresh(baseDelayMs: number) {
+      clearRefreshTimer();
+      if (cancelled || refreshPaused()) return;
+
+      refreshTimer = window.setTimeout(() => {
+        refreshTimer = undefined;
+        void loadDashboard();
+      }, refreshDelay.nextDelayMs(baseDelayMs));
+    }
+
     function pauseRefreshes() {
       clearRefreshTimer();
+      unsubscribeTileEvents?.();
+      unsubscribeTileEvents = undefined;
       requestAborter.abortActiveRequests();
       setHydratingItemIds(new Set());
     }
@@ -181,6 +225,7 @@ export default function App() {
     async function loadDashboard() {
       if (cancelled || refreshPaused()) return;
 
+      clearRefreshTimer();
       const shellSignal = requestAborter.beginShellRun();
 
       try {
@@ -188,37 +233,70 @@ export default function App() {
         const nextDashboard = (await response.json()) as DashboardRuntimeState;
 
         if (cancelled || shellSignal.aborted || refreshPaused()) return;
-        setDashboard(nextDashboard);
+        refreshDelay.recordSuccess();
+        performanceMarks.markShellLoad();
+        const restored = restoreDashboardTileSnapshots(
+          nextDashboard,
+          nextDashboard.state === "ready"
+            ? tileSnapshotStore.restore({
+                currentRevisionId: nextDashboard.currentRevisionId,
+                schemaVersion: nextDashboard.schemaVersion,
+              })
+            : [],
+        );
+        setDashboard(restored.dashboard);
 
         const currentRun = ++hydrationRun;
         if (
-          nextDashboard.state === "ready" &&
+          restored.dashboard.state === "ready" &&
-          nextDashboard.liveDatasourceHydration?.enabled !== false
+          restored.dashboard.liveDatasourceHydration?.enabled !== false
         ) {
-          const tiles = dashboardHydrationTiles(nextDashboard.document);
+          const tiles = dashboardHydrationTiles(restored.dashboard.document).filter(
+            (tile) => tileBackoff.canAttempt(dashboardTileKey(tile)),
+          );
           const tileSignal = requestAborter.beginTileRun();
-          setHydratingItemIds(new Set(tiles.map(dashboardTileKey)));
+          setHydratingItemIds(
-          hydrateDashboardTiles(tiles, currentRun, tileSignal);
+            new Set(
+              tiles
+                .map(dashboardTileKey)
+                .filter((key) => !restored.restoredItemIds.has(key)),
+            ),
+          );
+          hydrateDashboardTiles(
+            tiles,
+            currentRun,
+            tileSignal,
+            {
+              currentRevisionId: restored.dashboard.currentRevisionId,
+              schemaVersion: restored.dashboard.schemaVersion,
+            },
+          );
+          subscribeDashboardTileEvents(
+            currentRun,
+            tileSignal,
+            {
+              currentRevisionId: restored.dashboard.currentRevisionId,
+              schemaVersion: restored.dashboard.schemaVersion,
+            },
+          );
         } else {
           setHydratingItemIds(new Set());
         }
 
-        clearRefreshTimer();
-
         const refreshIntervalSeconds =
           nextDashboard.state === "ready"
             ? nextDashboard.document.metadata.refreshIntervalSeconds
             : undefined;
 
         if (refreshIntervalSeconds && !refreshPaused()) {
-          refreshTimer = window.setInterval(
+          lastRefreshIntervalMs = refreshIntervalSeconds * 1000;
-            () => void loadDashboard(),
+          scheduleNextRefresh(lastRefreshIntervalMs);
-            refreshIntervalSeconds * 1000,
-          );
         }
       } catch (error) {
         if (!isAbortError(error) && !cancelled) {
+          refreshDelay.recordFailure();
           console.error("Dashboard refresh failed", error);
+          scheduleNextRefresh(lastRefreshIntervalMs);
         }
       }
     }
@@ -227,31 +305,218 @@ export default function App() {
       tiles: DashboardTileReference[],
       run: number,
       signal: AbortSignal,
+      snapshotContext: DashboardTileSnapshotStoreContext,
     ) {
-      tiles.forEach((tile) => {
+      const modelIds = tiles.map(dashboardTileModelId);
-        void hydrateDashboardTile(tile, run, signal);
+
+      void runViewportAwareDashboardHydrationQueue({
+        batchSize: dashboardTileHydrationConcurrency,
+        collectVisibleModelIds: async () => {
+          await waitForDashboardRenderFrame(signal);
+          return collectVisibleDashboardModelIds({
+            clearTimeout: window.clearTimeout.bind(window),
+            createObserver: getDashboardIntersectionObserverFactory(),
+            documentTarget: document,
+            modelIds,
+            setTimeout: window.setTimeout.bind(window),
+            signal,
+            timeoutMs: dashboardViewportObservationTimeoutMs,
+          });
+        },
+        concurrency: dashboardTileHydrationConcurrency,
+        getModelId: dashboardTileModelId,
+        hydrate: async (tile) => {
+          const key = dashboardTileKey(tile);
+          const result = await hydrateDashboardTile(
+            tile,
+            run,
+            signal,
+            snapshotContext,
+          );
+
+          if (result === "ready") {
+            tileBackoff.recordSuccess(key);
+          } else if (result === "failed") {
+            tileBackoff.recordFailure(key);
+          }
+        },
+        hydrateBatch: async (batch) => {
+          const results = await hydrateDashboardTileBatch(
+            batch,
+            run,
+            signal,
+            snapshotContext,
+          );
+
+          for (const { result, tile } of results) {
+            const key = dashboardTileKey(tile);
+            if (result === "ready") {
+              tileBackoff.recordSuccess(key);
+            } else if (result === "failed") {
+              tileBackoff.recordFailure(key);
+            }
+          }
+        },
+        items: tiles,
+        onAllItemsSettled: () => performanceMarks.markAllTilesSettled(),
+        onVisibleItemsSettled: () => performanceMarks.markVisibleTilesReady(),
+        signal,
+        waitForIdle: () =>
+          waitForDashboardHydrationIdle({
+            ...getDashboardIdleCallbacks(),
+            clearTimeout: window.clearTimeout.bind(window),
+            setTimeout: window.setTimeout.bind(window),
+            signal,
+          }),
       });
     }
 
-    async function hydrateDashboardTile(
+    function subscribeDashboardTileEvents(
-      tile: DashboardTileReference,
       run: number,
       signal: AbortSignal,
+      snapshotContext: DashboardTileSnapshotStoreContext,
     ) {
-      const key = dashboardTileKey(tile);
+      unsubscribeTileEvents?.();
+      unsubscribeTileEvents = subscribeToDashboardTileEvents({
+        onTile: (event) => {
+          const tileResponse = event as DashboardTileResponse;
+          if (!isDashboardTileResponse(tileResponse)) return;
+          applyDashboardTileHydrationResponse(
+            tileResponse.tile,
+            tileResponse,
+            run,
+            signal,
+            snapshotContext,
+          );
+        },
+        onUnavailable: () => {
+          unsubscribeTileEvents = undefined;
+        },
+      });
+    }
 
+    async function hydrateDashboardTileBatch(
+      tiles: DashboardTileReference[],
+      run: number,
+      signal: AbortSignal,
+      snapshotContext: DashboardTileSnapshotStoreContext,
+    ): Promise<Array<{ result: DashboardTileHydrationResult; tile: DashboardTileReference }>> {
       try {
-        const response = await fetch(dashboardTileUrl(tile), { signal });
+        const response = await fetch("/api/dashboard/tiles", {
-        const tileResponse = (await response.json()) as DashboardTileResponse;
+          body: JSON.stringify({ tiles }),
+          headers: { "Content-Type": "application/json" },
+          method: "POST",
+          signal,
+        });
+        const batchResponse = (await response.json()) as DashboardTilesBatchResponse;
 
         if (
           cancelled ||
           signal.aborted ||
           run !== hydrationRun ||
           !response.ok ||
+          batchResponse.state !== "ready"
+        ) {
+          throw new Error("Dashboard tile batch hydration failed");
+        }
+
+        const responsesByKey = new Map(
+          batchResponse.tiles.map((tileResponse) => [
+            dashboardTileMatchKey(tileResponse.tile),
+            tileResponse,
+          ]),
+        );
+
+        return tiles.map((tile) => {
+          const key = dashboardTileKey(tile);
+          try {
+            return {
+              tile,
+              result: applyDashboardTileHydrationResponse(
+                tile,
+                responsesByKey.get(dashboardTileMatchKey(tile)),
+                run,
+                signal,
+                snapshotContext,
+              ),
+            };
+          } finally {
+            finishDashboardTileHydration(key, run, signal);
+          }
+        });
+      } catch (error) {
+        if (signal.aborted || cancelled || run !== hydrationRun) {
+          return tiles.map((tile) => ({ result: "aborted", tile }));
+        }
+
+        return Promise.all(
+          tiles.map(async (tile) => ({
+            tile,
+            result: await hydrateDashboardTile(tile, run, signal, snapshotContext),
+          })),
+        );
+      }
+    }
+
+    async function hydrateDashboardTile(
+      tile: DashboardTileReference,
+      run: number,
+      signal: AbortSignal,
+      snapshotContext: DashboardTileSnapshotStoreContext,
+    ): Promise<DashboardTileHydrationResult> {
+      const key = dashboardTileKey(tile);
+
+      try {
+        const response = await fetch(dashboardTileUrl(tile), { signal });
+        const tileResponse = (await response.json()) as DashboardTileResponse;
+
+        if (!response.ok) {
+          return "failed";
+        }
+
+        const result = applyDashboardTileHydrationResponse(
+          tile,
+          tileResponse,
+          run,
+          signal,
+          snapshotContext,
+        );
+        if (result !== "ready") {
+          return signal.aborted || cancelled || run !== hydrationRun
+            ? "aborted"
+            : "failed";
+        }
+
+        return "ready";
+      } catch (error) {
+        if (!isAbortError(error) && !cancelled) {
+          console.error("Dashboard tile hydration failed", error);
+          return "failed";
+        }
+        return "aborted";
+      } finally {
+        finishDashboardTileHydration(key, run, signal);
+      }
+    }
+
+    function applyDashboardTileHydrationResponse(
+      tile: DashboardTileReference,
+      tileResponse: DashboardTileResponse | undefined,
+      run: number,
+      signal: AbortSignal,
+      snapshotContext: DashboardTileSnapshotStoreContext,
+    ): DashboardTileHydrationResult {
+      if (
+        cancelled ||
+        signal.aborted ||
+        run !== hydrationRun ||
+        !tileResponse ||
+        dashboardTileMatchKey(tileResponse.tile) !== dashboardTileMatchKey(tile) ||
         tileResponse.state !== "ready"
       ) {
-          return;
+        return signal.aborted || cancelled || run !== hydrationRun
+          ? "aborted"
+          : "failed";
       }
 
       const readyTileResponse = tileResponse;
@@ -263,11 +528,19 @@ export default function App() {
             }
           : current,
       );
-      } catch (error) {
+      tileSnapshotStore.saveReadyTile({
-        if (!isAbortError(error) && !cancelled) {
+        ...snapshotContext,
-          console.error("Dashboard tile hydration failed", error);
+        response: readyTileResponse,
+      });
+      performanceMarks.markFirstTileReady();
+      return "ready";
     }
-      } finally {
+
+    function finishDashboardTileHydration(
+      key: string,
+      run: number,
+      signal: AbortSignal,
+    ) {
       if (!cancelled && !signal.aborted && run === hydrationRun) {
         setHydratingItemIds((current) => {
           const next = new Set(current);
@@ -276,7 +549,6 @@ export default function App() {
         });
       }
     }
-    }
 
     const detachRefreshLifecycle = attachDashboardRefreshLifecycle({
       documentTarget: document,
@@ -291,6 +563,7 @@ export default function App() {
     return () => {
       cancelled = true;
       clearRefreshTimer();
+      unsubscribeTileEvents?.();
       requestAborter.abortActiveRequests();
       detachRefreshLifecycle();
     };
@@ -326,6 +599,61 @@ function getThemeStorage(): Storage | undefined {
   }
 }
 
+function getTileSnapshotStorage(): Storage | undefined {
+  try {
+    return window.localStorage;
+  } catch {
+    return undefined;
+  }
+}
+
+type DashboardIdleWindow = Window & {
+  cancelIdleCallback?: (handle: number) => void;
+  requestIdleCallback?: (
+    callback: () => void,
+    options?: { timeout?: number },
+  ) => number;
+};
+
+function getDashboardIdleCallbacks() {
+  const idleWindow = window as DashboardIdleWindow;
+  return {
+    cancelIdleCallback: idleWindow.cancelIdleCallback?.bind(idleWindow),
+    requestIdleCallback: idleWindow.requestIdleCallback?.bind(idleWindow),
+  };
+}
+
+function getDashboardIntersectionObserverFactory():
+  | DashboardIntersectionObserverFactory
+  | undefined {
+  if (typeof window.IntersectionObserver === "undefined") return undefined;
+
+  return (callback) =>
+    new window.IntersectionObserver((entries) => {
+      callback(entries);
+    });
+}
+
+async function waitForDashboardRenderFrame(signal: AbortSignal): Promise<void> {
+  if (signal.aborted) return;
+
+  await new Promise<void>((resolve) => {
+    let settled = false;
+    let frame: number | undefined;
+
+    function finish() {
+      if (settled) return;
+      settled = true;
+      if (frame !== undefined) window.cancelAnimationFrame(frame);
+      signal.removeEventListener("abort", finish);
+      resolve();
+    }
+
+    signal.addEventListener("abort", finish, { once: true });
+    frame = window.requestAnimationFrame(finish);
+  });
+}
+
 function isAbortError(error: unknown): boolean {
   return (
     typeof error === "object" &&
@@ -335,6 +663,57 @@ function isAbortError(error: unknown): boolean {
   );
 }
 
+function isDashboardTileResponse(value: unknown): value is DashboardTileResponse {
+  return (
+    typeof value === "object" &&
+    value !== null &&
+    "state" in value &&
+    (value.state === "ready" ||
+      value.state === "not_found" ||
+      value.state === "disabled") &&
+    "tile" in value &&
+    typeof value.tile === "object" &&
+    value.tile !== null
+  );
+}
+
+export function restoreDashboardTileSnapshots(
+  dashboard: DashboardRuntimeState,
+  snapshots: RestoredDashboardTileSnapshot[],
+): {
+  dashboard: DashboardRuntimeState;
+  restoredItemIds: Set<string>;
+} {
+  if (dashboard.state !== "ready" || snapshots.length === 0) {
+    return {
+      dashboard,
+      restoredItemIds: new Set(),
+    };
+  }
+
+  return snapshots.filter((snapshot) =>
+    isPersistableDashboardTileSnapshot(snapshot.response.tile)
+  ).reduce(
+    (current, snapshot) => ({
+      dashboard: {
+        ...current.dashboard,
+        document: applyDashboardTile(
+          current.dashboard.document,
+          snapshot.response as Extract<DashboardTileResponse, { state: "ready" }>,
+        ),
+      },
+      restoredItemIds: new Set([
+        ...current.restoredItemIds,
+        dashboardTileKey(snapshot.response.tile),
+      ]),
+    }),
+    {
+      dashboard,
+      restoredItemIds: new Set<string>(),
+    },
+  );
+}
+
 function markHydratingItems(
   dashboard: UiDashboardPreview,
   hydratingItemIds?: ReadonlySet<string>,
@@ -415,7 +794,7 @@ export function dashboardHydrationTiles(
       })),
   );
 
-  return [...telemetry, ...services, ...modules, ...status];
+  return [...status, ...telemetry, ...modules, ...services];
 }
 
 function dashboardTileKey(tile: DashboardTileReference): string {
@@ -424,6 +803,15 @@ function dashboardTileKey(tile: DashboardTileReference): string {
   return `${tile.kind}:${tile.id}`;
 }
 
+export function dashboardTileMatchKey(tile: DashboardTileReference): string {
+  return JSON.stringify(tile);
+}
+
+function dashboardTileModelId(tile: DashboardTileReference): string {
+  if (tile.kind === "status") return `${tile.stripId}:${tile.id}`;
+  return tile.id;
+}
+
 function dashboardTileUrl(tile: DashboardTileReference): string {
   const parts = tile.kind === "status"
     ? ["api", "dashboard", "tile", tile.kind, tile.stripId, tile.id]

apps/web/src/lib/client/dashboard-refresh.test.ts

@@ -1,8 +1,21 @@
 import { describe, expect, test } from "vitest";
 import {
   attachDashboardRefreshLifecycle,
+  collectVisibleDashboardModelIds,
+  createDashboardPerformanceMarks,
+  createDashboardRefreshDelay,
+  dashboardPerformanceMarks,
+  createDashboardTileSnapshotStore,
+  createDashboardTileBackoff,
   createDashboardRequestAborter,
+  runDashboardHydrationQueue,
+  runDashboardHydrationBatchQueue,
+  runViewportAwareDashboardHydrationQueue,
   shouldPauseDashboardRefresh,
+  splitDashboardHydrationItemsByVisibility,
+  subscribeToDashboardTileEvents,
+  waitForDashboardHydrationIdle,
+  type DashboardIntersectionEntry,
 } from "./dashboard-refresh";
 
 describe("dashboard refresh lifecycle", () => {
@@ -73,4 +86,548 @@ describe("dashboard refresh lifecycle", () => {
 
     expect(calls).toEqual(["pause", "pause", "pause", "load", "load"]);
   });
+
+  test("limits tile hydration concurrency", async () => {
+    let active = 0;
+    let maxActive = 0;
+    const started: number[] = [];
+    const releases = new Map<number, () => void>();
+
+    const queue = runDashboardHydrationQueue({
+      concurrency: 2,
+      items: [1, 2, 3, 4],
+      signal: new AbortController().signal,
+      hydrate: async (item) => {
+        active += 1;
+        maxActive = Math.max(maxActive, active);
+        started.push(item);
+        await new Promise<void>((resolve) => releases.set(item, resolve));
+        active -= 1;
+      },
+    });
+
+    await waitFor(() => started.length === 2);
+    expect(started).toEqual([1, 2]);
+    expect(maxActive).toBe(2);
+
+    releases.get(1)?.();
+    await waitFor(() => started.length === 3);
+    expect(started).toEqual([1, 2, 3]);
+    expect(maxActive).toBe(2);
+
+    releases.get(2)?.();
+    releases.get(3)?.();
+    await waitFor(() => started.length === 4);
+    releases.get(4)?.();
+    await queue;
+    expect(maxActive).toBe(2);
+  });
+
+  test("hydrates queued work in fixed-size batches", async () => {
+    const batches: number[][] = [];
+
+    await runDashboardHydrationBatchQueue({
+      batchSize: 3,
+      hydrateBatch: (items) => {
+        batches.push(items);
+      },
+      items: [1, 2, 3, 4, 5, 6, 7],
+      signal: new AbortController().signal,
+    });
+
+    expect(batches).toEqual([
+      [1, 2, 3],
+      [4, 5, 6],
+      [7],
+    ]);
+  });
+
+  test("hydrates visible items before deferred items and waits for idle", async () => {
+    const calls: string[] = [];
+    const idleReleases: Array<() => void> = [];
+
+    const hydration = runViewportAwareDashboardHydrationQueue({
+      collectVisibleModelIds: async () => new Set(["b"]),
+      concurrency: 1,
+      getModelId: (item) => item,
+      hydrate: async (item) => {
+        calls.push(item);
+      },
+      items: ["a", "b", "c"],
+      onAllItemsSettled: () => calls.push("all-settled"),
+      onVisibleItemsSettled: () => calls.push("visible-settled"),
+      signal: new AbortController().signal,
+      waitForIdle: async () => {
+        calls.push("idle");
+        await new Promise<void>((resolve) => idleReleases.push(resolve));
+      },
+    });
+
+    await waitFor(() => calls.includes("idle"));
+    expect(calls).toEqual(["b", "visible-settled", "idle"]);
+
+    idleReleases.shift()?.();
+    await hydration;
+    expect(calls).toEqual([
+      "b",
+      "visible-settled",
+      "idle",
+      "a",
+      "c",
+      "all-settled",
+    ]);
+  });
+
+  test("splits visible hydration items while preserving document order", () => {
+    expect(splitDashboardHydrationItemsByVisibility({
+      getModelId: (item) => item.id,
+      items: [{ id: "status" }, { id: "telemetry" }, { id: "service" }],
+      visibleModelIds: new Set(["service", "status"]),
+    })).toEqual({
+      visible: [{ id: "status" }, { id: "service" }],
+      deferred: [{ id: "telemetry" }],
+    });
+  });
+
+  test("collects visible data-model-id elements with IntersectionObserver", async () => {
+    const elements = [
+      modelElement("status"),
+      modelElement("telemetry"),
+      modelElement("unrelated"),
+    ];
+    let callback:
+      | ((entries: DashboardIntersectionEntry[]) => void)
+      | undefined;
+    let finishObservation: (() => void) | undefined;
+    let disconnected = false;
+    const observed: string[] = [];
+
+    const visible = collectVisibleDashboardModelIds({
+      createObserver: (observerCallback) => {
+        callback = observerCallback;
+        return {
+          disconnect() {
+            disconnected = true;
+          },
+          observe(element) {
+            observed.push(element.getAttribute("data-model-id") || "");
+          },
+        };
+      },
+      documentTarget: {
+        querySelectorAll: () => elements,
+      },
+      modelIds: ["status", "telemetry"],
+      setTimeout: (handler) => {
+        finishObservation = handler;
+        return 1 as unknown as ReturnType<typeof setTimeout>;
+      },
+      clearTimeout: () => undefined,
+      signal: new AbortController().signal,
+    });
+
+    callback?.([
+      {
+        isIntersecting: false,
+        intersectionRatio: 0,
+        target: elements[0],
+      },
+      {
+        isIntersecting: true,
+        target: elements[1],
+      },
+    ]);
+    finishObservation?.();
+
+    expect(await visible).toEqual(new Set(["telemetry"]));
+    expect(observed).toEqual(["status", "telemetry"]);
+    expect(disconnected).toBe(true);
+  });
+
+  test("waits for requestIdleCallback when available", async () => {
+    let idleCallback: (() => void) | undefined;
+    let cancelledIdle: number | undefined;
+    const wait = waitForDashboardHydrationIdle({
+      cancelIdleCallback: (handle) => {
+        cancelledIdle = handle;
+      },
+      requestIdleCallback: (callback) => {
+        idleCallback = callback;
+        return 7;
+      },
+      signal: new AbortController().signal,
+    });
+
+    expect(cancelledIdle).toBeUndefined();
+    idleCallback?.();
+    await wait;
+    expect(cancelledIdle).toBe(7);
+  });
+
+  test("jitters refresh delays and slows repeated failures", () => {
+    const delay = createDashboardRefreshDelay({
+      jitterRatio: 0.1,
+      random: () => 1,
+    });
+
+    expect(delay.nextDelayMs(1_000)).toBe(1_100);
+    delay.recordFailure();
+    expect(delay.nextDelayMs(1_000)).toBe(2_200);
+    delay.recordFailure();
+    expect(delay.nextDelayMs(1_000)).toBe(4_400);
+    delay.recordSuccess();
+    expect(delay.nextDelayMs(1_000)).toBe(1_100);
+  });
+
+  test("marks dashboard performance milestones once per shell run", () => {
+    const marks: string[] = [];
+    const performanceMarks = createDashboardPerformanceMarks({
+      mark: (name) => marks.push(name),
+    });
+
+    performanceMarks.markShellLoad();
+    performanceMarks.markFirstTileReady();
+    performanceMarks.markFirstTileReady();
+    performanceMarks.markVisibleTilesReady();
+    performanceMarks.markAllTilesSettled();
+    performanceMarks.markShellLoad();
+    performanceMarks.markFirstTileReady();
+
+    expect(marks).toEqual([
+      dashboardPerformanceMarks.shellLoad,
+      dashboardPerformanceMarks.firstTileReady,
+      dashboardPerformanceMarks.visibleTilesReady,
+      dashboardPerformanceMarks.allTilesSettled,
+      dashboardPerformanceMarks.shellLoad,
+      dashboardPerformanceMarks.firstTileReady,
+    ]);
+  });
+
+  test("subscribes to dashboard tile events", () => {
+    const received: unknown[] = [];
+    let listener: ((event: MessageEvent<string>) => void) | undefined;
+    let closed = false;
+
+    const unsubscribe = subscribeToDashboardTileEvents({
+      createEventSource: (url) => {
+        expect(url).toBe("/api/dashboard/events");
+        return {
+          addEventListener(_type, eventListener) {
+            listener = eventListener;
+          },
+          close() {
+            closed = true;
+          },
+          onerror: null,
+        };
+      },
+      onTile: (tile) => received.push(tile),
+    });
+
+    listener?.({ data: JSON.stringify({ state: "ready" }) } as MessageEvent<string>);
+    expect(received).toEqual([{ state: "ready" }]);
+    unsubscribe();
+    expect(closed).toBe(true);
+  });
+
+  test("falls back when dashboard tile events are unavailable or fail", () => {
+    let unavailableCount = 0;
+
+    subscribeToDashboardTileEvents({
+      createEventSource: undefined,
+      onTile: () => undefined,
+      onUnavailable: () => {
+        unavailableCount += 1;
+      },
+    });
+
+    let errorHandler: (() => void) | null = null;
+    const unsubscribe = subscribeToDashboardTileEvents({
+      createEventSource: () => ({
+        addEventListener: () => undefined,
+        close: () => undefined,
+        get onerror() {
+          return errorHandler;
+        },
+        set onerror(handler) {
+          errorHandler = handler;
+        },
+      }),
+      onTile: () => undefined,
+      onUnavailable: () => {
+        unavailableCount += 1;
+      },
+    });
+
+    if (!errorHandler) throw new Error("expected error handler");
+    const triggerError = errorHandler as unknown as () => void;
+    triggerError();
+    unsubscribe();
+    expect(unavailableCount).toBe(2);
+  });
+
+  test("backs off failed tile keys and resets after success", () => {
+    const backoff = createDashboardTileBackoff();
+
+    backoff.recordFailure("telemetry:infra-ram", 1_000);
+    expect(backoff.canAttempt("telemetry:infra-ram", 15_999)).toBe(false);
+    expect(backoff.canAttempt("telemetry:infra-ram", 16_000)).toBe(true);
+
+    backoff.recordFailure("telemetry:infra-ram", 16_000);
+    expect(backoff.canAttempt("telemetry:infra-ram", 45_999)).toBe(false);
+    expect(backoff.canAttempt("telemetry:infra-ram", 46_000)).toBe(true);
+
+    backoff.recordFailure("telemetry:infra-ram", 46_000);
+    backoff.recordFailure("telemetry:infra-ram", 106_000);
+    expect(backoff.canAttempt("telemetry:infra-ram", 225_999)).toBe(false);
+    expect(backoff.canAttempt("telemetry:infra-ram", 226_000)).toBe(true);
+
+    backoff.recordSuccess("telemetry:infra-ram");
+    expect(backoff.canAttempt("telemetry:infra-ram", 107_000)).toBe(true);
+  });
+
+  test("stores only ready tile snapshots and restores them for matching revisions", () => {
+    const storage = createMemoryStorage();
+    let now = 1_000;
+    const store = createDashboardTileSnapshotStore(storage, {
+      now: () => now,
+    });
+
+    store.saveReadyTile({
+      currentRevisionId: "revision-a",
+      response: {
+        state: "ready",
+        tile: { kind: "telemetry", id: "infra-ram" },
+        item: {
+          id: "infra-ram",
+          label: "Infra RAM",
+          value: { kind: "percent", value: 42 },
+          detail: "live",
+          severity: "ok",
+        },
+      },
+      schemaVersion: "dashboard.v1",
+    });
+    store.saveTile({
+      currentRevisionId: "revision-a",
+      response: {
+        state: "not_found",
+        tile: { kind: "telemetry", id: "missing" },
+        message: "Missing",
+      },
+      schemaVersion: "dashboard.v1",
+    });
+    store.saveTile({
+      currentRevisionId: "revision-a",
+      response: {
+        state: "disabled",
+        tile: { kind: "module", id: "disabled-module" },
+        message: "Disabled",
+      },
+      schemaVersion: "dashboard.v1",
+    });
+
+    now = 61_000;
+
+    const restored = store.restore({
+      currentRevisionId: "revision-a",
+      schemaVersion: "dashboard.v1",
+    });
+
+    expect(restored).toEqual([
+      {
+        ageMs: 60_000,
+        response: {
+          state: "ready",
+          tile: { kind: "telemetry", id: "infra-ram" },
+          item: {
+            id: "infra-ram",
+            label: "Infra RAM",
+            value: { kind: "percent", value: 42 },
+            detail: "live - stale 60s",
+            severity: "stale",
+          },
+        },
+      },
+    ]);
+    expect(store.restore({
+      currentRevisionId: "revision-b",
+      schemaVersion: "dashboard.v1",
+    })).toEqual([]);
+  });
+
+  test("ignores malformed stored tile snapshots", () => {
+    const storage = createMemoryStorage();
+    storage.setItem(
+      "dimensionlab.dashboard.tiles.v1",
+      JSON.stringify({
+        currentRevisionId: "revision-a",
+        schemaVersion: "dashboard.v1",
+        tiles: [
+          {},
+          {
+            item: {
+              id: "incomplete",
+              detail: "live",
+            },
+            savedAt: 1_000,
+            tile: { kind: "telemetry", id: "incomplete" },
+          },
+          {
+            item: {
+              id: "infra-ram",
+              label: "Infra RAM",
+              value: { kind: "percent", value: 42 },
+              detail: "live",
+              severity: "ok",
+            },
+            savedAt: 1_000,
+            tile: { kind: "telemetry", id: "infra-ram" },
+          },
+          {
+            item: { id: "broken-detail", detail: 42 },
+            savedAt: 1_000,
+            tile: { kind: "telemetry", id: "broken-detail" },
+          },
+        ],
+        version: 1,
+      }),
+    );
+
+    const store = createDashboardTileSnapshotStore(storage, {
+      now: () => 16_000,
+    });
+
+    expect(store.restore({
+      currentRevisionId: "revision-a",
+      schemaVersion: "dashboard.v1",
+    })).toEqual([
+      {
+        ageMs: 15_000,
+        response: {
+          state: "ready",
+          tile: { kind: "telemetry", id: "infra-ram" },
+          item: {
+            id: "infra-ram",
+            label: "Infra RAM",
+            value: { kind: "percent", value: 42 },
+            detail: "live - stale 15s",
+            severity: "stale",
+          },
+        },
+      },
+    ]);
+  });
+
+  test("ignores restored aggregate health snapshots", () => {
+    const storage = createMemoryStorage();
+    storage.setItem(
+      "dimensionlab.dashboard.tiles.v1",
+      JSON.stringify({
+        currentRevisionId: "revision-a",
+        schemaVersion: "dashboard.v1",
+        tiles: [
+          {
+            item: {
+              id: "system-status",
+              label: "System Status",
+              value: "20 services down",
+              severity: "danger",
+            },
+            savedAt: 1_000,
+            tile: { kind: "status", stripId: "footer-status", id: "system-status" },
+          },
+          {
+            item: {
+              id: "runtime-health-summary",
+              kind: "summary",
+              title: "Runtime Health",
+              value: "20 services down",
+              detail: "0 warnings - 8 services ok",
+              severity: "danger",
+            },
+            savedAt: 1_000,
+            tile: { kind: "module", id: "runtime-health-summary" },
+          },
+          {
+            item: {
+              id: "infra-ram",
+              label: "Infra RAM",
+              value: { kind: "percent", value: 42 },
+              detail: "live",
+              severity: "ok",
+            },
+            savedAt: 1_000,
+            tile: { kind: "telemetry", id: "infra-ram" },
+          },
+        ],
+        version: 1,
+      }),
+    );
+
+    const store = createDashboardTileSnapshotStore(storage, {
+      now: () => 16_000,
+    });
+
+    expect(store.restore({
+      currentRevisionId: "revision-a",
+      schemaVersion: "dashboard.v1",
+    })).toEqual([
+      {
+        ageMs: 15_000,
+        response: {
+          state: "ready",
+          tile: { kind: "telemetry", id: "infra-ram" },
+          item: {
+            id: "infra-ram",
+            label: "Infra RAM",
+            value: { kind: "percent", value: 42 },
+            detail: "live - stale 15s",
+            severity: "stale",
+          },
+        },
+      },
+    ]);
+  });
 });
+
+async function waitFor(predicate: () => boolean) {
+  for (let attempt = 0; attempt < 20; attempt += 1) {
+    if (predicate()) return;
+    await Promise.resolve();
+  }
+
+  throw new Error("condition was not met");
+}
+
+function createMemoryStorage(): Storage {
+  const values = new Map<string, string>();
+  return {
+    get length() {
+      return values.size;
+    },
+    clear() {
+      values.clear();
+    },
+    getItem(key) {
+      return values.get(key) ?? null;
+    },
+    key(index) {
+      return [...values.keys()][index] ?? null;
+    },
+    removeItem(key) {
+      values.delete(key);
+    },
+    setItem(key, value) {
+      values.set(key, value);
+    },
+  };
+}
+
+function modelElement(modelId: string) {
+  return {
+    getAttribute(name: string) {
+      return name === "data-model-id" ? modelId : null;
+    },
+  };
+}

apps/web/src/lib/client/dashboard-refresh.ts

@@ -3,6 +3,39 @@ export interface DashboardRefreshPauseState {
   online: boolean;
 }
 
+export type DashboardTileReference =
+  | { kind: "telemetry"; id: string }
+  | { kind: "service"; groupId: string; id: string }
+  | { kind: "module"; id: string }
+  | { kind: "status"; stripId: string; id: string };
+
+type SnapshotMetricValue = Record<string, unknown>;
+
+export type DashboardTileSnapshotItem = {
+  detail?: string;
+  id: string;
+  label?: string;
+  severity?: string;
+  value?: SnapshotMetricValue | string;
+} & Record<string, unknown>;
+
+export type DashboardTileSnapshotResponse =
+  | {
+      state: "ready";
+      tile: DashboardTileReference;
+      item: DashboardTileSnapshotItem;
+    }
+  | {
+      state: "not_found";
+      tile: DashboardTileReference;
+      message: string;
+    }
+  | {
+      state: "disabled";
+      tile: DashboardTileReference;
+      message: string;
+    };
+
 export function shouldPauseDashboardRefresh(
   state: DashboardRefreshPauseState,
 ): boolean {
@@ -88,3 +121,766 @@ export function attachDashboardRefreshLifecycle(
     options.windowTarget.removeEventListener("pagehide", handlePageHide);
   };
 }
+
+export interface DashboardHydrationQueueOptions<TItem> {
+  concurrency: number;
+  hydrate: (item: TItem) => Promise<void> | void;
+  items: TItem[];
+  signal: AbortSignal;
+}
+
+export async function runDashboardHydrationQueue<TItem>(
+  options: DashboardHydrationQueueOptions<TItem>,
+): Promise<void> {
+  const concurrency = Math.max(1, Math.floor(options.concurrency));
+  let nextIndex = 0;
+
+  async function worker() {
+    while (!options.signal.aborted) {
+      const item = options.items[nextIndex];
+      nextIndex += 1;
+      if (item === undefined) return;
+
+      await options.hydrate(item);
+    }
+  }
+
+  const workerCount = Math.min(concurrency, options.items.length);
+  await Promise.all(
+    Array.from({ length: workerCount }, () => worker()),
+  );
+}
+
+export interface DashboardViewportHydrationQueueOptions<TItem> {
+  batchSize?: number;
+  collectVisibleModelIds: () => Promise<ReadonlySet<string>>;
+  concurrency: number;
+  getModelId: (item: TItem) => string;
+  hydrate: (item: TItem) => Promise<void> | void;
+  hydrateBatch?: (items: TItem[]) => Promise<void> | void;
+  items: TItem[];
+  onAllItemsSettled?: () => void;
+  onVisibleItemsSettled?: () => void;
+  signal: AbortSignal;
+  waitForIdle: () => Promise<void>;
+}
+
+export async function runViewportAwareDashboardHydrationQueue<TItem>(
+  options: DashboardViewportHydrationQueueOptions<TItem>,
+): Promise<void> {
+  const visibleModelIds = await options.collectVisibleModelIds();
+  if (options.signal.aborted) return;
+
+  const { visible, deferred } = splitDashboardHydrationItemsByVisibility({
+    getModelId: options.getModelId,
+    items: options.items,
+    visibleModelIds,
+  });
+
+  await runDashboardHydrationItems(options, visible);
+  if (options.signal.aborted) return;
+  options.onVisibleItemsSettled?.();
+
+  if (!deferred.length) {
+    options.onAllItemsSettled?.();
+    return;
+  }
+
+  await options.waitForIdle();
+  if (options.signal.aborted) return;
+
+  await runDashboardHydrationItems(options, deferred);
+  options.onAllItemsSettled?.();
+}
+
+async function runDashboardHydrationItems<TItem>(
+  options: DashboardViewportHydrationQueueOptions<TItem>,
+  items: TItem[],
+): Promise<void> {
+  if (!options.hydrateBatch) {
+    await runDashboardHydrationQueue({
+      concurrency: options.concurrency,
+      hydrate: options.hydrate,
+      items,
+      signal: options.signal,
+    });
+    return;
+  }
+
+  await runDashboardHydrationBatchQueue({
+    batchSize: options.batchSize ?? options.concurrency,
+    hydrateBatch: options.hydrateBatch,
+    items,
+    signal: options.signal,
+  });
+}
+
+export async function runDashboardHydrationBatchQueue<TItem>(options: {
+  batchSize: number;
+  hydrateBatch: (items: TItem[]) => Promise<void> | void;
+  items: TItem[];
+  signal: AbortSignal;
+}): Promise<void> {
+  const batchSize = Math.max(1, Math.floor(options.batchSize));
+
+  for (let index = 0; index < options.items.length; index += batchSize) {
+    if (options.signal.aborted) return;
+    await options.hydrateBatch(options.items.slice(index, index + batchSize));
+  }
+}
+
+export function splitDashboardHydrationItemsByVisibility<TItem>(options: {
+  getModelId: (item: TItem) => string;
+  items: TItem[];
+  visibleModelIds: ReadonlySet<string>;
+}): {
+  deferred: TItem[];
+  visible: TItem[];
+} {
+  const visible: TItem[] = [];
+  const deferred: TItem[] = [];
+
+  for (const item of options.items) {
+    if (options.visibleModelIds.has(options.getModelId(item))) {
+      visible.push(item);
+    } else {
+      deferred.push(item);
+    }
+  }
+
+  return { visible, deferred };
+}
+
+export interface DashboardModelElement {
+  getAttribute(name: string): string | null;
+}
+
+export interface DashboardViewportElementSource {
+  querySelectorAll(selector: string): ArrayLike<DashboardModelElement>;
+}
+
+export interface DashboardIntersectionEntry {
+  intersectionRatio?: number;
+  isIntersecting: boolean;
+  target: DashboardModelElement;
+}
+
+export interface DashboardIntersectionObserver {
+  disconnect(): void;
+  observe(element: DashboardModelElement): void;
+}
+
+export type DashboardIntersectionObserverFactory = (
+  callback: (entries: DashboardIntersectionEntry[]) => void,
+) => DashboardIntersectionObserver;
+
+type DashboardTimerHandle = ReturnType<typeof setTimeout>;
+
+export interface DashboardVisibleModelIdCollectorOptions {
+  clearTimeout?: (handle: DashboardTimerHandle) => void;
+  createObserver?: DashboardIntersectionObserverFactory;
+  documentTarget: DashboardViewportElementSource;
+  modelIds: Iterable<string>;
+  setTimeout?: (callback: () => void, timeoutMs: number) => DashboardTimerHandle;
+  signal: AbortSignal;
+  timeoutMs?: number;
+}
+
+export async function collectVisibleDashboardModelIds(
+  options: DashboardVisibleModelIdCollectorOptions,
+): Promise<Set<string>> {
+  const targetIds = new Set(options.modelIds);
+  if (!targetIds.size || options.signal.aborted) return new Set();
+
+  const elements = Array.from(
+    options.documentTarget.querySelectorAll("[data-model-id]"),
+  ).filter((element) => {
+    const modelId = element.getAttribute("data-model-id");
+    return modelId ? targetIds.has(modelId) : false;
+  });
+
+  if (!elements.length) return new Set();
+  if (!options.createObserver) return targetIds;
+  const createObserver = options.createObserver;
+
+  const setTimer =
+    options.setTimeout ||
+    ((callback: () => void, timeoutMs: number) =>
+      globalThis.setTimeout(callback, timeoutMs));
+  const clearTimer =
+    options.clearTimeout ||
+    ((handle: DashboardTimerHandle) => globalThis.clearTimeout(handle));
+  const timeoutMs = options.timeoutMs ?? 80;
+
+  return new Promise((resolve) => {
+    const visibleModelIds = new Set<string>();
+    let settled = false;
+    let timeoutHandle: DashboardTimerHandle | undefined;
+    let observer: DashboardIntersectionObserver | undefined;
+
+    function cleanup() {
+      if (timeoutHandle !== undefined) {
+        clearTimer(timeoutHandle);
+      }
+      observer?.disconnect();
+      options.signal.removeEventListener("abort", finish);
+    }
+
+    function finish() {
+      if (settled) return;
+      settled = true;
+      cleanup();
+      resolve(visibleModelIds);
+    }
+
+    observer = createObserver((entries) => {
+      for (const entry of entries) {
+        const modelId = entry.target.getAttribute("data-model-id");
+        if (
+          modelId &&
+          targetIds.has(modelId) &&
+          (entry.isIntersecting || (entry.intersectionRatio ?? 0) > 0)
+        ) {
+          visibleModelIds.add(modelId);
+        }
+      }
+    });
+
+    for (const element of elements) {
+      observer.observe(element);
+    }
+
+    options.signal.addEventListener("abort", finish, { once: true });
+    timeoutHandle = setTimer(finish, timeoutMs);
+  });
+}
+
+export interface DashboardHydrationIdleOptions {
+  cancelIdleCallback?: (handle: number) => void;
+  clearTimeout?: (handle: DashboardTimerHandle) => void;
+  requestIdleCallback?: (
+    callback: () => void,
+    options?: { timeout?: number },
+  ) => number;
+  setTimeout?: (callback: () => void, timeoutMs: number) => DashboardTimerHandle;
+  signal: AbortSignal;
+  timeoutMs?: number;
+}
+
+export async function waitForDashboardHydrationIdle(
+  options: DashboardHydrationIdleOptions,
+): Promise<void> {
+  if (options.signal.aborted) return;
+
+  const setTimer =
+    options.setTimeout ||
+    ((callback: () => void, timeoutMs: number) =>
+      globalThis.setTimeout(callback, timeoutMs));
+  const clearTimer =
+    options.clearTimeout ||
+    ((handle: DashboardTimerHandle) => globalThis.clearTimeout(handle));
+
+  await new Promise<void>((resolve) => {
+    let settled = false;
+    let idleHandle: number | undefined;
+    let timeoutHandle: DashboardTimerHandle | undefined;
+
+    function cleanup() {
+      if (idleHandle !== undefined) {
+        options.cancelIdleCallback?.(idleHandle);
+      }
+      if (timeoutHandle !== undefined) {
+        clearTimer(timeoutHandle);
+      }
+      options.signal.removeEventListener("abort", finish);
+    }
+
+    function finish() {
+      if (settled) return;
+      settled = true;
+      cleanup();
+      resolve();
+    }
+
+    options.signal.addEventListener("abort", finish, { once: true });
+    if (options.requestIdleCallback) {
+      idleHandle = options.requestIdleCallback(finish, {
+        timeout: options.timeoutMs ?? 1_000,
+      });
+    } else {
+      timeoutHandle = setTimer(finish, 0);
+    }
+  });
+}
+
+export interface DashboardRefreshDelayOptions {
+  failureMultiplierLimit?: number;
+  jitterRatio?: number;
+  random?: () => number;
+}
+
+export function createDashboardRefreshDelay(
+  options: DashboardRefreshDelayOptions = {},
+) {
+  const random = options.random || Math.random;
+  const jitterRatio = options.jitterRatio ?? 0.1;
+  const failureMultiplierLimit = options.failureMultiplierLimit ?? 8;
+  let consecutiveFailures = 0;
+
+  return {
+    nextDelayMs(baseDelayMs: number): number {
+      const failureMultiplier = consecutiveFailures
+        ? Math.min(2 ** consecutiveFailures, failureMultiplierLimit)
+        : 1;
+      const jitterFactor = 1 + ((random() * 2) - 1) * jitterRatio;
+      return Math.max(0, Math.round(baseDelayMs * failureMultiplier * jitterFactor));
+    },
+    recordFailure(): void {
+      consecutiveFailures += 1;
+    },
+    recordSuccess(): void {
+      consecutiveFailures = 0;
+    },
+  };
+}
+
+export interface DashboardPerformanceMarkOptions {
+  mark?: (name: string) => void;
+}
+
+export const dashboardPerformanceMarks = {
+  allTilesSettled: "dashboard:all-tiles-settled",
+  firstTileReady: "dashboard:first-tile-ready",
+  shellLoad: "dashboard:shell-load",
+  visibleTilesReady: "dashboard:visible-tiles-ready",
+} as const;
+
+export function createDashboardPerformanceMarks(
+  options: DashboardPerformanceMarkOptions = {},
+) {
+  const mark = options.mark ||
+    globalThis.performance?.mark?.bind(globalThis.performance);
+  let firstTileReadyMarked = false;
+
+  function safeMark(name: string) {
+    try {
+      mark?.(name);
+    } catch {
+      // Performance marks are diagnostics only.
+    }
+  }
+
+  return {
+    markAllTilesSettled(): void {
+      safeMark(dashboardPerformanceMarks.allTilesSettled);
+    },
+    markFirstTileReady(): void {
+      if (firstTileReadyMarked) return;
+      firstTileReadyMarked = true;
+      safeMark(dashboardPerformanceMarks.firstTileReady);
+    },
+    markShellLoad(): void {
+      firstTileReadyMarked = false;
+      safeMark(dashboardPerformanceMarks.shellLoad);
+    },
+    markVisibleTilesReady(): void {
+      safeMark(dashboardPerformanceMarks.visibleTilesReady);
+    },
+  };
+}
+
+export interface DashboardTileEventSource {
+  addEventListener(
+    type: "dashboard-tile",
+    listener: (event: MessageEvent<string>) => void,
+  ): void;
+  close(): void;
+  onerror: (() => void) | null;
+}
+
+export interface DashboardTileEventSubscriptionOptions {
+  createEventSource?: (url: string) => DashboardTileEventSource;
+  onTile: (data: unknown) => void;
+  onUnavailable?: () => void;
+  url?: string;
+}
+
+export function subscribeToDashboardTileEvents(
+  options: DashboardTileEventSubscriptionOptions,
+): () => void {
+  const createEventSource = options.createEventSource ||
+    (typeof globalThis.EventSource !== "undefined"
+      ? (url: string) => new globalThis.EventSource(url)
+      : undefined);
+
+  if (!createEventSource) {
+    options.onUnavailable?.();
+    return () => undefined;
+  }
+
+  const source = createEventSource(options.url || "/api/dashboard/events");
+  source.addEventListener("dashboard-tile", (event) => {
+    try {
+      options.onTile(JSON.parse(event.data));
+    } catch {
+      // Ignore malformed diagnostics from an optional live transport.
+    }
+  });
+  source.onerror = () => {
+    source.close();
+    options.onUnavailable?.();
+  };
+
+  return () => source.close();
+}
+
+const dashboardTileBackoffDelaysMs = [15_000, 30_000, 60_000, 120_000];
+
+export function createDashboardTileBackoff() {
+  const failures = new Map<string, { attempts: number; nextAttemptAt: number }>();
+
+  return {
+    canAttempt(key: string, now = Date.now()): boolean {
+      const failure = failures.get(key);
+      return !failure || now >= failure.nextAttemptAt;
+    },
+    recordFailure(key: string, now = Date.now()): void {
+      const previousAttempts = failures.get(key)?.attempts || 0;
+      const attempts = previousAttempts + 1;
+      const delay =
+        dashboardTileBackoffDelaysMs[
+          Math.min(attempts - 1, dashboardTileBackoffDelaysMs.length - 1)
+        ];
+      failures.set(key, {
+        attempts,
+        nextAttemptAt: now + delay,
+      });
+    },
+    recordSuccess(key: string): void {
+      failures.delete(key);
+    },
+  };
+}
+
+interface DashboardTileSnapshotRecord {
+  item: DashboardTileSnapshotItem;
+  savedAt: number;
+  tile: DashboardTileReference;
+}
+
+interface DashboardTileSnapshotPayload {
+  currentRevisionId: string;
+  schemaVersion: string;
+  tiles: DashboardTileSnapshotRecord[];
+  version: 1;
+}
+
+export interface DashboardTileSnapshotStoreContext {
+  currentRevisionId: string;
+  schemaVersion: string;
+}
+
+export interface DashboardTileSnapshotStoreOptions {
+  now?: () => number;
+}
+
+export interface RestoredDashboardTileSnapshot {
+  ageMs: number;
+  response: Extract<DashboardTileSnapshotResponse, { state: "ready" }>;
+}
+
+const dashboardTileSnapshotStorageKey = "dimensionlab.dashboard.tiles.v1";
+
+export function createDashboardTileSnapshotStore(
+  storage: Storage | undefined,
+  options: DashboardTileSnapshotStoreOptions = {},
+) {
+  const now = options.now || Date.now;
+
+  function read(): DashboardTileSnapshotPayload | null {
+    if (!storage) return null;
+
+    try {
+      const serialized = storage.getItem(dashboardTileSnapshotStorageKey);
+      if (!serialized) return null;
+
+      const payload = JSON.parse(serialized) as Partial<DashboardTileSnapshotPayload>;
+      if (
+        payload.version !== 1 ||
+        typeof payload.currentRevisionId !== "string" ||
+        typeof payload.schemaVersion !== "string" ||
+        !Array.isArray(payload.tiles)
+      ) {
+        return null;
+      }
+
+      return {
+        currentRevisionId: payload.currentRevisionId,
+        schemaVersion: payload.schemaVersion,
+        tiles: payload.tiles
+          .filter(isDashboardTileSnapshotRecord)
+          .filter((record) => isPersistableDashboardTileSnapshot(record.tile)),
+        version: 1,
+      };
+    } catch {
+      return null;
+    }
+  }
+
+  function write(payload: DashboardTileSnapshotPayload): void {
+    if (!storage) return;
+
+    try {
+      storage.setItem(dashboardTileSnapshotStorageKey, JSON.stringify(payload));
+    } catch {
+      // Best-effort warm-start cache; quota and privacy failures are non-fatal.
+    }
+  }
+
+  function matchingPayload(
+    context: DashboardTileSnapshotStoreContext,
+  ): DashboardTileSnapshotPayload {
+    const payload = read();
+    if (
+      payload &&
+      payload.currentRevisionId === context.currentRevisionId &&
+      payload.schemaVersion === context.schemaVersion
+    ) {
+      return payload;
+    }
+
+    return {
+      currentRevisionId: context.currentRevisionId,
+      schemaVersion: context.schemaVersion,
+      tiles: [],
+      version: 1,
+    };
+  }
+
+  function saveReadyTile(
+    input: DashboardTileSnapshotStoreContext & {
+      response: Extract<DashboardTileSnapshotResponse, { state: "ready" }>;
+    },
+  ): void {
+    if (!isPersistableDashboardTileSnapshot(input.response.tile)) return;
+
+    const payload = matchingPayload(input);
+    const key = dashboardTileSnapshotKey(input.response.tile);
+    const nextRecord: DashboardTileSnapshotRecord = {
+      item: input.response.item,
+      savedAt: now(),
+      tile: input.response.tile,
+    };
+    payload.tiles = [
+      nextRecord,
+      ...payload.tiles.filter((record) =>
+        dashboardTileSnapshotKey(record.tile) !== key
+      ),
+    ];
+    write(payload);
+  }
+
+  return {
+    restore(context: DashboardTileSnapshotStoreContext): RestoredDashboardTileSnapshot[] {
+      const payload = read();
+      if (
+        !payload ||
+        payload.currentRevisionId !== context.currentRevisionId ||
+        payload.schemaVersion !== context.schemaVersion
+      ) {
+        return [];
+      }
+
+      const restoredAt = now();
+      return payload.tiles.map((record) => ({
+        ageMs: Math.max(0, restoredAt - record.savedAt),
+        response: {
+          state: "ready",
+          tile: record.tile,
+          item: {
+            ...record.item,
+            detail: staleDashboardTileDetail(record.item.detail, restoredAt - record.savedAt),
+            severity: "stale",
+          },
+        },
+      }));
+    },
+    saveReadyTile,
+    saveTile(input: DashboardTileSnapshotStoreContext & {
+      response: DashboardTileSnapshotResponse;
+    }): void {
+      if (input.response.state === "ready") {
+        saveReadyTile({
+          currentRevisionId: input.currentRevisionId,
+          response: input.response,
+          schemaVersion: input.schemaVersion,
+        });
+      }
+    },
+  };
+}
+
+function dashboardTileSnapshotKey(tile: DashboardTileReference): string {
+  return JSON.stringify(tile);
+}
+
+export function isPersistableDashboardTileSnapshot(
+  tile: DashboardTileReference,
+): boolean {
+  if (tile.kind === "status" && tile.id === "system-status") return false;
+  if (tile.kind === "module" && tile.id === "runtime-health-summary") return false;
+  return true;
+}
+
+function isDashboardTileSnapshotRecord(
+  value: unknown,
+): value is DashboardTileSnapshotRecord {
+  if (!isSnapshotRecord(value)) return false;
+
+  return (
+    typeof value.savedAt === "number" &&
+    Number.isFinite(value.savedAt) &&
+    isDashboardTileReference(value.tile) &&
+    isDashboardTileSnapshotItem(value.tile, value.item)
+  );
+}
+
+function isDashboardTileReference(
+  value: unknown,
+): value is DashboardTileReference {
+  if (!isSnapshotRecord(value) || typeof value.kind !== "string") return false;
+
+  switch (value.kind) {
+    case "telemetry":
+    case "module":
+      return typeof value.id === "string";
+    case "service":
+      return typeof value.groupId === "string" && typeof value.id === "string";
+    case "status":
+      return typeof value.stripId === "string" && typeof value.id === "string";
+    default:
+      return false;
+  }
+}
+
+function isDashboardTileSnapshotItem(
+  tile: DashboardTileReference,
+  value: unknown,
+): value is DashboardTileSnapshotItem {
+  if (!isSnapshotRecord(value) || value.id !== tile.id) return false;
+
+  switch (tile.kind) {
+    case "telemetry":
+      return (
+        typeof value.label === "string" &&
+        isMetricValue(value.value) &&
+        isSeverity(value.severity) &&
+        isOptionalString(value.detail) &&
+        isOptionalString(value.description) &&
+        isOptionalString(value.icon) &&
+        isOptionalNumberArray(value.sparkline)
+      );
+    case "service":
+      return (
+        typeof value.label === "string" &&
+        typeof value.description === "string" &&
+        isSeverity(value.severity) &&
+        isOptionalString(value.detail) &&
+        isOptionalString(value.icon) &&
+        isOptionalLink(value.link)
+      );
+    case "module":
+      return (
+        (value.kind === "summary" ||
+          value.kind === "weather" ||
+          value.kind === "custom") &&
+        isOptionalString(value.title) &&
+        isOptionalString(value.label) &&
+        isOptionalString(value.value) &&
+        isOptionalString(value.detail) &&
+        isOptionalString(value.icon) &&
+        (value.severity === undefined || isSeverity(value.severity))
+      );
+    case "status":
+      return (
+        typeof value.label === "string" &&
+        typeof value.value === "string" &&
+        isOptionalLink(value.link) &&
+        (value.severity === undefined || isSeverity(value.severity))
+      );
+  }
+}
+
+function isSnapshotRecord(value: unknown): value is Record<string, unknown> {
+  return typeof value === "object" && value !== null;
+}
+
+function isMetricValue(value: unknown): boolean {
+  if (!isSnapshotRecord(value) || typeof value.kind !== "string") return false;
+
+  if (value.kind === "text") {
+    return (
+      typeof value.value === "string" &&
+      isOptionalString(value.unit)
+    );
+  }
+
+  const numericKinds = ["bytes", "latency", "number", "percent", "temperature"];
+  if (!numericKinds.includes(value.kind)) return false;
+  if (typeof value.value !== "number" || !Number.isFinite(value.value)) return false;
+  if (value.kind === "percent" && (value.value < 0 || value.value > 100)) {
+    return false;
+  }
+
+  const precision = value.precision;
+  return (
+    isOptionalString(value.unit) &&
+    (precision === undefined ||
+      (typeof precision === "number" &&
+        Number.isInteger(precision) &&
+        precision >= 0 &&
+        precision <= 4))
+  );
+}
+
+function isSeverity(value: unknown): boolean {
+  return (
+    value === "neutral" ||
+    value === "ok" ||
+    value === "warning" ||
+    value === "danger" ||
+    value === "stale" ||
+    value === "unavailable"
+  );
+}
+
+function isOptionalString(value: unknown): boolean {
+  return value === undefined || typeof value === "string";
+}
+
+function isOptionalNumberArray(value: unknown): boolean {
+  return (
+    value === undefined ||
+    (Array.isArray(value) &&
+      value.every((item) => typeof item === "number" && Number.isFinite(item)))
+  );
+}
+
+function isOptionalLink(value: unknown): boolean {
+  return (
+    value === undefined ||
+    (isSnapshotRecord(value) &&
+      typeof value.href === "string" &&
+      isOptionalString(value.label) &&
+      (value.external === undefined || typeof value.external === "boolean"))
+  );
+}
+
+function staleDashboardTileDetail(
+  detail: string | undefined,
+  ageMs: number,
+): string | undefined {
+  if (!detail) return detail;
+  const ageSeconds = Math.max(0, Math.floor(ageMs / 1_000));
+  return ageSeconds > 0 ? `${detail} - stale ${ageSeconds}s` : detail;
+}

apps/web/src/lib/dashboard-seed/dimensionlab.test.ts

@@ -128,7 +128,6 @@ const verifiedSeedIconIds = new Set([
   "mdi:pulse",
   "mdi:robot-outline",
   "mdi:router-network",
-  "mdi:text-box-search",
   "mdi:thermometer",
   "mdi:web",
   "mdi:weather-sunny",

apps/web/src/lib/dashboard-seed/dimensionlab.ts

@@ -431,14 +431,6 @@ export const dimensionLabDashboardFixture: DashboardDocument = {
         href: "https://models.dimensionlab.net",
         datasource: uptimeMonitor(7),
       }),
-      service({
-        id: "prompt-registry",
-        label: "Prompt Registry",
-        description: "Shared prompts, traces, evals",
-        icon: "mdi:text-box-search",
-        href: "https://prompts.dimensionlab.net",
-        datasource: uptimeMonitor(20),
-      }),
     ]),
     group("systems", "Systems", [
       service({

apps/web/src/lib/server/dashboard.test.ts

@@ -66,9 +66,6 @@ describe("dashboard runtime loader", () => {
     expect(runtime.document.statusStrips[0]?.items.map((item) => item.id)).toContain(
       "auto-refresh",
     );
-    expect(runtime.document.serviceGroups.flatMap((group) => group.services).map((service) => service.id)).toContain(
-      "prompt-registry",
-    );
     expect(store.listRevisions()).toHaveLength(2);
     expect(store.getActiveDashboard()?.revision.actor).toBe("initial-seed");
   });
@@ -138,13 +135,5 @@ function olderDimensionLabSeed() {
     ...strip,
     items: strip.items.filter((item) => item.id !== "auto-refresh"),
   }));
-  document.serviceGroups = document.serviceGroups.map((group) =>
-    group.id === "ai-automation"
-      ? {
-          ...group,
-          services: group.services.filter((service) => service.id !== "prompt-registry"),
-        }
-      : group,
-  );
   return document;
 }

apps/web/src/lib/server/datasources/dashboard-datasources.test.ts

@@ -3,7 +3,7 @@ import {
   DASHBOARD_SCHEMA_VERSION,
   type DashboardDocument,
 } from "@dimensionlab/dashboard-model";
-import { resolveDashboardDatasources } from ".";
+import { resolveDashboardDatasources, resolveDashboardTile } from ".";
 
 describe("dashboard datasource resolution", () => {
   test("hydrates telemetry, service health, weather, and summary data from live adapters", async () => {
@@ -124,6 +124,100 @@ describe("dashboard datasource resolution", () => {
     expect(resolved).not.toBe(testDocument);
     expect(testDocument.telemetry[0].value.value).toBe(1);
   });
+
+  test("shares service health snapshots across aggregate tile hydration", async () => {
+    let resolveFetch: ((response: Response) => void) | undefined;
+    const fetch = vi.fn((input: RequestInfo | URL) => {
+      const url = String(input);
+      if (url !== "https://service.example/health") {
+        throw new Error(`Unhandled test request: ${url}`);
+      }
+
+      return new Promise<Response>((resolve) => {
+        resolveFetch = resolve;
+      });
+    });
+    const document = testDashboard();
+
+    const moduleTile = resolveDashboardTile(
+      document,
+      { kind: "module", id: "runtime-health-summary" },
+      { fetch },
+    );
+    const statusTile = resolveDashboardTile(
+      document,
+      { kind: "status", stripId: "footer", id: "system-status" },
+      { fetch },
+    );
+
+    await Promise.resolve();
+    expect(fetch).toHaveBeenCalledTimes(1);
+
+    resolveFetch?.(jsonResponse({ status: "UP", ping: 42 }));
+    expect(await moduleTile).toMatchObject({
+      state: "ready",
+      item: {
+        id: "runtime-health-summary",
+        severity: "ok",
+        value: "all systems operational",
+      },
+    });
+    expect(await statusTile).toMatchObject({
+      state: "ready",
+      item: {
+        id: "system-status",
+        severity: "ok",
+        value: "All systems operational",
+      },
+    });
+  });
+
+  test("isolates service health snapshots by datasource fetch context", async () => {
+    const firstFetch = vi.fn(async () =>
+      jsonResponse({
+        status: "UP",
+        ping: 42,
+      })
+    );
+    const secondFetch = vi.fn(async () =>
+      jsonResponse({
+        status: "DOWN",
+        ping: 0,
+      })
+    );
+    const document = testDashboard();
+    document.serviceGroups[0].services[0] = {
+      ...document.serviceGroups[0].services[0],
+      id: "api-isolated",
+      datasource: {
+        type: "external",
+        adapter: "http-status",
+        reference: "GET https://service.example/isolated",
+      },
+    };
+
+    await resolveDashboardTile(
+      document,
+      { kind: "status", stripId: "footer", id: "system-status" },
+      { fetch: firstFetch },
+    );
+    const second = await resolveDashboardTile(
+      document,
+      { kind: "status", stripId: "footer", id: "system-status" },
+      { fetch: secondFetch },
+    );
+
+    expect(firstFetch).toHaveBeenCalledTimes(1);
+    expect(secondFetch).toHaveBeenCalledTimes(1);
+    expect(second).toMatchObject({
+      state: "ready",
+      item: {
+        id: "system-status",
+        severity: "danger",
+        value: "1 service down",
+      },
+    });
+  });
 });
 
 const testDocument = testDashboard();

apps/web/src/lib/server/datasources/index.ts

@@ -41,6 +41,7 @@ export type DashboardTileResolution =
 
 export interface DatasourceResolutionOptions {
   fetch?: DatasourceFetch;
+  now?: () => number;
   prometheusBaseUrl?: string;
   prometheusRangeSeconds?: number;
   prometheusStepSeconds?: number;
@@ -113,9 +114,7 @@ export async function resolveDashboardTile(
     const item = module.id === "runtime-health-summary"
       ? runtimeHealthSummary(
           module,
-          await Promise.all(
+          await serviceGroupsSnapshot(document, context),
-            document.serviceGroups.map((group) => resolveServiceGroup(group, context)),
-          ),
         )
       : await resolveModule(module, context);
 
@@ -132,7 +131,7 @@ export async function resolveDashboardTile(
     item: await resolveStatusTile(
       statusItem,
       document.metadata.refreshIntervalSeconds,
-      document.serviceGroups,
+      document,
       context,
     ),
   };
@@ -140,14 +139,27 @@ export async function resolveDashboardTile(
 
 interface DatasourceContext {
   fetch: DatasourceFetch;
+  fetchIdentity: number;
+  now: () => number;
   prometheusBaseUrl: string;
   prometheusRangeSeconds: number;
   prometheusStepSeconds: number;
   requestTimeoutMs: number;
+  serviceGroupsSnapshot?: Promise<ServiceGroup[]>;
 }
 
 type DatasourceFetch = (input: string, init?: RequestInit) => Promise<Response>;
 
+interface ServiceGroupsSnapshotEntry {
+  expiresAt: number;
+  snapshot: Promise<ServiceGroup[]>;
+}
+
+const serviceGroupsSnapshotTtlMs = 30_000;
+const serviceGroupsSnapshotCache = new Map<string, ServiceGroupsSnapshotEntry>();
+const datasourceFetchIdentities = new WeakMap<DatasourceFetch, number>();
+let nextDatasourceFetchIdentity = 1;
+
 interface PrometheusVectorResult {
   metric?: Record<string, string>;
   value?: [number, string];
@@ -159,8 +171,12 @@ interface PrometheusMatrixResult {
 }
 
 function datasourceContext(options: DatasourceResolutionOptions): DatasourceContext {
+  const fetch = options.fetch || globalThis.fetch;
+
   return {
-    fetch: options.fetch || globalThis.fetch,
+    fetch,
+    fetchIdentity: datasourceFetchIdentity(fetch),
+    now: options.now || Date.now,
     prometheusBaseUrl:
       options.prometheusBaseUrl ||
       process.env.PROMETHEUS_BASE_URL ||
@@ -171,6 +187,68 @@ function datasourceContext(options: DatasourceResolutionOptions): DatasourceCont
   };
 }
 
+function datasourceFetchIdentity(fetch: DatasourceFetch): number {
+  const existing = datasourceFetchIdentities.get(fetch);
+  if (existing) return existing;
+
+  const next = nextDatasourceFetchIdentity;
+  nextDatasourceFetchIdentity += 1;
+  datasourceFetchIdentities.set(fetch, next);
+  return next;
+}
+
+function serviceGroupsSnapshot(
+  document: DashboardDocument,
+  context: DatasourceContext,
+): Promise<ServiceGroup[]> {
+  if (context.serviceGroupsSnapshot) return context.serviceGroupsSnapshot;
+
+  const key = serviceGroupsSnapshotKey(document, context);
+  const now = context.now();
+  const cached = serviceGroupsSnapshotCache.get(key);
+  if (cached && cached.expiresAt > now) {
+    context.serviceGroupsSnapshot = cached.snapshot;
+    return cached.snapshot;
+  }
+
+  const snapshot = Promise.all(
+    document.serviceGroups.map((group) => resolveServiceGroup(group, context)),
+  );
+  context.serviceGroupsSnapshot = snapshot;
+  serviceGroupsSnapshotCache.set(key, {
+    expiresAt: now + serviceGroupsSnapshotTtlMs,
+    snapshot,
+  });
+  snapshot.catch(() => {
+    if (serviceGroupsSnapshotCache.get(key)?.snapshot === snapshot) {
+      serviceGroupsSnapshotCache.delete(key);
+    }
+  });
+  return snapshot;
+}
+
+function serviceGroupsSnapshotKey(
+  document: DashboardDocument,
+  context: DatasourceContext,
+): string {
+  return JSON.stringify(
+    {
+      fetchIdentity: context.fetchIdentity,
+      prometheusBaseUrl: context.prometheusBaseUrl,
+      prometheusRangeSeconds: context.prometheusRangeSeconds,
+      prometheusStepSeconds: context.prometheusStepSeconds,
+      requestTimeoutMs: context.requestTimeoutMs,
+      serviceGroups: document.serviceGroups.map((group) => ({
+        id: group.id,
+        services: group.services.map((service) => ({
+          datasource: service.datasource,
+          id: service.id,
+        })),
+      })),
+    },
+  );
+}
+
 async function resolveTelemetryCard(
   card: TelemetryCard,
   context: DatasourceContext,
@@ -485,13 +563,11 @@ function resolveStatusItem(
 async function resolveStatusTile(
   item: StatusItem,
   refreshIntervalSeconds: number | undefined,
-  serviceGroups: ServiceGroup[],
+  document: DashboardDocument,
   context: DatasourceContext,
 ): Promise<StatusItem> {
   if (item.id === "system-status") {
-    const resolvedGroups = await Promise.all(
+    const resolvedGroups = await serviceGroupsSnapshot(document, context);
-      serviceGroups.map((group) => resolveServiceGroup(group, context)),
-    );
     const health = serviceHealthSummary(resolvedGroups);
     return {
       ...structuredClone(item),

apps/web/src/lib/workspace-boundary.test.ts

@@ -1,4 +1,6 @@
-import { existsSync, readFileSync } from "node:fs";
+import { spawnSync } from "node:child_process";
+import { chmodSync, existsSync, mkdtempSync, readFileSync, writeFileSync } from "node:fs";
+import { tmpdir } from "node:os";
 import { join } from "node:path";
 import { describe, expect, test } from "vitest";
 
@@ -42,6 +44,8 @@ describe("workspace boundaries", () => {
       "turbo run check test:unit build @dimensionlab/ui#build-storybook @dimensionlab/web#test:e2e",
     );
     expect(webPackage.scripts).not.toHaveProperty("test:qa");
+    expect(webPackage.scripts?.test).toBe("bun --bun vitest run");
+    expect(webPackage.scripts?.["test:unit"]).toBe("bun --bun vitest run");
     expect(turboConfig.tasks).not.toHaveProperty("test:qa");
     expect(turboConfig.globalDependencies).toEqual(
       expect.arrayContaining(["bun.lock", "tsconfig.base.json"]),
@@ -253,6 +257,186 @@ describe("workspace boundaries", () => {
       "rm -rf build && vite build && bun build src/server/index.ts --target bun --outdir build",
     );
   });
+
+  test("defines Forgejo CI and main-branch deploy automation", () => {
+    const workflow = readFileSync(
+      join(root, ".forgejo/workflows/dimensionlab-website.yml"),
+      "utf8",
+    );
+
+    expect(workflow).toContain("name: Dimension Lab website");
+    expect(workflow).toContain("pull_request:");
+    expect(workflow).toContain("push:");
+    expect(workflow).toContain("branches:");
+    expect(workflow).toContain("- main");
+    expect(workflow).toContain("runs-on: docker");
+    expect(workflow).toContain("bun install --frozen-lockfile");
+    expect(workflow).toContain("bun run check");
+    expect(workflow).toContain("bun run test");
+    expect(workflow).toContain("bun run build");
+    expect(workflow).toContain("needs: ci");
+    expect(workflow).toContain("runs-on: deploy");
+    expect(workflow).toContain("github.event_name == 'push'");
+    expect(workflow).toContain("github.ref == 'refs/heads/main'");
+    expect(workflow).toContain(
+      "git remote add origin git@git.dimensionlab.net:vince/dimensionlab-website.git",
+    );
+    expect(workflow).toContain('git fetch --force --prune --depth=1 origin "$GITHUB_SHA"');
+    expect(workflow).toContain("podman inspect dimensionlab-website");
+    expect(workflow).toContain("DEPLOY_CONTAINER_CLI: podman");
+    expect(workflow).toContain("PODMAN_SYSTEMD_UNIT");
+    expect(workflow).toContain("scripts/deploy-dimensionlab-website.sh");
+  });
+
+  test("keeps production deployment behind a guarded script", () => {
+    const deployScript = readFileSync(
+      join(root, "scripts/deploy-dimensionlab-website.sh"),
+      "utf8",
+    );
+
+    expect(deployScript).toContain("refs/heads/main");
+    expect(deployScript).toContain("dimensionlab-website.service");
+    expect(deployScript).toContain("localhost/dimensionlab-website");
+    expect(deployScript).toContain("apps/web/Containerfile");
+    expect(deployScript).toContain("rollback-");
+    expect(deployScript).toContain("https://dimensionlab.net");
+    expect(deployScript).toContain("/api/dashboard/tiles");
+    expect(deployScript).toContain("--dry-run");
+    expect(deployScript).toContain("DEPLOY_RESTART_STRATEGY");
+  });
+
+  test("rolls back the latest image when production smoke checks fail", () => {
+    const result = runDeployScriptWithFakes(
+      {
+        curl: failingCurlCommand,
+        git: fakeGitCommand,
+        podman: fakePodmanCommand("dimensionlab-website.service"),
+      },
+      {
+        DEPLOY_CONTAINER_CLI: "podman",
+        DEPLOY_EVENT_NAME: "push",
+        DEPLOY_REF: "refs/heads/main",
+        DEPLOY_RESTART_STRATEGY: "quadlet-container",
+        DEPLOY_SHA: "1234567890abcdef",
+        DEPLOY_SMOKE_TIMEOUT_SECONDS: "0",
+      },
+    );
+
+    expect(result.status).toBe(1);
+    expect(result.stderr).toContain("smoke checks failed");
+    expect(result.stderr).toContain("rolling back to localhost/dimensionlab-website:rollback-");
+    expect(result.log).toMatch(
+      /tag localhost\/dimensionlab-website:latest localhost\/dimensionlab-website:rollback-\d{14}/,
+    );
+    expect(result.log).toMatch(
+      /tag localhost\/dimensionlab-website:rollback-\d{14} localhost\/dimensionlab-website:latest/,
+    );
+    expect(result.log.match(/^stop dimensionlab-website$/gm)).toHaveLength(2);
+  });
+
+  test("rolls back the latest image when the container fails to restart", () => {
+    const result = runDeployScriptWithFakes(
+      {
+        curl: passingCurlCommand,
+        git: fakeGitCommand,
+        podman: fakePodmanCommand("dimensionlab-website.service"),
+      },
+      {
+        DEPLOY_CONTAINER_CLI: "podman",
+        DEPLOY_CONTAINER_START_TIMEOUT_SECONDS: "1",
+        DEPLOY_TEST_CONTAINER_IMAGE: "wrong",
+        DEPLOY_EVENT_NAME: "push",
+        DEPLOY_REF: "refs/heads/main",
+        DEPLOY_RESTART_STRATEGY: "quadlet-container",
+        DEPLOY_SHA: "1234567890abcdef",
+      },
+    );
+
+    expect(result.status).toBe(1);
+    expect(result.stderr).toContain("did not restart on localhost/dimensionlab-website:latest");
+    expect(result.stderr).toContain("rolling back to localhost/dimensionlab-website:rollback-");
+    expect(result.stderr).toContain("rollback image is running");
+    expect(result.log).toMatch(
+      /tag localhost\/dimensionlab-website:rollback-\d{14} localhost\/dimensionlab-website:latest/,
+    );
+    expect(result.log.match(/^stop dimensionlab-website$/gm)).toHaveLength(2);
+  });
+
+  test.each([
+    {
+      env: { DEPLOY_EVENT_NAME: "pull_request", DEPLOY_REF: "refs/heads/main" },
+      message: "production deploys only run for push",
+    },
+    {
+      env: { DEPLOY_EVENT_NAME: "push", DEPLOY_REF: "refs/heads/codex/test" },
+      message: "expected refs/heads/main",
+    },
+  ])("refuses guarded deploy contexts before host mutations", ({ env, message }) => {
+    const result = runDeployScriptWithFakes(
+      {
+        curl: passingCurlCommand,
+        git: fakeGitCommand,
+        podman: fakePodmanCommand("dimensionlab-website.service"),
+      },
+      {
+        DEPLOY_CONTAINER_CLI: "podman",
+        DEPLOY_SHA: "1234567890abcdef",
+        ...env,
+      },
+    );
+
+    expect(result.status).toBe(1);
+    expect(result.stderr).toContain(message);
+    expect(result.log).toBe("");
+  });
+
+  test("refuses stop-based deploys unless the container belongs to the expected unit", () => {
+    const result = runDeployScriptWithFakes(
+      {
+        curl: passingCurlCommand,
+        git: fakeGitCommand,
+        podman: fakePodmanCommand("other.service"),
+      },
+      {
+        DEPLOY_CONTAINER_CLI: "podman",
+        DEPLOY_EVENT_NAME: "push",
+        DEPLOY_REF: "refs/heads/main",
+        DEPLOY_RESTART_STRATEGY: "quadlet-container",
+        DEPLOY_SHA: "1234567890abcdef",
+      },
+    );
+
+    expect(result.status).toBe(1);
+    expect(result.stderr).toContain(
+      "refusing to stop dimensionlab-website; expected PODMAN_SYSTEMD_UNIT=dimensionlab-website.service",
+    );
+    expect(result.log).not.toContain("build ");
+    expect(result.log).not.toContain("stop dimensionlab-website");
+  });
+
+  test("checks the expected unit before auto falls back to stopping the container", () => {
+    const result = runDeployScriptWithFakes(
+      {
+        curl: passingCurlCommand,
+        git: fakeGitCommand,
+        podman: fakePodmanCommand("other.service"),
+        systemctl: fakeSystemctlCommand({ active: false, show: true }),
+      },
+      {
+        DEPLOY_CONTAINER_CLI: "podman",
+        DEPLOY_EVENT_NAME: "push",
+        DEPLOY_REF: "refs/heads/main",
+        DEPLOY_RESTART_STRATEGY: "auto",
+        DEPLOY_SHA: "1234567890abcdef",
+      },
+    );
+
+    expect(result.status).toBe(1);
+    expect(result.stderr).toContain(
+      "refusing to stop dimensionlab-website; expected PODMAN_SYSTEMD_UNIT=dimensionlab-website.service",
+    );
+    expect(result.log).not.toContain("stop dimensionlab-website");
+  });
 });
 
 type WorkspacePackageExport =
@@ -269,3 +453,126 @@ function expectPackageExport(
 ): void {
   expect(actual).toMatchObject(expected);
 }
+
+function runDeployScriptWithFakes(
+  commands: Record<string, string>,
+  env: Record<string, string>,
+): { log: string; status: number | null; stderr: string; stdout: string } {
+  const tempDir = mkdtempSync(join(tmpdir(), "dimensionlab-deploy-test-"));
+  const logPath = join(tempDir, "commands.log");
+
+  for (const [name, source] of Object.entries(commands)) {
+    const commandPath = join(tempDir, name);
+    writeFileSync(commandPath, source);
+    chmodSync(commandPath, 0o755);
+  }
+
+  const result = spawnSync("bash", [join(root, "scripts/deploy-dimensionlab-website.sh")], {
+    cwd: root,
+    encoding: "utf8",
+    env: {
+      ...process.env,
+      ...env,
+      DEPLOY_TEST_LOG: logPath,
+      PATH: `${tempDir}:${process.env.PATH ?? ""}`,
+    },
+  });
+
+  return {
+    log: existsSync(logPath) ? readFileSync(logPath, "utf8") : "",
+    status: result.status,
+    stderr: result.stderr,
+    stdout: result.stdout,
+  };
+}
+
+const fakeGitCommand = `#!/usr/bin/env bash
+case "$1" in
+  branch)
+    echo main
+    ;;
+  rev-parse)
+    echo 1234567890abcdef
+    ;;
+  config|submodule)
+    exit 0
+    ;;
+esac
+`;
+
+function fakePodmanCommand(systemdUnit: string): string {
+  return `#!/usr/bin/env bash
+state_file="$DEPLOY_TEST_LOG.state"
+[ -f "$state_file" ] || printf 'initial' > "$state_file"
+printf '%s\\n' "$*" >> "$DEPLOY_TEST_LOG"
+if [ "$1" = "image" ] && [ "$2" = "inspect" ]; then
+  if [ "$4" = "--format" ]; then
+    case "$3" in
+      *:rollback-*)
+        echo sha256:old
+        ;;
+      *)
+        if [ "$(cat "$state_file")" = "rollback" ]; then
+          echo sha256:old
+        else
+          echo sha256:new
+        fi
+        ;;
+    esac
+  fi
+  exit 0
+fi
+if [ "$1" = "tag" ] && [ "$2" != "localhost/dimensionlab-website:latest" ]; then
+  printf 'rollback' > "$state_file"
+fi
+if [ "$1" = "inspect" ]; then
+  case "$*" in
+    *PODMAN_SYSTEMD_UNIT*)
+      echo ${systemdUnit}
+      ;;
+    *State.Running*)
+      echo true
+      ;;
+    *'.Image'*|*'{{.Image}}'*)
+      if [ "$(cat "$state_file")" = "rollback" ]; then
+        echo sha256:old
+      elif [ "\${DEPLOY_TEST_CONTAINER_IMAGE:-new}" = "wrong" ]; then
+        echo sha256:wrong
+      else
+        echo sha256:new
+      fi
+      ;;
+  esac
+fi
+`;
+}
+
+const failingCurlCommand = `#!/usr/bin/env bash
+printf 'curl %s\\n' "$*" >> "$DEPLOY_TEST_LOG"
+exit 22
+`;
+
+const passingCurlCommand = `#!/usr/bin/env bash
+printf 'curl %s\\n' "$*" >> "$DEPLOY_TEST_LOG"
+if [ "$*" = *'/api/dashboard/tiles'* ]; then
+  printf '{"state":"ready","tiles":[]}'
+fi
+`;
+
+function fakeSystemctlCommand(options: { active: boolean; show: boolean }): string {
+  const activeStatus = options.active ? 0 : 3;
+  const showStatus = options.show ? 0 : 1;
+
+  return `#!/usr/bin/env bash
+printf 'systemctl %s\\n' "$*" >> "$DEPLOY_TEST_LOG"
+if [ "$1" = "--user" ] && [ "$2" = "is-active" ]; then
+  exit ${activeStatus}
+fi
+if [ "$1" = "--user" ] && [ "$2" = "show" ]; then
+  exit ${showStatus}
+fi
+if [ "$1" = "--user" ] && [ "$2" = "restart" ]; then
+  exit 0
+fi
+`;
+}

apps/web/src/server/index.test.ts

@@ -0,0 +1,77 @@
+import { afterAll, describe, expect, test, vi } from "vitest";
+import { handleRequest } from "./index";
+
+describe("server request routing", () => {
+  const consoleInfo = vi.spyOn(console, "info").mockImplementation(() => undefined);
+
+  afterAll(() => {
+    consoleInfo.mockRestore();
+  });
+
+  test("routes dashboard tile batch requests", async () => {
+    const response = await handleRequest(
+      new Request("https://example.test/api/dashboard/tiles", {
+        method: "POST",
+        body: JSON.stringify({
+          tiles: [
+            {
+              kind: "status",
+              stripId: "footer-status",
+              id: "auto-refresh",
+            },
+          ],
+        }),
+      }),
+    );
+
+    expect(response.status).toBe(200);
+    await expect(response.json()).resolves.toMatchObject({
+      state: "ready",
+      tiles: [
+        {
+          state: "ready",
+          tile: {
+            kind: "status",
+            stripId: "footer-status",
+            id: "auto-refresh",
+          },
+        },
+      ],
+    });
+  });
+
+  test("rejects non-post dashboard tile batch requests", async () => {
+    const response = await handleRequest(
+      new Request("https://example.test/api/dashboard/tiles", {
+        method: "GET",
+      }),
+    );
+
+    expect(response.status).toBe(405);
+    expect(response.headers.get("allow")).toBe("POST");
+  });
+
+  test("routes dashboard event stream requests", async () => {
+    const response = await handleRequest(
+      new Request("https://example.test/api/dashboard/events", {
+        method: "GET",
+      }),
+    );
+
+    expect(response.status).toBe(200);
+    expect(response.headers.get("content-type")).toBe(
+      "text/event-stream; charset=utf-8",
+    );
+  });
+
+  test("rejects non-get dashboard event stream requests", async () => {
+    const response = await handleRequest(
+      new Request("https://example.test/api/dashboard/events", {
+        method: "POST",
+      }),
+    );
+
+    expect(response.status).toBe(405);
+    expect(response.headers.get("allow")).toBe("GET");
+  });
+});

apps/web/src/server/index.ts

@@ -1,6 +1,11 @@
 import { extname, normalize } from "node:path";
 import { handleAgentDashboardRoute } from "./routes/agent-dashboard";
-import { handleDashboardRoute, handleDashboardTileRoute } from "./routes/dashboard";
+import {
+  handleDashboardEventsRoute,
+  handleDashboardRoute,
+  handleDashboardTileRoute,
+  handleDashboardTilesRoute,
+} from "./routes/dashboard";
 
 const host = process.env.HOST || "0.0.0.0";
 const port = Number(process.env.PORT || 3000);
@@ -23,6 +28,16 @@ export async function handleRequest(request: Request): Promise<Response> {
     return handleDashboardRoute();
   }
 
+  if (url.pathname === "/api/dashboard/tiles") {
+    if (request.method !== "POST") return methodNotAllowed(["POST"]);
+    return handleDashboardTilesRoute(request);
+  }
+
+  if (url.pathname === "/api/dashboard/events") {
+    if (request.method !== "GET") return methodNotAllowed(["GET"]);
+    return handleDashboardEventsRoute({ signal: request.signal });
+  }
+
   if (url.pathname.startsWith("/api/dashboard/tile/")) {
     if (request.method !== "GET") return methodNotAllowed(["GET"]);
     return handleDashboardTileRoute(url.pathname);

apps/web/src/server/routes/dashboard.test.ts

@@ -1,8 +1,28 @@
-import { describe, expect, test, vi } from "vitest";
+import { afterAll, afterEach, describe, expect, test, vi } from "vitest";
 import { dimensionLabDashboardFixture } from "$lib/dashboard-seed/dimensionlab";
-import { loadDashboardResponse, loadDashboardTileResponse } from "./dashboard";
+import {
+  createDashboardTileCache,
+  dashboardTileCacheKey,
+  handleDashboardEventsRoute,
+  handleDashboardTilesRoute,
+  handleDashboardTileRoute,
+  loadDashboardResponse,
+  loadDashboardTilesResponse,
+  loadDashboardTileResponse,
+  type DashboardTileResolutionLogEvent,
+} from "./dashboard";
 
 describe("dashboard API route", () => {
+  const consoleInfo = vi.spyOn(console, "info").mockImplementation(() => undefined);
+
+  afterEach(() => {
+    consoleInfo.mockClear();
+  });
+
+  afterAll(() => {
+    consoleInfo.mockRestore();
+  });
+
   test("returns the ready dashboard shell without hydrating live datasources", async () => {
     const fetch = vi.spyOn(globalThis, "fetch").mockRejectedValue(
       new Error("live datasource fetch should not run for the shell response"),
@@ -145,6 +165,555 @@ describe("dashboard API route", () => {
     });
   });
 
+  test("caches ready tile responses until the tile ttl expires", async () => {
+    const cache = createDashboardTileCache();
+    let now = 1_000;
+    const fetch = vi.fn(async () => {
+      now += 7;
+      return jsonResponse({
+        status: "UP",
+        ping: 42,
+      });
+    });
+    const tile = { kind: "service", groupId: "essentials", id: "vaultwarden" } as const;
+
+    const first = await loadDashboardTileResponse(tile, {
+      fetch,
+      now: () => now,
+      refreshSeedDocument: true,
+      seedIfEmpty: true,
+      tileCache: cache,
+    });
+    const second = await loadDashboardTileResponse(tile, {
+      fetch,
+      now: () => now + 29_999,
+      refreshSeedDocument: true,
+      seedIfEmpty: true,
+      tileCache: cache,
+    });
+    now += 30_001;
+    const third = await loadDashboardTileResponse(tile, {
+      fetch,
+      now: () => now,
+      refreshSeedDocument: true,
+      seedIfEmpty: true,
+      tileCache: cache,
+    });
+
+    expect(first).toEqual(second);
+    expect(third.state).toBe("ready");
+    expect(fetch).toHaveBeenCalledTimes(2);
+  });
+
+  test("logs tile duration and cache hit or miss metadata", async () => {
+    let now = 1_000;
+    const logs: DashboardTileResolutionLogEvent[] = [];
+    const tile = { kind: "service", groupId: "essentials", id: "vaultwarden" } as const;
+    const service = dimensionLabDashboardFixture.serviceGroups
+      .flatMap((group) => group.services)
+      .find((item) => item.id === tile.id);
+    if (!service) throw new Error("missing service fixture");
+
+    let cacheCalls = 0;
+    const tileCache = {
+      async resolve() {
+        cacheCalls += 1;
+        if (cacheCalls === 1) now += 7;
+        const cacheState = cacheCalls === 1 ? "miss" as const : "hit" as const;
+
+        return {
+          cache: cacheState,
+          coalesced: false,
+          response: {
+            state: "ready" as const,
+            tile,
+            item: service,
+          },
+        };
+      },
+    };
+
+    await loadDashboardTileResponse(tile, {
+      logTileResolution: (event) => logs.push(event),
+      now: () => now,
+      refreshSeedDocument: true,
+      seedIfEmpty: true,
+      tileCache,
+    });
+    now += 10;
+    await loadDashboardTileResponse(tile, {
+      logTileResolution: (event) => logs.push(event),
+      now: () => now,
+      refreshSeedDocument: true,
+      seedIfEmpty: true,
+      tileCache,
+    });
+
+    expect(logs).toEqual([
+      expect.objectContaining({
+        cache: "miss",
+        coalesced: false,
+        durationMs: 7,
+        status: "ready",
+        tileKey: dashboardTileCacheKey(tile),
+      }),
+      expect.objectContaining({
+        cache: "hit",
+        coalesced: false,
+        durationMs: 0,
+        status: "ready",
+        tileKey: dashboardTileCacheKey(tile),
+      }),
+    ]);
+  });
+
+  test("keeps telemetry tiles cached for fifteen seconds", async () => {
+    const cache = createDashboardTileCache();
+    let now = 1_000;
+    const fetch = telemetryFetch();
+    const tile = { kind: "telemetry", id: "infra-ram" } as const;
+
+    await loadDashboardTileResponse(tile, {
+      fetch,
+      now: () => now,
+      prometheusBaseUrl: "https://prometheus.example",
+      refreshSeedDocument: true,
+      seedIfEmpty: true,
+      tileCache: cache,
+    });
+    await loadDashboardTileResponse(tile, {
+      fetch,
+      now: () => now + 14_999,
+      prometheusBaseUrl: "https://prometheus.example",
+      refreshSeedDocument: true,
+      seedIfEmpty: true,
+      tileCache: cache,
+    });
+    now += 15_001;
+    await loadDashboardTileResponse(tile, {
+      fetch,
+      now: () => now,
+      prometheusBaseUrl: "https://prometheus.example",
+      refreshSeedDocument: true,
+      seedIfEmpty: true,
+      tileCache: cache,
+    });
+
+    expect(fetch).toHaveBeenCalledTimes(4);
+  });
+
+  test("keeps weather module tiles cached for ten minutes", async () => {
+    const cache = createDashboardTileCache();
+    let now = 1_000;
+    const fetch = vi.fn(async () =>
+      jsonResponse({
+        current: {
+          apparent_temperature: 19,
+          temperature_2m: 20,
+          weather_code: 0,
+          wind_speed_10m: 11,
+        },
+      }),
+    );
+    const tile = { kind: "module", id: "weather-amsterdam" } as const;
+
+    await loadDashboardTileResponse(tile, {
+      fetch,
+      now: () => now,
+      refreshSeedDocument: true,
+      seedIfEmpty: true,
+      tileCache: cache,
+    });
+    await loadDashboardTileResponse(tile, {
+      fetch,
+      now: () => now + 599_999,
+      refreshSeedDocument: true,
+      seedIfEmpty: true,
+      tileCache: cache,
+    });
+    now += 600_001;
+    await loadDashboardTileResponse(tile, {
+      fetch,
+      now: () => now,
+      refreshSeedDocument: true,
+      seedIfEmpty: true,
+      tileCache: cache,
+    });
+
+    expect(fetch).toHaveBeenCalledTimes(2);
+  });
+
+  test("coalesces concurrent tile requests for the same cache key", async () => {
+    const cache = createDashboardTileCache();
+    const logs: DashboardTileResolutionLogEvent[] = [];
+    let resolveFetch: ((response: Response) => void) | undefined;
+    const fetch = vi.fn(() =>
+      new Promise<Response>((resolve) => {
+        resolveFetch = resolve;
+      })
+    );
+    const tile = { kind: "service", groupId: "essentials", id: "vaultwarden" } as const;
+
+    const first = loadDashboardTileResponse(tile, {
+      fetch,
+      logTileResolution: (event) => logs.push(event),
+      now: () => 1_000,
+      refreshSeedDocument: true,
+      seedIfEmpty: true,
+      tileCache: cache,
+    });
+    const second = loadDashboardTileResponse(tile, {
+      fetch,
+      logTileResolution: (event) => logs.push(event),
+      now: () => 1_000,
+      refreshSeedDocument: true,
+      seedIfEmpty: true,
+      tileCache: cache,
+    });
+
+    await Promise.resolve();
+    expect(fetch).toHaveBeenCalledTimes(1);
+
+    resolveFetch?.(jsonResponse({ status: "UP", ping: 42 }));
+    expect(await first).toEqual(await second);
+    expect(logs).toEqual([
+      expect.objectContaining({
+        cache: "miss",
+        coalesced: false,
+        status: "ready",
+        tileKey: dashboardTileCacheKey(tile),
+      }),
+      expect.objectContaining({
+        cache: "miss",
+        coalesced: true,
+        status: "ready",
+        tileKey: dashboardTileCacheKey(tile),
+      }),
+    ]);
+  });
+
+  test("logs coalesced metadata when shared tile requests fail", async () => {
+    const cache = createDashboardTileCache();
+    let rejectLoad: ((error: Error) => void) | undefined;
+    const load = vi.fn(() =>
+      new Promise<never>((_resolve, reject) => {
+        rejectLoad = reject;
+      })
+    );
+
+    const first = cache.resolve("tile-a", 30_000, 1_000, load);
+    const second = cache.resolve("tile-a", 30_000, 1_000, load);
+
+    await Promise.resolve();
+    expect(load).toHaveBeenCalledTimes(1);
+
+    rejectLoad?.(new TypeError("upstream failed"));
+    await expect(first).rejects.toThrow("upstream failed");
+    await expect(second).rejects.toMatchObject({
+      cache: "miss",
+      coalesced: true,
+      cause: expect.any(TypeError),
+    });
+  });
+
+  test("logs coalesced metadata for failed tile cache resolution", async () => {
+    const logs: DashboardTileResolutionLogEvent[] = [];
+    const tile = { kind: "service", groupId: "essentials", id: "vaultwarden" } as const;
+
+    await expect(
+      loadDashboardTileResponse(tile, {
+        logTileResolution: (event) => logs.push(event),
+        refreshSeedDocument: true,
+        seedIfEmpty: true,
+        tileCache: {
+          async resolve() {
+            throw {
+              cache: "miss",
+              cause: new TypeError("coalesced cache failed"),
+              coalesced: true,
+            };
+          },
+        },
+      }),
+    ).rejects.toThrow("coalesced cache failed");
+
+    expect(logs).toEqual([
+      expect.objectContaining({
+        cache: "miss",
+        coalesced: true,
+        errorCategory: "TypeError",
+        status: "error",
+        tileKey: dashboardTileCacheKey(tile),
+      }),
+    ]);
+  });
+
+  test("logs error categories for failed tile cache resolution", async () => {
+    const logs: DashboardTileResolutionLogEvent[] = [];
+    const tile = { kind: "service", groupId: "essentials", id: "vaultwarden" } as const;
+
+    await expect(
+      loadDashboardTileResponse(tile, {
+        logTileResolution: (event) => logs.push(event),
+        refreshSeedDocument: true,
+        seedIfEmpty: true,
+        tileCache: {
+          async resolve() {
+            throw new TypeError("cache failed");
+          },
+        },
+      }),
+    ).rejects.toThrow("cache failed");
+
+    expect(logs).toEqual([
+      expect.objectContaining({
+        cache: "miss",
+        coalesced: false,
+        errorCategory: "TypeError",
+        status: "error",
+        tileKey: dashboardTileCacheKey(tile),
+      }),
+    ]);
+  });
+
+  test("uses structured tile cache keys when identifiers contain delimiters", () => {
+    expect(
+      dashboardTileCacheKey({ kind: "service", groupId: "a:b", id: "c" }),
+    ).not.toBe(
+      dashboardTileCacheKey({ kind: "service", groupId: "a", id: "b:c" }),
+    );
+    expect(
+      dashboardTileCacheKey({ kind: "status", stripId: "a:b", id: "c" }),
+    ).not.toBe(
+      dashboardTileCacheKey({ kind: "status", stripId: "a", id: "b:c" }),
+    );
+  });
+
+  test("uses thirty-second status aggregate and five-minute static status ttl buckets", async () => {
+    const cache = createDashboardTileCache();
+    const fetch = vi.fn(async () => jsonResponse({ status: "UP", ping: 42 }));
+    const serviceCheckCount = dimensionLabDashboardFixture.serviceGroups
+      .flatMap((group) => group.services).length;
+
+    await loadDashboardTileResponse(
+      { kind: "status", stripId: "footer-status", id: "system-status" },
+      {
+        fetch,
+        now: () => 1_000,
+        refreshSeedDocument: true,
+        seedIfEmpty: true,
+        tileCache: cache,
+      },
+    );
+    await loadDashboardTileResponse(
+      { kind: "status", stripId: "footer-status", id: "system-status" },
+      {
+        fetch,
+        now: () => 31_001,
+        refreshSeedDocument: true,
+        seedIfEmpty: true,
+        tileCache: cache,
+      },
+    );
+
+    await loadDashboardTileResponse(
+      { kind: "status", stripId: "footer-status", id: "auto-refresh" },
+      {
+        fetch,
+        now: () => 1_000,
+        refreshSeedDocument: true,
+        seedIfEmpty: true,
+        tileCache: cache,
+      },
+    );
+    await loadDashboardTileResponse(
+      { kind: "status", stripId: "footer-status", id: "auto-refresh" },
+      {
+        fetch,
+        now: () => 300_999,
+        refreshSeedDocument: true,
+        seedIfEmpty: true,
+        tileCache: cache,
+      },
+    );
+    await loadDashboardTileResponse(
+      { kind: "status", stripId: "footer-status", id: "auto-refresh" },
+      {
+        fetch,
+        now: () => 301_001,
+        refreshSeedDocument: true,
+        seedIfEmpty: true,
+        tileCache: cache,
+      },
+    );
+
+    expect(fetch).toHaveBeenCalledTimes(serviceCheckCount * 2);
+  });
+
+  test("serves tile route responses with short private cache headers", async () => {
+    const response = await handleDashboardTileRoute(
+      "/api/dashboard/tile/service/essentials/vaultwarden",
+      {
+        fetch: vi.fn(async () => jsonResponse({ status: "UP", ping: 42 })),
+        refreshSeedDocument: true,
+        seedIfEmpty: true,
+        tileCache: createDashboardTileCache(),
+      },
+    );
+
+    expect(response.headers.get("cache-control")).toBe(
+      "private, max-age=5, stale-while-revalidate=30",
+    );
+  });
+
+  test("serves batch tile route responses", async () => {
+    const response = await handleDashboardTilesRoute(
+      new Request("https://example.test/api/dashboard/tiles", {
+        method: "POST",
+        body: JSON.stringify({
+          tiles: [
+            {
+              kind: "status",
+              stripId: "footer-status",
+              id: "auto-refresh",
+            },
+          ],
+        }),
+      }),
+      {
+        refreshSeedDocument: true,
+        seedIfEmpty: true,
+        tileCache: createDashboardTileCache(),
+      },
+    );
+
+    expect(response.status).toBe(200);
+    expect(response.headers.get("cache-control")).toBe(
+      "private, max-age=5, stale-while-revalidate=30",
+    );
+    await expect(response.json()).resolves.toMatchObject({
+      state: "ready",
+      tiles: [
+        {
+          state: "ready",
+          tile: {
+            kind: "status",
+            stripId: "footer-status",
+            id: "auto-refresh",
+          },
+        },
+      ],
+    });
+  });
+
+  test("streams ready dashboard tile events", async () => {
+    const controller = new AbortController();
+    const response = await handleDashboardEventsRoute({
+      refreshSeedDocument: true,
+      seedIfEmpty: true,
+      signal: controller.signal,
+      tileCache: {
+        async resolve(key) {
+          controller.abort();
+          const tile = JSON.parse(key);
+
+          return {
+            cache: "miss",
+            coalesced: false,
+            response: {
+              state: "ready",
+              tile,
+              item: {
+                id: tile.id,
+                label: "Status",
+                severity: "ok",
+                value: "ok",
+              },
+            },
+          };
+        },
+      },
+    });
+
+    expect(response.headers.get("content-type")).toBe(
+      "text/event-stream; charset=utf-8",
+    );
+    expect(response.headers.get("cache-control")).toBe("no-cache");
+    const body = await response.text();
+    expect(body).toContain("event: dashboard-tile");
+    expect(body).toContain('"state":"ready"');
+    expect(body).toContain('"tile"');
+  });
+
+  test("rejects invalid batch tile requests", async () => {
+    const response = await handleDashboardTilesRoute(
+      new Request("https://example.test/api/dashboard/tiles", {
+        method: "POST",
+        body: JSON.stringify({
+          tiles: [{ kind: "service", id: "missing-group" }],
+        }),
+      }),
+    );
+
+    expect(response.status).toBe(400);
+  });
+
+  test("resolves batch tile responses with server concurrency capped at six", async () => {
+    let active = 0;
+    let maxActive = 0;
+    const started: string[] = [];
+    const releases = new Map<string, () => void>();
+    const service = dimensionLabDashboardFixture.serviceGroups
+      .flatMap((group) => group.services)[0];
+    if (!service) throw new Error("missing service fixture");
+    const tiles = Array.from({ length: 7 }, (_, index) => ({
+      kind: "service" as const,
+      groupId: "essentials",
+      id: `service-${index}`,
+    }));
+
+    const batch = loadDashboardTilesResponse(tiles, {
+      refreshSeedDocument: true,
+      seedIfEmpty: true,
+      tileCache: {
+        async resolve(key) {
+          active += 1;
+          maxActive = Math.max(maxActive, active);
+          started.push(key);
+          await new Promise<void>((resolve) => releases.set(key, resolve));
+          active -= 1;
+
+          return {
+            cache: "miss",
+            coalesced: false,
+            response: {
+              state: "ready",
+              tile: JSON.parse(key),
+              item: service,
+            },
+          };
+        },
+      },
+    });
+
+    await waitFor(() => started.length === 6);
+    expect(maxActive).toBe(6);
+    releases.get(started[0])?.();
+    await waitFor(() => started.length === 7);
+    for (const release of releases.values()) release();
+
+    await expect(batch).resolves.toMatchObject({
+      state: "ready",
+      tiles: expect.arrayContaining([
+        expect.objectContaining({
+          state: "ready",
+          tile: tiles[0],
+        }),
+      ]),
+    });
+    expect(maxActive).toBe(6);
+  });
+
   test("does not hydrate tile routes when live datasources are disabled", async () => {
     const previous = process.env.DISABLE_LIVE_DATASOURCES;
     process.env.DISABLE_LIVE_DATASOURCES = "1";
@@ -179,3 +748,52 @@ function jsonResponse(payload: unknown): Response {
     headers: { "content-type": "application/json" },
   });
 }
+
+async function waitFor(predicate: () => boolean) {
+  for (let attempt = 0; attempt < 20; attempt += 1) {
+    if (predicate()) return;
+    await Promise.resolve();
+  }
+
+  throw new Error("condition was not met");
+}
+
+function telemetryFetch() {
+  return vi.fn(async (input: RequestInfo | URL) => {
+    const url = String(input);
+
+    if (url.startsWith("https://prometheus.example/api/v1/query_range")) {
+      return jsonResponse({
+        status: "success",
+        data: {
+          result: [
+            {
+              metric: { host: "linux-infra" },
+              values: [
+                [1771430000, "10"],
+                [1771430060, "20"],
+                [1771430120, "42"],
+              ],
+            },
+          ],
+        },
+      });
+    }
+
+    if (url.startsWith("https://prometheus.example/api/v1/query")) {
+      return jsonResponse({
+        status: "success",
+        data: {
+          result: [
+            {
+              metric: { host: "linux-infra" },
+              value: [1771430400, "42"],
+            },
+          ],
+        },
+      });
+    }
+
+    throw new Error(`Unhandled test request: ${url}`);
+  });
+}

apps/web/src/server/routes/dashboard.ts

@@ -1,3 +1,4 @@
+import type { DashboardDocument } from "@dimensionlab/dashboard-model";
 import {
   loadDashboardRuntime,
   type DashboardRuntimeOptions,
@@ -19,6 +20,111 @@ export interface LoadDashboardResponseOptions
     DatasourceResolutionOptions {
   disableLiveDatasources?: boolean;
   hydrateLiveDatasources?: boolean;
+  logTileResolution?: (event: DashboardTileResolutionLogEvent) => void;
+  now?: () => number;
+  tileCache?: DashboardTileCache;
+}
+
+export interface DashboardEventsRouteOptions extends LoadDashboardResponseOptions {
+  signal?: AbortSignal;
+}
+
+interface DashboardTileCacheEntry {
+  expiresAt: number;
+  response: DashboardTileResolution;
+}
+
+export interface DashboardTileCache {
+  resolve(
+    key: string,
+    ttlMs: number,
+    now: number,
+    load: () => Promise<DashboardTileResolution>,
+  ): Promise<DashboardTileCacheResult>;
+}
+
+const dashboardTileResponseHeaders = {
+  "Cache-Control": "private, max-age=5, stale-while-revalidate=30",
+};
+
+const dashboardBatchTileConcurrency = 6;
+
+const defaultDashboardTileCache = createDashboardTileCache();
+
+interface DashboardTileCacheResult {
+  cache: "hit" | "miss";
+  coalesced: boolean;
+  response: DashboardTileResolution;
+}
+
+export interface DashboardTileResolutionLogEvent {
+  cache: "bypass" | "hit" | "miss";
+  coalesced: boolean;
+  durationMs: number;
+  errorCategory?: string;
+  status: DashboardTileResolution["state"] | "error";
+  tileKey: string;
+}
+
+export interface DashboardTilesBatchResponse {
+  state: "ready";
+  tiles: DashboardTileResolution[];
+}
+
+export function createDashboardTileCache(): DashboardTileCache {
+  const entries = new Map<string, DashboardTileCacheEntry>();
+  const inFlight = new Map<string, Promise<DashboardTileResolution>>();
+
+  return {
+    async resolve(key, ttlMs, now, load) {
+      const cached = entries.get(key);
+      if (cached && cached.expiresAt > now) {
+        return {
+          cache: "hit",
+          coalesced: false,
+          response: cached.response,
+        };
+      }
+
+      const active = inFlight.get(key);
+      if (active) {
+        return active
+          .then((response) => ({
+            cache: "miss" as const,
+            coalesced: true,
+            response,
+          }))
+          .catch((error) => {
+            throw new DashboardTileCacheResolutionError(error, {
+              cache: "miss",
+              coalesced: true,
+            });
+          });
+      }
+
+      const request = load()
+        .then((response) => {
+          if (response.state === "ready") {
+            entries.set(key, {
+              expiresAt: now + ttlMs,
+              response,
+            });
+          }
+
+          return response;
+        })
+        .finally(() => {
+          inFlight.delete(key);
+        });
+
+      inFlight.set(key, request);
+      return request.then((response) => ({
+        cache: "miss" as const,
+        coalesced: false,
+        response,
+      }));
+    },
+  };
 }
 
 export async function loadDashboardResponse(
@@ -67,13 +173,26 @@ export async function loadDashboardTileResponse(
     options.disableLiveDatasources ||
     process.env.DISABLE_LIVE_DATASOURCES === "1"
   ) {
-    return {
+    const tileKey = dashboardTileCacheKey(tile);
+    const startedAt = options.now?.() ?? Date.now();
+    const response = {
       state: "disabled",
       tile,
       message: "Live datasource hydration is disabled.",
-    };
+    } satisfies DashboardTileResolution;
+    logDashboardTileResolution(options, {
+      cache: "bypass",
+      coalesced: false,
+      durationMs: elapsedDashboardTileMs(startedAt, options),
+      status: response.state,
+      tileKey,
+    });
+    return response;
   }
 
+  const tileKey = dashboardTileCacheKey(tile);
+  const startedAt = options.now?.() ?? Date.now();
+
   const dashboard = loadDashboardRuntime(undefined, {
     refreshSeedDocument: options.refreshSeedDocument ?? true,
     seedIfEmpty: options.seedIfEmpty ?? true,
@@ -81,21 +200,71 @@ export async function loadDashboardTileResponse(
   });
 
   if (dashboard.state !== "ready") {
-    return {
+    const response = {
       state: "not_found",
       tile,
       message: `Dashboard is not ready: ${dashboard.state}`,
-    };
+    } satisfies DashboardTileResolution;
+    logDashboardTileResolution(options, {
+      cache: "bypass",
+      coalesced: false,
+      durationMs: elapsedDashboardTileMs(startedAt, options),
+      status: response.state,
+      tileKey,
+    });
+    return response;
   }
 
-  return resolveDashboardTile(dashboard.document, tile, options);
+  const cache = options.tileCache || defaultDashboardTileCache;
+  const now = options.now?.() ?? Date.now();
+
+  try {
+    const result = await cache.resolve(
+      tileKey,
+      dashboardTileTtlMs(dashboard.document, tile),
+      now,
+      () => resolveDashboardTile(dashboard.document, tile, options),
+    );
+    logDashboardTileResolution(options, {
+      cache: result.cache,
+      coalesced: result.coalesced,
+      durationMs: elapsedDashboardTileMs(startedAt, options),
+      status: result.response.state,
+      tileKey,
+    });
+    return result.response;
+  } catch (error) {
+    const cacheError = dashboardTileCacheResolutionError(error);
+    logDashboardTileResolution(options, {
+      cache: cacheError?.cache ?? "miss",
+      coalesced: cacheError?.coalesced ?? false,
+      durationMs: elapsedDashboardTileMs(startedAt, options),
+      errorCategory: dashboardTileErrorCategory(cacheError?.cause ?? error),
+      status: "error",
+      tileKey,
+    });
+    throw cacheError?.cause ?? error;
+  }
+}
+
+export async function loadDashboardTilesResponse(
+  tiles: DashboardTileReference[],
+  options: LoadDashboardResponseOptions = {},
+): Promise<DashboardTilesBatchResponse> {
+  return {
+    state: "ready",
+    tiles: await resolveDashboardTilesBatch(tiles, options),
+  };
 }
 
 export async function handleDashboardRoute(): Promise<Response> {
   return Response.json(await loadDashboardResponse());
 }
 
-export async function handleDashboardTileRoute(pathname: string): Promise<Response> {
+export async function handleDashboardTileRoute(
+  pathname: string,
+  options: LoadDashboardResponseOptions = {},
+): Promise<Response> {
   const tile = parseDashboardTilePath(pathname);
   if (!tile) {
     return Response.json(
@@ -104,12 +273,310 @@ export async function handleDashboardTileRoute(pathname: string): Promise<Respon
     );
   }
 
-  const response = await loadDashboardTileResponse(tile);
+  const response = await loadDashboardTileResponse(tile, options);
   return Response.json(response, {
     status: response.state === "not_found" ? 404 : 200,
+    headers: dashboardTileResponseHeaders,
   });
 }
 
+export async function handleDashboardTilesRoute(
+  request: Request,
+  options: LoadDashboardResponseOptions = {},
+): Promise<Response> {
+  const tiles = await parseDashboardTilesBatchRequest(request);
+  if (!tiles) {
+    return Response.json(
+      { ok: false, message: "Invalid dashboard tiles request" },
+      { status: 400 },
+    );
+  }
+
+  return Response.json(await loadDashboardTilesResponse(tiles, options), {
+    headers: dashboardTileResponseHeaders,
+  });
+}
+
+export async function handleDashboardEventsRoute(
+  options: DashboardEventsRouteOptions = {},
+): Promise<Response> {
+  const stream = new ReadableStream<Uint8Array>({
+    async start(controller) {
+      const encoder = new TextEncoder();
+      const dashboard = await loadDashboardResponse({
+        ...options,
+        hydrateLiveDatasources: false,
+      });
+
+      if (dashboard.state !== "ready" || options.signal?.aborted) {
+        controller.close();
+        return;
+      }
+
+      try {
+        for (const tile of dashboardEventTiles(dashboard.document)) {
+          if (options.signal?.aborted) break;
+          const resolution = await loadDashboardTileResponse(tile, options);
+          if (resolution.state === "ready") {
+            controller.enqueue(
+              encoder.encode(dashboardTileEventChunk(resolution)),
+            );
+          }
+        }
+      } finally {
+        controller.close();
+      }
+    },
+  });
+
+  return new Response(stream, {
+    headers: {
+      "Cache-Control": "no-cache",
+      "Connection": "keep-alive",
+      "Content-Type": "text/event-stream; charset=utf-8",
+      "X-Accel-Buffering": "no",
+    },
+  });
+}
+
+export function dashboardTileCacheKey(tile: DashboardTileReference): string {
+  return JSON.stringify(tile);
+}
+
+function logDashboardTileResolution(
+  options: LoadDashboardResponseOptions,
+  event: DashboardTileResolutionLogEvent,
+): void {
+  if (options.logTileResolution) {
+    options.logTileResolution(event);
+    return;
+  }
+
+  console.info("dashboard.tile", event);
+}
+
+function elapsedDashboardTileMs(
+  startedAt: number,
+  options: LoadDashboardResponseOptions,
+): number {
+  const now = options.now?.() ?? Date.now();
+  return Math.max(0, now - startedAt);
+}
+
+function dashboardTileErrorCategory(error: unknown): string {
+  if (
+    typeof error === "object" &&
+    error !== null &&
+    "name" in error &&
+    typeof error.name === "string"
+  ) {
+    return error.name;
+  }
+
+  return "unknown";
+}
+
+async function resolveDashboardTilesBatch(
+  tiles: DashboardTileReference[],
+  options: LoadDashboardResponseOptions,
+): Promise<DashboardTileResolution[]> {
+  const results: DashboardTileResolution[] = new Array(tiles.length);
+  let nextIndex = 0;
+
+  async function worker() {
+    while (nextIndex < tiles.length) {
+      const index = nextIndex;
+      nextIndex += 1;
+      results[index] = await loadDashboardTileResponse(tiles[index], options);
+    }
+  }
+
+  await Promise.all(
+    Array.from(
+      { length: Math.min(dashboardBatchTileConcurrency, tiles.length) },
+      () => worker(),
+    ),
+  );
+
+  return results;
+}
+
+async function parseDashboardTilesBatchRequest(
+  request: Request,
+): Promise<DashboardTileReference[] | null> {
+  try {
+    const body = await request.json() as unknown;
+    if (
+      typeof body !== "object" ||
+      body === null ||
+      !("tiles" in body) ||
+      !Array.isArray(body.tiles)
+    ) {
+      return null;
+    }
+
+    const tiles = body.tiles.map(parseDashboardTileReference);
+    return tiles.every((tile): tile is DashboardTileReference => tile !== null)
+      ? tiles
+      : null;
+  } catch {
+    return null;
+  }
+}
+
+function parseDashboardTileReference(value: unknown): DashboardTileReference | null {
+  if (typeof value !== "object" || value === null || !("kind" in value)) {
+    return null;
+  }
+
+  if (
+    (value.kind === "telemetry" || value.kind === "module") &&
+    "id" in value &&
+    typeof value.id === "string"
+  ) {
+    return { kind: value.kind, id: value.id };
+  }
+
+  if (
+    value.kind === "service" &&
+    "groupId" in value &&
+    typeof value.groupId === "string" &&
+    "id" in value &&
+    typeof value.id === "string"
+  ) {
+    return {
+      kind: "service",
+      groupId: value.groupId,
+      id: value.id,
+    };
+  }
+
+  if (
+    value.kind === "status" &&
+    "stripId" in value &&
+    typeof value.stripId === "string" &&
+    "id" in value &&
+    typeof value.id === "string"
+  ) {
+    return {
+      kind: "status",
+      stripId: value.stripId,
+      id: value.id,
+    };
+  }
+
+  return null;
+}
+
+function dashboardEventTiles(document: DashboardDocument): DashboardTileReference[] {
+  const status = document.statusStrips.flatMap((strip) =>
+    strip.items.map((item): DashboardTileReference => ({
+      kind: "status",
+      stripId: strip.id,
+      id: item.id,
+    })),
+  );
+  const telemetry = document.telemetry
+    .filter((card) => card.datasource?.type === "external")
+    .map((card): DashboardTileReference => ({ kind: "telemetry", id: card.id }));
+  const modules = (document.modules || [])
+    .filter((module) =>
+      module.datasource?.type === "external" ||
+      module.id === "runtime-health-summary"
+    )
+    .map((module): DashboardTileReference => ({ kind: "module", id: module.id }));
+  const services = document.serviceGroups.flatMap((group) =>
+    group.services
+      .filter((service) => service.datasource?.type === "external")
+      .map((service): DashboardTileReference => ({
+        kind: "service",
+        groupId: group.id,
+        id: service.id,
+      })),
+  );
+
+  return [...status, ...telemetry, ...modules, ...services];
+}
+
+function dashboardTileEventChunk(resolution: DashboardTileResolution): string {
+  return [
+    "event: dashboard-tile",
+    `data: ${JSON.stringify(resolution)}`,
+    "",
+    "",
+  ].join("\n");
+}
+
+class DashboardTileCacheResolutionError extends Error {
+  readonly cache: "hit" | "miss";
+  readonly coalesced: boolean;
+  override readonly cause: unknown;
+
+  constructor(
+    cause: unknown,
+    metadata: Pick<DashboardTileCacheResult, "cache" | "coalesced">,
+  ) {
+    super("Dashboard tile cache resolution failed");
+    this.name = "DashboardTileCacheResolutionError";
+    this.cause = cause;
+    this.cache = metadata.cache;
+    this.coalesced = metadata.coalesced;
+  }
+}
+
+function dashboardTileCacheResolutionError(
+  error: unknown,
+): DashboardTileCacheResolutionFailure | null {
+  if (
+    typeof error === "object" &&
+    error !== null &&
+    "cache" in error &&
+    (error.cache === "hit" || error.cache === "miss") &&
+    "coalesced" in error &&
+    typeof error.coalesced === "boolean" &&
+    "cause" in error
+  ) {
+    return {
+      cache: error.cache,
+      cause: error.cause,
+      coalesced: error.coalesced,
+    };
+  }
+
+  return null;
+}
+
+interface DashboardTileCacheResolutionFailure {
+  cache: "hit" | "miss";
+  cause: unknown;
+  coalesced: boolean;
+}
+
+function dashboardTileTtlMs(
+  document: DashboardDocument,
+  tile: DashboardTileReference,
+): number {
+  if (tile.kind === "telemetry") return 15_000;
+  if (tile.kind === "service") return 30_000;
+
+  if (tile.kind === "module") {
+    const module = document.modules?.find((item) => item.id === tile.id);
+    if (
+      module?.datasource?.type === "external" &&
+      module.datasource.adapter === "weather"
+    ) {
+      return 10 * 60_000;
+    }
+    return 30_000;
+  }
+
+  if (["system-status", "uptime", "load-avg"].includes(tile.id)) {
+    return 30_000;
+  }
+
+  return 5 * 60_000;
+}
+
 function parseDashboardTilePath(pathname: string): DashboardTileReference | null {
   const parts = pathname.split("/").filter(Boolean);
   const [, dashboard, tileRoot, kind, firstId, secondId] = parts;

apps/web/tests/e2e/dashboard.spec.ts

@@ -16,7 +16,6 @@ const linkedServiceIds = [
   "open-webui",
   "comfyui",
   "models",
-  "prompt-registry",
   "adminer",
   "assistant",
   "suna",
@@ -104,7 +103,7 @@ test.describe("dashboard page QA gate", () => {
     expect(metrics.runtimeBottom).toBeLessThanOrEqual(956);
     expect(metrics.footerBottom).toBeLessThanOrEqual(956);
     expect(metrics.telemetryCardCount).toBe(16);
-    expect(metrics.serviceRowCount).toBe(28);
+    expect(metrics.serviceRowCount).toBe(27);
     expect(metrics.footerCellCount).toBe(5);
     expect(metrics.clippedItems).toEqual([]);
   });

scripts/deploy-dimensionlab-website.sh

@@ -0,0 +1,362 @@
+#!/usr/bin/env bash
+set -Eeuo pipefail
+
+APP_NAME="${APP_NAME:-dimensionlab-website}"
+SERVICE_NAME="${SERVICE_NAME:-dimensionlab-website.service}"
+CONTAINER_NAME="${CONTAINER_NAME:-dimensionlab-website}"
+IMAGE_REPO="${IMAGE_REPO:-localhost/dimensionlab-website}"
+CONTAINERFILE="${CONTAINERFILE:-apps/web/Containerfile}"
+PUBLIC_URL="${PUBLIC_URL:-https://dimensionlab.net/}"
+TILE_BATCH_URL="${TILE_BATCH_URL:-https://dimensionlab.net/api/dashboard/tiles}"
+DEPLOY_RESTART_STRATEGY="${DEPLOY_RESTART_STRATEGY:-auto}"
+DEPLOY_SMOKE_TIMEOUT_SECONDS="${DEPLOY_SMOKE_TIMEOUT_SECONDS:-120}"
+DEPLOY_CONTAINER_START_TIMEOUT_SECONDS="${DEPLOY_CONTAINER_START_TIMEOUT_SECONDS:-90}"
+
+dry_run=false
+rollback_tag=""
+release_tag=""
+latest_tag="${IMAGE_REPO}:latest"
+container_cli=""
+deployment_started=false
+rollback_done=false
+rollback_in_progress=false
+
+usage() {
+  cat <<USAGE
+Usage: $0 [--dry-run]
+
+Build and deploy ${APP_NAME} for the production ${SERVICE_NAME} unit.
+
+Options:
+  --dry-run  Print the deployment actions without changing the host.
+USAGE
+}
+
+log() {
+  printf '[deploy:%s] %s\n' "$APP_NAME" "$*"
+}
+
+fail() {
+  printf '[deploy:%s] ERROR: %s\n' "$APP_NAME" "$*" >&2
+  if [ "${deployment_started:-false}" = "true" ] && [ "${rollback_in_progress:-false}" != "true" ]; then
+    rollback || true
+  fi
+  exit 1
+}
+
+run() {
+  if "$dry_run"; then
+    printf '[deploy:%s] DRY-RUN:' "$APP_NAME"
+    printf ' %q' "$@"
+    printf '\n'
+    return 0
+  fi
+
+  "$@"
+}
+
+for arg in "$@"; do
+  case "$arg" in
+    --dry-run)
+      dry_run=true
+      ;;
+    -h|--help)
+      usage
+      exit 0
+      ;;
+    *)
+      usage >&2
+      fail "unknown argument: $arg"
+      ;;
+  esac
+done
+
+deployment_ref() {
+  printf '%s' "${DEPLOY_REF:-${GITHUB_REF:-${FORGEJO_REF:-}}}"
+}
+
+deployment_event() {
+  printf '%s' "${DEPLOY_EVENT_NAME:-${GITHUB_EVENT_NAME:-${FORGEJO_EVENT_NAME:-}}}"
+}
+
+require_main_push() {
+  local event
+  local ref
+
+  event="$(deployment_event)"
+  ref="$(deployment_ref)"
+
+  if [ -n "$event" ] && [ "$event" != "push" ]; then
+    fail "refusing to deploy for event '$event'; production deploys only run for push"
+  fi
+
+  if [ -n "$ref" ]; then
+    [ "$ref" = "refs/heads/main" ] || fail "refusing to deploy ref '$ref'; expected refs/heads/main"
+    return 0
+  fi
+
+  local branch
+  branch="$(git branch --show-current 2>/dev/null || true)"
+  [ "$branch" = "main" ] || fail "refusing to deploy branch '$branch'; expected main"
+}
+
+select_container_cli() {
+  if [ -n "${DEPLOY_CONTAINER_CLI:-}" ]; then
+    command -v "$DEPLOY_CONTAINER_CLI" >/dev/null 2>&1 || fail "container CLI not found: $DEPLOY_CONTAINER_CLI"
+    container_cli="$DEPLOY_CONTAINER_CLI"
+    return 0
+  fi
+
+  if command -v podman >/dev/null 2>&1; then
+    container_cli="podman"
+    return 0
+  fi
+
+  if command -v docker >/dev/null 2>&1; then
+    container_cli="docker"
+    return 0
+  fi
+
+  fail "podman or docker is required"
+}
+
+current_sha() {
+  if [ -n "${DEPLOY_SHA:-${GITHUB_SHA:-}}" ]; then
+    printf '%s' "${DEPLOY_SHA:-${GITHUB_SHA:-}}"
+    return 0
+  fi
+
+  git rev-parse HEAD
+}
+
+tag_existing_latest_for_rollback() {
+  rollback_tag="${IMAGE_REPO}:rollback-$(date -u +%Y%m%d%H%M%S)"
+  if "$container_cli" image inspect "$latest_tag" >/dev/null 2>&1; then
+    log "tagging current latest image as $rollback_tag"
+    run "$container_cli" tag "$latest_tag" "$rollback_tag"
+  else
+    log "no existing $latest_tag image found; rollback image tag will not be created"
+    rollback_tag=""
+  fi
+}
+
+container_systemd_unit() {
+  "$container_cli" inspect "$CONTAINER_NAME" \
+    --format '{{ index .Config.Labels "PODMAN_SYSTEMD_UNIT" }}' 2>/dev/null || true
+}
+
+require_container_managed_by_service() {
+  local unit
+
+  if "$dry_run"; then
+    log "DRY-RUN: would require $CONTAINER_NAME to be managed by $SERVICE_NAME"
+    return 0
+  fi
+
+  unit="$(container_systemd_unit)"
+  [ "$unit" = "$SERVICE_NAME" ] || fail "refusing to stop $CONTAINER_NAME; expected PODMAN_SYSTEMD_UNIT=$SERVICE_NAME, got '${unit:-unset}'"
+}
+
+validate_restart_strategy() {
+  case "$DEPLOY_RESTART_STRATEGY" in
+    systemctl)
+      if ! "$dry_run" && ! systemctl --user show "$SERVICE_NAME" >/dev/null 2>&1; then
+        fail "systemctl --user cannot access $SERVICE_NAME"
+      fi
+      ;;
+    quadlet-container|kill-container)
+      require_container_managed_by_service
+      ;;
+    auto)
+      if "$dry_run"; then
+        log "DRY-RUN: would validate automatic restart strategy"
+      elif ! command -v systemctl >/dev/null 2>&1 || ! systemctl --user show "$SERVICE_NAME" >/dev/null 2>&1; then
+        require_container_managed_by_service
+      fi
+      ;;
+    *)
+      fail "unknown DEPLOY_RESTART_STRATEGY: $DEPLOY_RESTART_STRATEGY"
+      ;;
+  esac
+}
+
+initialize_submodules() {
+  log "initializing submodules"
+  run git config --global url."https://git.dimensionlab.net/".insteadOf "ssh://git@git.dimensionlab.net/"
+  run git submodule update --init --recursive
+}
+
+build_image() {
+  local sha
+  local short_sha
+
+  sha="$(current_sha)"
+  short_sha="${sha:0:12}"
+  release_tag="${IMAGE_REPO}:${short_sha}"
+
+  [ -f "$CONTAINERFILE" ] || fail "containerfile not found: $CONTAINERFILE"
+
+  log "building $release_tag and $latest_tag from $CONTAINERFILE"
+  run "$container_cli" build -f "$CONTAINERFILE" -t "$release_tag" -t "$latest_tag" .
+}
+
+restart_service() {
+  log "restarting $SERVICE_NAME with strategy $DEPLOY_RESTART_STRATEGY"
+
+  case "$DEPLOY_RESTART_STRATEGY" in
+    systemctl)
+      run systemctl --user restart "$SERVICE_NAME"
+      ;;
+    quadlet-container|kill-container)
+      require_container_managed_by_service
+      run "$container_cli" stop "$CONTAINER_NAME"
+      ;;
+    auto)
+      if command -v systemctl >/dev/null 2>&1 && systemctl --user is-active "$SERVICE_NAME" >/dev/null 2>&1; then
+        run systemctl --user restart "$SERVICE_NAME"
+      else
+        require_container_managed_by_service
+        run "$container_cli" stop "$CONTAINER_NAME"
+      fi
+      ;;
+    *)
+      fail "unknown DEPLOY_RESTART_STRATEGY: $DEPLOY_RESTART_STRATEGY"
+      ;;
+  esac
+}
+
+latest_image_id() {
+  "$container_cli" image inspect "$latest_tag" --format '{{.Id}}' 2>/dev/null || true
+}
+
+container_image_id() {
+  "$container_cli" inspect "$CONTAINER_NAME" --format '{{.Image}}' 2>/dev/null || true
+}
+
+container_running() {
+  local running
+  running="$("$container_cli" inspect "$CONTAINER_NAME" --format '{{.State.Running}}' 2>/dev/null || true)"
+  [ "$running" = "true" ]
+}
+
+wait_for_container_restart() {
+  local expected_image
+
+  if "$dry_run"; then
+    log "DRY-RUN: would wait for $CONTAINER_NAME to run $latest_tag"
+    return 0
+  fi
+
+  expected_image="$(latest_image_id)"
+  [ -n "$expected_image" ] || fail "could not resolve image id for $latest_tag"
+
+  wait_for_container_image "$expected_image" "new image" || fail "$CONTAINER_NAME did not restart on $latest_tag within ${DEPLOY_CONTAINER_START_TIMEOUT_SECONDS}s"
+}
+
+wait_for_container_image() {
+  local expected_image="$1"
+  local label="$2"
+  local deadline
+
+  deadline=$((SECONDS + DEPLOY_CONTAINER_START_TIMEOUT_SECONDS))
+
+  while [ "$SECONDS" -lt "$deadline" ]; do
+    if container_running && [ "$(container_image_id)" = "$expected_image" ]; then
+      log "$CONTAINER_NAME is running the $label"
+      return 0
+    fi
+
+    sleep 2
+  done
+
+  return 1
+}
+
+smoke_get() {
+  local url="$1"
+  curl -fsS --max-time 10 -o /dev/null "$url"
+}
+
+smoke_tiles() {
+  local response
+
+  response="$(
+    curl -fsS --max-time 20 \
+      -H "content-type: application/json" \
+      --data '{"tiles":[{"kind":"status","stripId":"footer-status","id":"system-status"}]}' \
+      "$TILE_BATCH_URL"
+  )"
+
+  [[ "$response" == *'"state":"ready"'* ]]
+}
+
+wait_for_smoke() {
+  local deadline
+
+  if "$dry_run"; then
+    log "DRY-RUN: would smoke check $PUBLIC_URL and $TILE_BATCH_URL"
+    return 0
+  fi
+
+  deadline=$((SECONDS + DEPLOY_SMOKE_TIMEOUT_SECONDS))
+  until smoke_get "$PUBLIC_URL" && smoke_tiles; do
+    if [ "$SECONDS" -ge "$deadline" ]; then
+      fail "smoke checks failed for $PUBLIC_URL and $TILE_BATCH_URL"
+    fi
+
+    sleep 3
+  done
+
+  log "smoke checks passed"
+}
+
+rollback() {
+  local rollback_image
+
+  if [ "$deployment_started" != "true" ] || [ -z "$rollback_tag" ] || [ "$rollback_done" = "true" ]; then
+    return 0
+  fi
+
+  rollback_done=true
+  rollback_in_progress=true
+  printf '[deploy:%s] rolling back to %s\n' "$APP_NAME" "$rollback_tag" >&2
+  rollback_image="$("$container_cli" image inspect "$rollback_tag" --format '{{.Id}}' 2>/dev/null || true)"
+  "$container_cli" tag "$rollback_tag" "$latest_tag" || true
+  if container_running; then
+    restart_service || true
+  else
+    printf '[deploy:%s] waiting for %s to recover with rollback image\n' "$APP_NAME" "$SERVICE_NAME" >&2
+  fi
+  if [ -n "$rollback_image" ] && wait_for_container_image "$rollback_image" "rollback image"; then
+    printf '[deploy:%s] rollback image is running\n' "$APP_NAME" >&2
+  else
+    printf '[deploy:%s] ERROR: rollback image did not become healthy\n' "$APP_NAME" >&2
+  fi
+  rollback_in_progress=false
+}
+
+on_error() {
+  local status=$?
+  rollback
+  exit "$status"
+}
+
+trap on_error ERR
+
+require_main_push
+select_container_cli
+validate_restart_strategy
+
+log "using container CLI: $container_cli"
+log "target image: $latest_tag"
+log "target service: $SERVICE_NAME"
+
+initialize_submodules
+tag_existing_latest_for_rollback
+build_image
+deployment_started=true
+restart_service
+wait_for_container_restart
+wait_for_smoke
+
+log "deployment finished"