Merge pull request #435 from StroepWafel/improved-global-search

Improved global search

package.json

@@ -84,7 +84,7 @@
     "codemirror": "^6.0.1",
     "color": "^5.0.0",
     "dompurify": "^3.2.4",
-    "embeddia": "^1.2.1",
+    "embeddia": "^1.3.0",
     "embla-carousel-autoplay": "^8.5.2",
     "embla-carousel-svelte": "^8.5.2",
     "esbuild": "^0.25.3",

src/background.ts

@@ -1,4 +1,5 @@
 import browser from "webextension-polyfill";
+import semver from "semver";
 import type { SettingsState } from "@/types/storage";
 import { fetchNews } from "./background/news";
 import {
@@ -506,6 +507,50 @@ function SetStorageValue(object: any) {
   }
 }
 
+/** One-time migration for 3.6.5: opt upgraders into Global Search + indexing + transparency defaults. */
+const GLOBAL_SEARCH_PLUGIN_SETTINGS_KEY = "plugin.global-search.settings";
+const GLOBAL_SEARCH_MIGRATION_VERSION = "3.6.5";
+
+async function migrateGlobalSearchDefaultsFor365Upgrade(
+  previousVersion: string,
+): Promise<void> {
+  try {
+    const currRaw = browser.runtime.getManifest().version;
+    const prev = semver.coerce(previousVersion);
+    const curr = semver.coerce(currRaw);
+    if (
+      prev == null ||
+      curr == null ||
+      semver.lt(curr, GLOBAL_SEARCH_MIGRATION_VERSION) ||
+      !semver.lt(prev, GLOBAL_SEARCH_MIGRATION_VERSION)
+    ) {
+      return;
+    }
+
+    const got = await browser.storage.local.get(GLOBAL_SEARCH_PLUGIN_SETTINGS_KEY);
+    const existing = (got[GLOBAL_SEARCH_PLUGIN_SETTINGS_KEY] ?? {}) as Record<
+      string,
+      unknown
+    >;
+
+    await browser.storage.local.set({
+      [GLOBAL_SEARCH_PLUGIN_SETTINGS_KEY]: {
+        ...existing,
+        enabled: true,
+        transparencyEffects: true,
+        runIndexingOnLoad: true,
+        passiveIndexing: true,
+      },
+    });
+
+    console.info(
+      `[BetterSEQTA+] Migration ${GLOBAL_SEARCH_MIGRATION_VERSION}: Global Search and related settings enabled (from ${previousVersion}).`,
+    );
+  } catch (e) {
+    console.warn("[BetterSEQTA+] Global Search 3.6.5 settings migration failed:", e);
+  }
+}
+
 browser.runtime.onInstalled.addListener(function (event) {
   browser.storage.local.remove(["justupdated"]);
   browser.storage.local.remove(["data"]);
@@ -513,6 +558,10 @@ browser.runtime.onInstalled.addListener(function (event) {
   if (event.reason == "install" || event.reason == "update") {
     browser.storage.local.set({ justupdated: true });
   }
+
+  if (event.reason === "update" && event.previousVersion) {
+    void migrateGlobalSearchDefaultsFor365Upgrade(event.previousVersion);
+  }
 });
 
 initCloudSettingsAutoSync({ reloadSeqtaPages });

src/plugins/built-in/globalSearch/lazy.ts

@@ -7,6 +7,7 @@ import {
 } from "../../core/settingsHelpers";
 import { isSeqtaEngageExperience } from "@/seqta/utils/isSeqtaEngage";
 import styles from "./src/core/styles.css?inline";
+import { resetSearchIndexes } from "./src/indexing/resetIndexes";
 
 // Platform-aware default hotkey
 const getDefaultHotkey = () => {
@@ -35,78 +36,35 @@ const settings = defineSettings({
     title: "Index on Page Load",
     description: "Run content indexing when SEQTA loads",
   }),
+  passiveIndexing: booleanSetting({
+    default: true,
+    title: "Index Browsed Content",
+    description:
+      "Capture safe text from SEQTA pages you visit so they're searchable. Sensitive routes (settings, files, login) are always excluded.",
+  }),
   resetIndex: buttonSetting({
     title: "Reset Index",
     description: "Reset the search index and storage",
     trigger: async () => {
-      const confirmed = confirm("Are you sure you want to reset the search index and storage?");
+      const confirmed = confirm(
-
+        "Reset the search index and all stored Global Search data?\n\nAfter this, reload this SEQTA tab so indexing can run again and rebuild the index.",
-      if (confirmed) {
+      );
-        try {
+      if (!confirmed) return;
-          // Dynamically import modules to avoid loading heavy dependencies
-          const { VectorWorkerManager } = await import("./src/indexing/worker/vectorWorkerManager");
-          const { resetDatabase } = await import("./src/indexing/db");
-          
-          // Reset vector worker first
-          try {
-            const workerManager = VectorWorkerManager.getInstance();
-            await workerManager.resetWorker();
-            console.log("Vector worker reset successfully");
-          } catch (e) {
-            console.warn("Failed to reset vector worker:", e);
-          }
-
-          // Close all database connections properly before deletion
-          try {
-            await resetDatabase();
-            console.log("betterseqta-index database closed and reset");
-          } catch (e) {
-            console.warn("Failed to reset betterseqta-index database:", e);
-          }
-
-          // Wait a bit for connections to fully close
-          await new Promise(resolve => setTimeout(resolve, 100));
-
-          // Delete embeddiaDB (vector search database)
-          const deleteDb = (dbName: string) => {
-            return new Promise<void>((resolve, reject) => {
-              const req = indexedDB.deleteDatabase(dbName);
-              req.onsuccess = () => {
-                console.log(`Successfully deleted database: ${dbName}`);
-                resolve();
-              };
-              req.onerror = () => {
-                console.error(`Error deleting database ${dbName}:`, req.error);
-                reject(req.error);
-              };
-              req.onblocked = () => {
-                console.warn(`Database ${dbName} deletion blocked - connections still open`);
-                // Wait and retry once
-                setTimeout(() => {
-                  const retryReq = indexedDB.deleteDatabase(dbName);
-                  retryReq.onsuccess = () => {
-                    console.log(`Successfully deleted database on retry: ${dbName}`);
-                    resolve();
-                  };
-                  retryReq.onerror = () => reject(retryReq.error);
-                  retryReq.onblocked = () => {
-                    reject(new Error(`One database is open, failed to remove: ${dbName}. Please close other tabs and try again.`));
-                  };
-                }, 500);
-              };
-            });
-          };
 
       try {
-            await deleteDb("embeddiaDB");
+        // `resetSearchIndexes` is a tiny statically-imported helper: no
-            await deleteDb("betterseqta-index");
+        // dynamic chunks to chase, so the button keeps working even when
-            alert("Search index and storage have been reset successfully.");
+        // the settings page has been open across an extension update.
+        await resetSearchIndexes();
+        alert(
+          "Search index and storage were reset.\n\nReload this tab to regenerate the index.",
+        );
       } catch (e) {
-            alert("Failed to reset one or more databases: " + String(e) + "\n\nTry closing other browser tabs and try again.");
+        alert(
-          }
+          "Failed to reset index: " +
-        } catch (e) {
+            String(e) +
-          alert("Failed to reset index: " + String(e));
+            "\n\nTry closing other browser tabs and try again.",
-        }
+        );
       }
     },
   }),

src/plugins/built-in/globalSearch/src/components/SearchBar.svelte

@@ -48,6 +48,13 @@
   let calculatorResult = $state<string | null>(null);
   let resultsList = $state<HTMLUListElement>();
 
+  // Monotonic counter so a slow async search (vector reranking) cannot
+  // overwrite results from a newer keystroke. Without this guard, the user
+  // observes results "flickering" — e.g. typing `world w` finds the assessment
+  // but `world wa` triggers a new search whose vector pass returns later than
+  // the `world w` pass and clobbers the more relevant matches.
+  let searchRequestId = 0;
+
   const updateCalculatorState = (hasResult: string | null) => {
     calculatorResult = hasResult;
   };
@@ -166,9 +173,10 @@
     });
 
     const term = searchTerm.trim().toLowerCase();
+    const requestId = ++searchRequestId;
 
     if (commandsFuse && dynamicContentFuse) {
-      combinedResults = await doSearch(
+      const results = await doSearch(
         term,
         commandsFuse,
         commandIdToItemMap,
@@ -176,7 +184,16 @@
         dynamicIdToItemMap,
         true, // sortByRecent
       );
+
+      // Drop the result if the user has typed since this search started, or
+      // if the current term no longer matches what we searched for. This
+      // keeps the visible list anchored to the latest query.
+      if (requestId !== searchRequestId) return;
+      if (searchTerm.trim().toLowerCase() !== term) return;
+
+      combinedResults = results;
     } else {
+      if (requestId !== searchRequestId) return;
       combinedResults = [];
     }
 

src/plugins/built-in/globalSearch/src/components/items/GenericItem.svelte

@@ -0,0 +1,89 @@
+<script lang="ts">
+  import HighlightedText from '../../utils/HighlightedText.svelte';
+  import type { DynamicContentItem } from '../../utils/dynamicItems';
+  import type { FuseResultMatch } from '../../core/types';
+
+  const { item, isSelected, searchTerm, matches, onclick } = $props<{
+    item: DynamicContentItem;
+    isSelected: boolean;
+    searchTerm: string;
+    matches?: readonly FuseResultMatch[];
+    onclick: () => void;
+  }>();
+
+  const categoryLabel = (category: string): string => {
+    if (!category) return '';
+    return category.charAt(0).toUpperCase() + category.slice(1);
+  };
+
+  const gradientForCategory = (category: string): string => {
+    switch (category) {
+      case 'courses':
+        return 'from-[#7c5fe0] to-[#4d2bb8]';
+      case 'notices':
+        return 'from-[#f6c453] to-[#d39007]';
+      case 'documents':
+        return 'from-[#4FBBFE] to-[#2090F3]';
+      case 'folio':
+        return 'from-[#22c55e] to-[#0f9b3a]';
+      case 'portals':
+        return 'from-[#22d3ee] to-[#0e7490]';
+      case 'reports':
+        return 'from-[#f97316] to-[#c2410c]';
+      case 'goals':
+        return 'from-[#10b981] to-[#047857]';
+      case 'passive':
+        return 'from-[#6b7280] to-[#374151]';
+      default:
+        return 'from-[#4FBBFE] to-[#2090F3]';
+    }
+  };
+
+  const fallbackIcon = (category: string): string => {
+    switch (category) {
+      case 'courses':
+        return '\ueb4d';
+      case 'notices':
+        return '\ueb24';
+      case 'documents':
+        return '\ueb6f';
+      case 'folio':
+        return '\ueb16';
+      case 'portals':
+        return '\ueb01';
+      case 'reports':
+        return '\ueb70';
+      case 'goals':
+        return '\uea15';
+      case 'passive':
+        return '\ueb71';
+      default:
+        return '\ue924';
+    }
+  };
+</script>
+
+<button
+  class="w-full flex flex-col px-2 py-1.5 rounded-lg select-none cursor-pointer group transition-colors duration-100 ring-0 dark:ring-zinc-600/50
+  {isSelected ? 'bg-zinc-900/5 dark:bg-white/10 text-zinc-900 dark:text-white dark:ring-[1px] dark:shadow' : 'hover:bg-zinc-500/5 dark:hover:bg-white/5 text-zinc-800 dark:text-zinc-200'}"
+  onclick={onclick}
+>
+  <div class="flex items-center w-full">
+    <div
+      class="flex-none scale-90 w-8 h-8 text-xl font-IconFamily flex items-center justify-center text-white rounded-md bg-gradient-to-br {gradientForCategory(item.category)}"
+    >
+      {item.metadata?.icon || fallbackIcon(item.category)}
+    </div>
+    <span class="ml-4 text-lg truncate">
+      <HighlightedText text={item.text} term={searchTerm} matches={matches} />
+    </span>
+    <span class="flex-none ml-auto text-xs text-zinc-500 dark:text-zinc-400">
+      {item.metadata?.subjectCode || categoryLabel(item.category)}
+    </span>
+  </div>
+  {#if item.content}
+    <div class="mt-1 ml-12 text-sm text-zinc-600 dark:text-zinc-400 line-clamp-2 text-start">
+      <HighlightedText text={item.content} term={searchTerm} matches={matches} />
+    </div>
+  {/if}
+</button>

src/plugins/built-in/globalSearch/src/core/commands.ts

@@ -25,11 +25,11 @@ async function getCurrentLesson() {
   try {
     const response = await fetch(`${location.origin}/seqta/student/load/timetable?`, {
       method: "POST",
+      credentials: "include",
       headers: { "Content-Type": "application/json" },
       body: JSON.stringify({
         from: todayFormatted,
         until: todayFormatted,
-        student: 69,
       }),
     });
     

src/plugins/built-in/globalSearch/src/core/index.ts

@@ -15,6 +15,10 @@ import { cleanupSearchBar, mountSearchBar } from "./mountSearchBar";
 import { IndexedDbManager } from "embeddia";
 import { VectorWorkerManager } from "../indexing/worker/vectorWorkerManager";
 import { checkAndHandleUpdate } from "../utils/versionCheck";
+import {
+  getStoredPassiveItems,
+  installPassiveObserver,
+} from "../indexing/passiveObserver";
 
 // Platform-aware default hotkey
 const getDefaultHotkey = () => {
@@ -43,11 +47,19 @@ const settings = defineSettings({
     title: "Index on Page Load",
     description: "Run content indexing when SEQTA loads",
   }),
+  passiveIndexing: booleanSetting({
+    default: true,
+    title: "Index Browsed Content",
+    description:
+      "Capture safe text from SEQTA pages you visit so they're searchable. Sensitive routes (settings, files, login) are always excluded.",
+  }),
   resetIndex: buttonSetting({
     title: "Reset Index",
     description: "Reset the search index and storage",
     trigger: async () => {
-      const confirmed = confirm("Are you sure you want to reset the search index and storage?");
+      const confirmed = confirm(
+        "Reset the search index and all stored Global Search data?\n\nAfter this, reload this SEQTA tab so indexing can run again and rebuild the index.",
+      );
 
       if (confirmed) {
         try {
@@ -106,7 +118,9 @@ const settings = defineSettings({
           try {
             await deleteDb("embeddiaDB");
             await deleteDb("betterseqta-index");
-            alert("Search index and storage have been reset successfully.");
+            alert(
+              "Search index and storage were reset.\n\nReload this tab to regenerate the index.",
+            );
           } catch (e) {
             alert("Failed to reset one or more databases: " + String(e) + "\n\nTry closing other browser tabs and try again.");
           }
@@ -131,6 +145,9 @@ class GlobalSearchPlugin extends BasePlugin<typeof settings> {
   @Setting(settings.runIndexingOnLoad)
   runIndexingOnLoad!: boolean;
 
+  @Setting(settings.passiveIndexing)
+  passiveIndexing!: boolean;
+
   @Setting(settings.resetIndex)
   resetIndex!: () => void;
 }
@@ -150,26 +167,35 @@ const globalSearchPlugin: Plugin<typeof settings> = {
   run: async (api) => {
     const appRef = { current: null };
 
-    // Check for extension updates and clear caches if needed
+    // Run the version check BEFORE we open any IndexedDB connections.
-    // Use a timeout to avoid blocking initialization
+    // On a normal load (no version change) this is just a string compare
-    setTimeout(async () => {
+    // and a manifest read, so the cost is negligible. On a real update,
+    // we want the database wipe to complete before `IndexedDbManager`
+    // grabs a handle on `embeddiaDB`, otherwise the delete request comes
+    // back blocked.
     try {
       const wasUpdated = await checkAndHandleUpdate();
       if (wasUpdated) {
-          console.log("[Global Search] Extension updated - caches cleared");
+        console.log(
+          "[Global Search] Extension updated — search index reset; the next indexing pass will repopulate.",
+        );
       }
     } catch (error: any) {
-        // Handle CSS preload errors and other failures gracefully
+      // Firefox sometimes refuses CSS preloads or asset reads; we never
-        // These can happen in Firefox or when assets aren't available
+      // want this path to take the whole plugin down.
-        if (error?.message?.includes("preload CSS") || 
+      if (
+        error?.message?.includes("preload CSS") ||
         error?.message?.includes("MIME type") ||
-            error?.message?.includes("NS_ERROR_CORRUPTED_CONTENT")) {
+        error?.message?.includes("NS_ERROR_CORRUPTED_CONTENT")
-          console.debug("[Global Search] Version check skipped due to asset loading restrictions:", error.message);
+      ) {
+        console.debug(
+          "[Global Search] Version check skipped due to asset loading restrictions:",
+          error.message,
+        );
       } else {
         console.warn("[Global Search] Failed to check for updates:", error);
       }
     }
-    }, 100);
 
     try {
       await IndexedDbManager.create("embeddiaDB", "embeddiaObjectStore", {
@@ -210,6 +236,17 @@ const globalSearchPlugin: Plugin<typeof settings> = {
         const workerManager = VectorWorkerManager.getInstance();
         console.log("Streaming active:", workerManager.isStreamingActive());
       },
+      passiveItems: async () => {
+        const items = await getStoredPassiveItems();
+        console.log(`Captured ${items.length} passive items`);
+        return items;
+      },
+      runSelfTests: async () => {
+        const { runGlobalSearchSelfTests } = await import(
+          "../indexing/selfTests"
+        );
+        return runGlobalSearchSelfTests();
+      },
       checkIndexedDBSize: async () => {
         try {
           const estimate = await navigator.storage.estimate();
@@ -232,6 +269,14 @@ const globalSearchPlugin: Plugin<typeof settings> = {
       }
     };
 
+    if (api.settings.passiveIndexing) {
+      try {
+        installPassiveObserver();
+      } catch (error) {
+        console.warn("[Global Search] Passive observer install failed:", error);
+      }
+    }
+
     if (api.settings.runIndexingOnLoad) {
       setTimeout(async () => {
         await runIndexing();

src/plugins/built-in/globalSearch/src/core/mountSearchBar.ts

@@ -8,7 +8,12 @@ import browser from "webextension-polyfill";
 export function mountSearchBar(
   titleElement: Element,
   api: any,
-  appRef: { current: any; storageChangeHandler?: any; progressHandler?: any },
+  appRef: {
+    current: any;
+    storageChangeHandler?: any;
+    progressHandler?: any;
+    clearDoneFlashTimer?: () => void;
+  },
 ) {
   if (titleElement.querySelector(".search-trigger")) {
     return;
@@ -18,74 +23,215 @@ export function mountSearchBar(
   let currentHotkey = isValidHotkey(api.settings.searchHotkey) ? api.settings.searchHotkey : "ctrl+k";
   let hotkeyDisplay = formatHotkeyForDisplay(currentHotkey);
 
+  // Search trigger + progress UI live in one wrapper so the auto-margin
+  // pushes the whole group to the left edge of the topbar instead of
+  // stranding the progress text on the far right of the screen.
+  const searchWrapper = document.createElement("div");
+  searchWrapper.className = "search-trigger-wrapper";
+
+  // Anchor stacks button + slim progress strip in one rounded chip (see
+  // `.search-trigger-anchor` in styles.css).
+  const searchAnchor = document.createElement("div");
+  searchAnchor.className = "search-trigger-anchor";
+
   const searchButton = document.createElement("div");
   searchButton.className = "search-trigger";
 
-  // Create progress indicator container
-  const progressContainer = document.createElement("div");
-  progressContainer.className = "search-progress-container";
-  progressContainer.style.cssText = "display: flex; align-items: center; gap: 8px; margin-left: 8px; min-width: 120px;";
-  
-  // Create progress bar
   const progressBarWrapper = document.createElement("div");
   progressBarWrapper.className = "search-progress-bar-wrapper";
-  progressBarWrapper.style.cssText = "flex: 1; height: 4px; background: rgba(0, 0, 0, 0.1); border-radius: 2px; overflow: hidden; display: none;";
+
+  const progressTrack = document.createElement("div");
+  progressTrack.className = "search-progress-track";
 
   const progressBar = document.createElement("div");
   progressBar.className = "search-progress-bar";
-  progressBar.style.cssText = "height: 100%; background: linear-gradient(90deg, #3b82f6, #2563eb, #3b82f6); transition: width 0.3s ease-out; width: 0%; position: relative;";
+  progressTrack.appendChild(progressBar);
+  progressBarWrapper.appendChild(progressTrack);
 
-  // Add shimmer effect
+  // Use a block-level <div> so the label reliably participates in flex
-  const shimmer = document.createElement("div");
+  // layout. A <span> defaults to `display: inline`, which silently ignores
-  shimmer.style.cssText = "position: absolute; inset: 0; background: linear-gradient(90deg, transparent, rgba(255,255,255,0.3), transparent); animation: shimmer 2s infinite;";
+  // `max-width`, `overflow`, and `text-overflow: ellipsis`, and was the
-  progressBar.appendChild(shimmer);
+  // reason the label appeared blank when the bar was visible.
-  progressBarWrapper.appendChild(progressBar);
+  const progressText = document.createElement("div");
-  
-  // Create progress text
-  const progressText = document.createElement("span");
   progressText.className = "search-progress-text";
-  progressText.style.cssText = "font-size: 11px; color: #666; white-space: nowrap; display: none;";
+  progressText.setAttribute("aria-live", "polite");
 
-  progressContainer.appendChild(progressBarWrapper);
+  searchAnchor.appendChild(searchButton);
-  progressContainer.appendChild(progressText);
+  searchAnchor.appendChild(progressBarWrapper);
+  searchWrapper.appendChild(searchAnchor);
+  searchWrapper.appendChild(progressText);
 
   // Indexing state
   let isIndexing = false;
+  /** True while indexing has run until it finishes/fails — used for Done! flash only */
+  let ranIndexingCycle = false;
   let completedJobs = 0;
   let totalJobs = 0;
   let indexingStatus: string | null = null;
+  let doneFlashTimer: ReturnType<typeof setTimeout> | null = null;
+  let doneFadeTimer: ReturnType<typeof setTimeout> | null = null;
+  /** Captures `wasIndexing && !indexing` for the current dispatcher tick */
+  let indexingJustStoppedFlag = false;
+
+  const DONE_HOLD_MS = 5000;
+  const DONE_FADE_MS = 550;
+
+  /** Treat as failure copy — plain “Done!” would be misleading */
+  const statusLooksRough = (s: string) =>
+    /\b(fail|error|cancel)\b/i.test(s);
+
+  const truncateStatus = (s: string, max = 44) =>
+    s.length > max ? s.slice(0, max - 1) + "…" : s;
+
+  const clearDoneFlashTimer = () => {
+    if (doneFlashTimer) {
+      clearTimeout(doneFlashTimer);
+      doneFlashTimer = null;
+    }
+    if (doneFadeTimer) {
+      clearTimeout(doneFadeTimer);
+      doneFadeTimer = null;
+    }
+  };
 
   const updateProgressDisplay = () => {
-    if (isIndexing && totalJobs > 0) {
+    const indexingStoppedThisTick = indexingJustStoppedFlag;
+    indexingJustStoppedFlag = false;
+
+    const active = isIndexing && totalJobs > 0;
+
+    // Stray pulses (missing total, 0 completed, etc.) used to hit the idle
+    // branch and call clearDoneFlashTimer(), killing the Done! hold/fade.
+    if (doneFlashTimer !== null || doneFadeTimer !== null) {
+      if (!active) {
+        return;
+      }
+      clearDoneFlashTimer();
+    }
+
+    const completionEligible =
+      ranIndexingCycle &&
+      !active &&
+      totalJobs > 0 &&
+      (completedJobs >= totalJobs || indexingStoppedThisTick);
+
+    if (active) {
+      clearDoneFlashTimer();
+      progressBarWrapper.classList.remove("is-rough-complete");
+      progressText.classList.remove(
+        "is-rough",
+        "is-fading-done",
+        "is-done-message",
+      );
       const percentage = Math.round((completedJobs / totalJobs) * 100);
       progressBar.style.width = `${Math.max(2, percentage)}%`;
-      progressBarWrapper.style.display = "block";
+      progressBarWrapper.classList.add("is-active");
+      searchAnchor.classList.add("is-indexing");
+      searchButton.classList.add("is-indexing");
 
       if (indexingStatus) {
-        progressText.textContent = indexingStatus.length > 20 ? indexingStatus.substring(0, 20) + "..." : indexingStatus;
+        progressText.textContent = `${truncateStatus(indexingStatus)} · ${percentage}%`;
-        progressText.style.display = "block";
       } else {
-        progressText.textContent = `${completedJobs}/${totalJobs} (${percentage}%)`;
+        progressText.textContent = `Indexing ${completedJobs}/${totalJobs} (${percentage}%)`;
-        progressText.style.display = "block";
       }
+      progressText.classList.add("is-active");
+      return;
+    }
+
+    if (completionEligible) {
+      // Duplicate end-of-run ticks must not reschedule hold/fade timers
+      if (doneFlashTimer !== null || doneFadeTimer !== null) {
+        return;
+      }
+
+      const rough =
+        indexingStatus != null && statusLooksRough(indexingStatus);
+
+      progressBar.style.width = "0%";
+      progressBarWrapper.classList.remove("is-active");
+      searchAnchor.classList.remove("is-indexing");
+      searchButton.classList.remove("is-indexing");
+      progressText.classList.remove("is-fading-done");
+
+      progressText.textContent = rough ? truncateStatus(indexingStatus!, 52) : "Done!";
+      if (rough) {
+        progressText.classList.add("is-rough");
+        progressBarWrapper.classList.add("is-rough-complete");
       } else {
-      progressBarWrapper.style.display = "none";
+        progressText.classList.remove("is-rough");
-      progressText.style.display = "none";
+        progressBarWrapper.classList.remove("is-rough-complete");
       }
+      progressText.classList.add("is-active", "is-done-message");
+
+      doneFlashTimer = setTimeout(() => {
+        doneFlashTimer = null;
+        progressText.classList.add("is-fading-done");
+        doneFadeTimer = setTimeout(() => {
+          doneFadeTimer = null;
+          ranIndexingCycle = false;
+          indexingStatus = null;
+          progressBar.style.width = "0%";
+          progressBarWrapper.classList.remove("is-active");
+          progressBarWrapper.classList.remove("is-rough-complete");
+          searchAnchor.classList.remove("is-indexing");
+          searchButton.classList.remove("is-indexing");
+          progressText.classList.remove(
+            "is-active",
+            "is-rough",
+            "is-fading-done",
+            "is-done-message",
+          );
+          progressText.textContent = "";
+        }, DONE_FADE_MS);
+      }, DONE_HOLD_MS);
+      return;
+    }
+
+    clearDoneFlashTimer();
+    progressBarWrapper.classList.remove("is-active");
+    progressBarWrapper.classList.remove("is-rough-complete");
+    searchAnchor.classList.remove("is-indexing");
+    searchButton.classList.remove("is-indexing");
+    progressText.classList.remove(
+      "is-active",
+      "is-rough",
+      "is-fading-done",
+      "is-done-message",
+    );
+    progressBar.style.width = "0%";
+    progressText.textContent = "";
+    ranIndexingCycle = false;
+    indexingStatus = null;
   };
 
   // Listen for indexing progress events
   const progressHandler = (event: CustomEvent) => {
-    const { completed, total, indexing, status } = event.detail;
+    const { completed, total, indexing, status } = event.detail as {
-    completedJobs = completed || 0;
+      completed?: number;
-    totalJobs = total || 0;
+      total?: number;
-    isIndexing = indexing || false;
+      indexing?: boolean;
-    indexingStatus = status || null;
+      status?: string;
+    };
+    const wasIndexing = isIndexing;
+
+    completedJobs = completed ?? 0;
+    totalJobs = total ?? 0;
+    isIndexing = Boolean(indexing);
+    indexingStatus = status ?? null;
+    indexingJustStoppedFlag = wasIndexing && !isIndexing;
+
+    if (!wasIndexing && isIndexing) ranIndexingCycle = true;
+    if (wasIndexing && !isIndexing) ranIndexingCycle = true;
+    if (totalJobs > 0 && completedJobs >= totalJobs && !isIndexing) {
+      ranIndexingCycle = true;
+    }
+
     updateProgressDisplay();
   };
 
   window.addEventListener('indexing-progress', progressHandler as EventListener);
   appRef.progressHandler = progressHandler;
+  appRef.clearDoneFlashTimer = clearDoneFlashTimer;
   
   const updateSearchButtonDisplay = () => {
     searchButton.innerHTML = /* html */ `
@@ -99,8 +245,7 @@ export function mountSearchBar(
   };
 
   updateSearchButtonDisplay();
-  titleElement.appendChild(searchButton);
+  titleElement.appendChild(searchWrapper);
-  titleElement.appendChild(progressContainer);
 
   // Listen for hotkey setting changes
   const handleStorageChange = (changes: any, area: string) => {
@@ -139,7 +284,12 @@ export function mountSearchBar(
   }
 }
 
-export function cleanupSearchBar(appRef: { current: any; storageChangeHandler?: any; progressHandler?: any }) {
+export function cleanupSearchBar(appRef: {
+  current: any;
+  storageChangeHandler?: any;
+  progressHandler?: any;
+  clearDoneFlashTimer?: () => void;
+}) {
   if (appRef.current) {
     try {
       unmount(appRef.current);
@@ -149,23 +299,29 @@ export function cleanupSearchBar(appRef: { current: any; storageChangeHandler?:
     }
   }
 
+  try {
+    appRef.clearDoneFlashTimer?.();
+  } catch {
+    /* ignore */
+  }
+  appRef.clearDoneFlashTimer = undefined;
+
   // Remove progress event listener
   if (appRef.progressHandler) {
     window.removeEventListener('indexing-progress', appRef.progressHandler as EventListener);
     appRef.progressHandler = null;
   }
 
-  // Remove search trigger button
+  // Remove search trigger wrapper (which contains the button and progress UI)
-  const searchTrigger = document.querySelector(".search-trigger");
+  const searchWrapper = document.querySelector(".search-trigger-wrapper");
-  if (searchTrigger) {
+  if (searchWrapper) {
-    searchTrigger.remove();
+    searchWrapper.remove();
   }
 
-  // Remove progress container
+  // Defensive cleanup for older mounts that may have left the trigger or
-  const progressContainer = document.querySelector(".search-progress-container");
+  // progress container as direct children of the topbar.
-  if (progressContainer) {
+  document.querySelector(".search-trigger")?.remove();
-    progressContainer.remove();
+  document.querySelector(".search-progress-container")?.remove();
-  }
 
   // Remove search root
   const searchRoot = document.querySelector("div[data-search-root]");

src/plugins/built-in/globalSearch/src/core/styles.css

@@ -1,15 +1,72 @@
+/*
+ * Wrapper that owns the auto-margin so the whole search-trigger-and-progress
+ * group sits at the left of the SEQTA topbar. Previously, only the
+ * `.search-trigger` had `margin-right: auto`, which pushed the progress text
+ * all the way to the far right of the screen.
+ */
+.search-trigger-wrapper {
+  display: flex !important;
+  align-items: center;
+  gap: 12px;
+  margin-left: 10px;
+  margin-right: auto !important;
+  /* Allow the bar's bottom portion to peek out below the wrapper without
+     getting clipped by the topbar's flex line. */
+  overflow: visible;
+}
+
+/*
+ * Stacks the clickable row and the progress strip as one visual “chip”
+ * so the bar is flush under the button (no floating gap).
+ */
+.search-trigger-anchor {
+  display: inline-flex;
+  flex-direction: column;
+  align-items: stretch;
+  vertical-align: middle;
+  border-radius: 8px;
+  overflow: hidden;
+  box-shadow:
+    0 1px 0 rgba(255, 255, 255, 0.06) inset,
+    0 3px 8px rgba(0, 0, 0, 0.12);
+}
+
+.dark .search-trigger-anchor {
+  box-shadow:
+    0 1px 0 rgba(255, 255, 255, 0.04) inset,
+    0 3px 10px rgba(0, 0, 0, 0.45);
+}
+
+.search-trigger-anchor.is-indexing {
+  /* Very soft “rear card” edge — tweak opacity if SEQTA chrome is noisy */
+  box-shadow:
+    0 1px 0 rgba(255, 255, 255, 0.06) inset,
+    0 3px 8px rgba(0, 0, 0, 0.14),
+    1px 3px 0 rgba(139, 92, 246, 0.14),
+    0 2px 6px rgba(0, 0, 0, 0.08);
+}
+
+.dark .search-trigger-anchor.is-indexing {
+  box-shadow:
+    0 1px 0 rgba(255, 255, 255, 0.05) inset,
+    0 4px 12px rgba(0, 0, 0, 0.5),
+    1px 3px 0 rgba(167, 139, 250, 0.12),
+    0 2px 8px rgba(0, 0, 0, 0.25);
+}
+
 .search-trigger {
   display: flex;
   align-items: center;
   justify-content: center;
+  flex: none;
   height: 32px;
-  margin-left: 10px;
+  border-radius: 0;
-  border-radius: 8px;
   cursor: pointer;
-  transition: all 0.2s ease;
+  transition:
-  margin-right: auto !important;
+    background-color 0.2s ease,
+    border-color 0.2s ease;
   padding: 3px 12px;
-  box-shadow: 0 2px 5px rgba(0, 0, 0, 0.2);
+  box-shadow: none;
   backdrop-filter: blur(4px);
   user-select: none;
 
@@ -28,10 +85,12 @@
   }
 }
 
-/* Light mode styles */
+/* Light mode chip */
 .search-trigger {
-  background-color: rgba(248, 250, 252, 0.05) !important;
   border: 1px solid rgba(0, 0, 0, 0.1) !important;
+  border-bottom: none;
+  border-radius: 8px 8px 0 0;
+  background-color: rgba(248, 250, 252, 0.94) !important;
   color: #555 !important;
 
   p {
@@ -44,8 +103,10 @@
 }
 
 .dark .search-trigger {
-  background-color: rgba(0, 0, 0, 0.03) !important;
   border: 1px solid rgba(255, 255, 255, 0.1) !important;
+  border-bottom: none;
+  border-radius: 8px 8px 0 0;
+  background-color: rgba(24, 24, 27, 0.92) !important;
   color: #aaa !important;
 
   p {
@@ -57,7 +118,17 @@
   }
 }
 
+/*
+ * Idle: full pill rounding + closed bottom border on the anchor chip.
+ */
+.search-trigger-anchor:not(.is-indexing) .search-trigger {
+  border-radius: 8px !important;
+  border-bottom: 1px solid rgba(0, 0, 0, 0.1) !important;
+}
 
+.dark .search-trigger-anchor:not(.is-indexing) .search-trigger {
+  border-bottom: 1px solid rgba(255, 255, 255, 0.1) !important;
+}
 .highlight {
   background-color: rgba(255, 213, 0, 0.3);
   font-weight: 500;
@@ -83,57 +154,139 @@
   animation: shimmer 2s infinite;
 }
 
-/* Progress indicator next to search trigger */
+/*
-.search-progress-container {
+ * Thin track flush under `.search-trigger` — same width as chip, shared
-  display: flex;
+ * `overflow:hidden` rounding on `.search-trigger-anchor`.
-  align-items: center;
+ */
-  gap: 8px;
-  margin-left: 8px;
-  min-width: 120px;
-  max-width: 200px;
-  height: 32px;
-}
-
 .search-progress-bar-wrapper {
-  flex: 1;
+  flex: none;
-  height: 4px;
+  height: 0;
-  background: rgba(0, 0, 0, 0.1);
+  min-height: 0;
-  border-radius: 2px;
+  border: none;
+  background: transparent;
+  border-radius: 0;
   overflow: hidden;
-  display: none;
+  opacity: 1;
-  min-width: 60px;
+  transform: none;
+  pointer-events: none;
+  transition: height 0.22s cubic-bezier(0.2, 0.7, 0.3, 1);
 }
 
-.dark .search-progress-bar-wrapper {
+.search-progress-bar-wrapper.is-active {
-  background: rgba(255, 255, 255, 0.1);
+  height: 4px;
+}
+
+.search-progress-track {
+  box-sizing: border-box;
+  height: 100%;
+  width: 100%;
+  position: relative;
+  overflow: hidden;
+  background: rgba(15, 23, 42, 0.08);
+}
+
+.dark .search-progress-track {
+  background: rgba(248, 250, 252, 0.1);
 }
 
 .search-progress-bar {
-  height: 100%;
-  background: linear-gradient(90deg, #3b82f6, #2563eb, #3b82f6);
-  transition: width 0.3s ease-out;
-  width: 0%;
   position: relative;
-  border-radius: 2px;
+  height: 100%;
+  width: 0%;
+  background: linear-gradient(90deg, #38bdf8, #2563eb);
+  transition:
+    width 0.35s cubic-bezier(0.2, 0.7, 0.35, 1),
+    background 0.25s ease;
+}
+
+.search-progress-bar-wrapper.is-rough-complete .search-progress-track {
+  background: rgba(185, 28, 28, 0.12);
+}
+
+.dark .search-progress-bar-wrapper.is-rough-complete .search-progress-track {
+  background: rgba(248, 113, 113, 0.12);
+}
+
+.search-progress-bar-wrapper.is-rough-complete .search-progress-bar {
+  background: linear-gradient(90deg, #f87171, #dc2626);
 }
 
 .search-progress-bar::after {
   content: '';
   position: absolute;
   inset: 0;
-  background: linear-gradient(90deg, transparent, rgba(255,255,255,0.3), transparent);
+  background: linear-gradient(
+    90deg,
+    transparent,
+    rgba(255, 255, 255, 0.28),
+    transparent
+  );
   animation: shimmer 2s infinite;
-  border-radius: 2px;
+}
+/*
+ * Progress label sits as a flex child immediately to the right of the
+ * search button (gap is provided by .search-trigger-wrapper). It's hidden
+ * by default and fades in once an indexing pass is active.
+ */
+.search-progress-text {
+  display: block;
+  font-size: 12px;
+  color: #475569;
+  white-space: nowrap;
+  font-weight: 500;
+  opacity: 0;
+  transform: translateX(-4px);
+  transition: opacity 0.2s ease, transform 0.2s ease, color 0.25s ease;
+  pointer-events: none;
+  max-width: 240px;
+  overflow: hidden;
+  text-overflow: ellipsis;
+  line-height: 1.35;
+  letter-spacing: 0.01em;
+  flex: 0 0 auto;
+  align-self: center;
 }
 
-.search-progress-text {
+/* While indexing: same neutral label colour as default (only “Done!” is green). */
-  font-size: 11px;
+.search-progress-text.is-active {
-  color: #666;
+  opacity: 1;
-  white-space: nowrap;
+  transform: translateX(0);
-  display: none;
+  color: #475569;
-  font-weight: 500;
+}
+
+/* Completed pass — green text only here, not on the strip or chip */
+.search-progress-text.is-active.is-done-message {
+  font-weight: 600;
+  letter-spacing: 0.02em;
+  color: #15803d !important;
+}
+
+.dark .search-progress-text.is-active.is-done-message {
+  color: #4ade80 !important;
+}
+
+/* After DONE_HOLD_MS, fade out before DOM teardown */
+.search-progress-text.is-active.is-fading-done {
+  opacity: 0;
+  transform: translateX(-4px);
+  transition:
+    opacity 0.5s ease,
+    transform 0.45s ease,
+    color 0.25s ease;
 }
 
 .dark .search-progress-text {
-  color: #999;
+  color: #cbd5e1;
+}
+
+.dark .search-progress-text.is-active {
+  color: #cbd5e1;
+}
+
+.search-progress-text.is-active.is-rough {
+  color: #b91c1c;
+}
+
+.dark .search-progress-text.is-active.is-rough {
+  color: #fca5a5;
 }

src/plugins/built-in/globalSearch/src/indexing/__fixtures__/seqtaResponses.ts

@@ -0,0 +1,161 @@
+/**
+ * Representative SEQTA response shapes captured from a real `/seqta/student/`
+ * session via the websiteskimmer recorder. These are static fixtures used
+ * by `selfTests.ts` to verify our extractors and the passive observer
+ * remain compatible with the upstream API as it evolves.
+ *
+ * NOTE: These fixtures are scrubbed of any secrets and reduced in size; the
+ * structure (keys, types, nesting) faithfully matches what SEQTA returns
+ * but the values are illustrative rather than real student data.
+ */
+
+export const subjectsListPayload = [
+  {
+    code: "2026S1",
+    description: "Sample Semester 1 timetable",
+    active: 1,
+    id: 77,
+    subjects: [
+      {
+        code: "ENGG1",
+        classunit: 29248,
+        description: "English GEN 1",
+        metaclass: 29611,
+        title: "English GEN 1",
+        programme: 3830,
+        marksbook_type: "numeric",
+      },
+      {
+        code: "MASA1",
+        classunit: 29247,
+        description: "Mathematics Specialist 1",
+        metaclass: 29610,
+        title: "Mathematics Specialist 1",
+        programme: 3831,
+        marksbook_type: "numeric",
+      },
+    ],
+  },
+];
+
+export const coursesPayload = {
+  c: "ENGG1#1",
+  t: "English GEN 1",
+  i: 3830,
+  m: 29611,
+  document:
+    '{"document":{"modules":[{"uuid":"1641cf87-ae08-4bcb-832d-d5709d84d0c5"}]}}',
+  w: [
+    [
+      { t: "", h: "", i: 248293, l: "", n: 0, o: "" },
+      {
+        t: "",
+        i: 248316,
+        l: '<p><a href="http://ed.ted.com/on/r80lnJL0#watch">http://ed.ted.com/on/r80lnJL0#watch</a></p>',
+        n: 1,
+        o: "",
+      },
+    ],
+    [{ t: "Lesson 2", h: "<h1>Module 2</h1>", i: 248294, l: "", n: 0, o: "" }],
+  ],
+};
+
+export const messagesListPayload = {
+  hasMore: false,
+  messages: [
+    {
+      date: "2026-04-29 04:26:25.075868+00",
+      attachments: false,
+      read: 1,
+      sender: "Jacob Johannesburg",
+      subject: "test",
+      sender_type: "student",
+      attachmentCount: 0,
+      id: 81469,
+      sender_id: 3111,
+    },
+  ],
+  ts: "2026-04-30 03:25:02.27900",
+};
+
+export const documentsPayload = [
+  {
+    docs: [
+      {
+        file: 49555,
+        filename: "School Glossary.docx",
+        size: "14931",
+        context_uuid: "3162189c-2052-4f83-ad83-a66c57460ea2",
+        mimetype:
+          "application/vnd.openxmlformats-officedocument.wordprocessingml.document",
+        created_date: "2021-08-04 12:55:55.102653+00",
+        title: "School Glossary",
+        uuid: "3162189c-2052-4f83-ad83-a66c57460ea2",
+        created_by: "537",
+      },
+    ],
+    id: 9,
+    category: "Document repository",
+  },
+];
+
+export const noticesPayload = [
+  {
+    id: 12345,
+    title: "Lunchtime sport tomorrow",
+    contents: "<p>Bring shoes.</p>",
+    staff: "Mr Coach",
+    staff_id: 246,
+    label: 9,
+    label_title: "All Students",
+    colour: "#ff5722",
+  },
+];
+
+export const portalsPayload = [
+  {
+    is_power_portal: false,
+    inherit_styles: true,
+    icon: "colour-cerulean",
+    id: 328,
+    label: "Mathletics",
+    priority: 20,
+    uuid: "9d20f40c-fdc9-4aa3-91f1-905d86e240c4",
+    url: "www.mathletics.com/",
+  },
+];
+
+export const folioListPayload = {
+  me: "Jacob Johannesburg",
+  list: [
+    {
+      student: "Jacob Johannesburg",
+      id: 203,
+      published: "2026-04-14 20:02:50",
+      title: "My folio",
+    },
+  ],
+};
+
+export const folioEntryPayload = {
+  forum: 478,
+  contents:
+    '[[embed:raw|<p>Some <strong>reflection</strong> text.</p>]] Plain trailing text.',
+  created: "2026-04-14 10:32:34.264641+00",
+  allow_comments: true,
+  author: { year: "Year 10", name: "Jacob Johannesburg", id: 3111 },
+  files: [],
+  id: 203,
+  published: "2026-04-14 20:02:50",
+  title: "My folio",
+  updated: "2026-04-14 10:32:50.696678+00",
+};
+
+/**
+ * Settings payload contains tenant-wide configuration including third-party
+ * URLs and API keys. The passive observer must NEVER index this route.
+ */
+export const settingsPayload = {
+  "global.dropbox.api.key": { value: "xxx-do-not-index" },
+  "global.ai.api.baseurl": { value: "https://example.com" },
+};

src/plugins/built-in/globalSearch/src/indexing/actions.ts

@@ -28,6 +28,40 @@ interface AssessmentMetadata {
 
 type ActionHandler<T = any> = (item: IndexItem & { metadata: T }) => void;
 
+/**
+ * Navigate to a SEQTA SPA hash route in the most reliable way available.
+ *
+ * Setting `location.hash` works when the destination module is already
+ * registered with SEQTA's hashchange router (as is the case for the
+ * existing `message`/`assessment` actions, which then poke at the live
+ * DOM). For navigations that switch to a module the SPA may not have
+ * loaded yet (courses, forums, folios, portals, documents, reports,
+ * goals, notices, ...) we instead assign through `location.href` against
+ * the canonical `${origin}/` base. The path stays `/`, so the browser
+ * still treats this as a hash-only change in practice — but if anything
+ * went sideways with the path, we get a clean reload that bootstraps the
+ * SPA fresh, which is far less surprising than a blank screen.
+ */
+function navigateToHashRoute(routeWithLeadingSlash: string): void {
+  const target = `${location.origin}/#?page=${routeWithLeadingSlash}`;
+  window.location.href = target;
+}
+
+function navigateInCurrentSeqtaApp(routeWithLeadingSlash: string): void {
+  window.location.hash = `#?page=${routeWithLeadingSlash}`;
+}
+
+/**
+ * Final-fallback hub when an item has no usable deep-link metadata.
+ *
+ * `/dashboard` is the standard SEQTA Learn landing page and is the
+ * destination the websiteskimmer recording captured for unknown routes.
+ * `/home` is BetterSEQTA-Plus's custom replacement which only renders
+ * after our content script has hooked the SPA — using it as a fallback
+ * from a fresh nav can produce a blank frame.
+ */
+const FALLBACK_ROUTE = "/dashboard";
+
 export const actionMap: Record<string, ActionHandler<any>> = {
   message: (async (item: IndexItem & { metadata: MessageMetadata }) => {
     window.location.hash = `#?page=/messages`;
@@ -81,32 +115,34 @@ export const actionMap: Record<string, ActionHandler<any>> = {
       }
     }
     
-    // Try to extract metadata values using multiple methods to handle XrayWrapper
+    // Try to extract metadata values using multiple methods to handle XrayWrapper.
+    // The metadata bag is intentionally typed loosely here because Firefox's
+    // XrayWrapper occasionally surfaces extra/casing-variant keys we still
+    // want to read defensively.
     const getMetadataValue = (key: string, altKey?: string): any => {
+      const bag = metadata as unknown as Record<string, any>;
       try {
-        // Try direct access first
+        const value = bag[key];
-        const value = metadata[key];
         if (value !== undefined && value !== null) {
           return value;
         }
         if (altKey) {
-          const altValue = metadata[altKey];
+          const altValue = bag[altKey];
           if (altValue !== undefined && altValue !== null) {
             return altValue;
           }
         }
-        // Try accessing via Object.keys iteration (works around XrayWrapper)
         try {
-          const keys = Object.keys(metadata);
+          const keys = Object.keys(bag);
           for (const k of keys) {
             if (k === key || k === altKey) {
-              const val = metadata[k];
+              const val = bag[k];
               if (val !== undefined && val !== null) {
                 return val;
               }
             }
           }
-        } catch (e) {
+        } catch {
           // Object.keys might fail on XrayWrapper, that's okay
         }
         return undefined;
@@ -189,14 +225,218 @@ export const actionMap: Record<string, ActionHandler<any>> = {
   }) as ActionHandler<any>,
 
   subjectassessment: ((item: IndexItem) => {
-    window.location.href = `/#?page=/assessments/${item.metadata.programme}:${item.metadata.subjectId}`;
+    navigateToHashRoute(
+      `/assessments/${item.metadata.programme}:${item.metadata.subjectId}`,
+    );
   }) as ActionHandler<any>,
 
   subjectcourse: ((item: IndexItem) => {
-    window.location.href = `/#?page=/courses/${item.metadata.programme}:${item.metadata.subjectId}`;
+    navigateToHashRoute(
+      `/courses/${item.metadata.programme}:${item.metadata.subjectId}`,
+    );
   }) as ActionHandler<any>,
 
   forum: ((item: IndexItem) => {
-    window.location.href = `/#?page=/forums/${item.metadata.forumId}`;
+    navigateToHashRoute(`/forums/${item.metadata.forumId}`);
+  }) as ActionHandler<any>,
+
+  course: ((item: IndexItem) => {
+    const programme = item.metadata?.programme;
+    const metaclass = item.metadata?.metaclass ?? item.metadata?.subjectId;
+    if (programme !== undefined && metaclass !== undefined) {
+      navigateToHashRoute(`/courses/${programme}:${metaclass}`);
+      return;
+    }
+    if (item.metadata?.route) {
+      navigateToHashRoute(String(item.metadata.route));
+      return;
+    }
+    navigateToHashRoute(FALLBACK_ROUTE);
+  }) as ActionHandler<any>,
+
+  notice: ((_item: IndexItem) => {
+    // SEQTA's notices route doesn't honour `&date=` from the hash, so just
+    // open the listing.
+    navigateToHashRoute("/notices");
+  }) as ActionHandler<any>,
+
+  document: ((_item: IndexItem) => {
+    // We don't trigger downloads automatically: opening the documents page
+    // gives users full SEQTA controls (preview, download, share) without
+    // needing the JWT-stamped streaming URL we deliberately avoid storing.
+    navigateToHashRoute("/documents");
+  }) as ActionHandler<any>,
+
+  folio: ((_item: IndexItem) => {
+    // SEQTA's folio SPA does not expose a per-id route; the previous
+    // `?page=/folios/read?id=N` shape contained a literal `?` inside the
+    // `page` query value and was unmatchable, which sent users to the
+    // dashboard. Always land on the read view and let the user pick.
+    navigateToHashRoute("/folios/read");
+  }) as ActionHandler<any>,
+
+  portal: ((item: IndexItem) => {
+    // SEQTA renders portals via the in-app viewer at `?page=/portals/<uuid>`
+    // (verified via the websiteskimmer capture). Prefer that so SSO/headers
+    // are preserved; only pop the external URL as a fallback if we don't
+    // have a UUID; final fallback to the dashboard rather than blanking.
+    const uuid = item.metadata?.portalUuid;
+    if (typeof uuid === "string" && uuid) {
+      navigateToHashRoute(`/portals/${uuid}`);
+      return;
+    }
+    const url = item.metadata?.url;
+    if (typeof url === "string" && url) {
+      window.open(url, "_blank", "noopener,noreferrer");
+      return;
+    }
+    navigateToHashRoute(FALLBACK_ROUTE);
+  }) as ActionHandler<any>,
+
+  report: ((_item: IndexItem) => {
+    navigateToHashRoute("/reports");
+  }) as ActionHandler<any>,
+
+  goal: ((item: IndexItem) => {
+    const year = item.metadata?.year;
+    if (year !== undefined) {
+      navigateToHashRoute(`/goals/${year}`);
+    } else {
+      navigateToHashRoute("/goals");
+    }
+  }) as ActionHandler<any>,
+
+  /**
+   * Routes for passively-captured items.
+   *
+   * The passive observer captures whatever `/seqta/student/...` JSON the
+   * page is fetching, so we can't trust a single category to imply a
+   * single SEQTA SPA route. Instead, derive the destination from the API
+   * route the entity came from, augmented with entity-shaped hints
+   * (programme/metaclass/year/uuid/...) that the observer hoists into
+   * metadata. We never replay the original POST: actions are user-driven
+   * and must stay safe even though the observer's own denylist excludes
+   * `save/*` and friends.
+   */
+  passive: ((item: IndexItem) => {
+    const md = (item.metadata ?? {}) as Record<string, unknown>;
+    const route = typeof md.route === "string" ? (md.route as string) : "";
+    const sourcePage =
+      typeof md.sourcePage === "string" ? (md.sourcePage as string) : "";
+    const routeParts = route
+      .replace(/^\/seqta\/student\/?/, "")
+      .replace(/^load\//, "")
+      .split("/")
+      .filter(Boolean)
+      .map((part) => part.toLowerCase());
+    const tail = routeParts[0] ?? "";
+    const child = routeParts[1] ?? "";
+
+    const num = (key: string): number | undefined => {
+      const value = md[key];
+      if (typeof value === "number" && Number.isFinite(value)) return value;
+      if (typeof value === "string" && value && Number.isFinite(Number(value))) {
+        return Number(value);
+      }
+      return undefined;
+    };
+    const str = (key: string): string | undefined => {
+      const value = md[key];
+      return typeof value === "string" && value ? value : undefined;
+    };
+
+    const programme = num("programme") ?? num("programmeId") ?? num("programmeID");
+    const metaclass =
+      num("metaclass") ?? num("metaclassId") ?? num("metaclassID");
+    const portalUuid = str("portalUuid") ?? str("uuid");
+    const forumId = num("forumId") ?? num("forum");
+    const year = num("year");
+    const assessmentId =
+      num("assessmentId") ?? num("assessmentID") ?? num("id");
+    const messageId = num("messageId");
+
+    if (sourcePage === "/messages") {
+      navigateInCurrentSeqtaApp("/messages");
+      return;
+    }
+
+    switch (tail) {
+      case "courses":
+        if (programme !== undefined && metaclass !== undefined) {
+          navigateToHashRoute(`/courses/${programme}:${metaclass}`);
+          return;
+        }
+        break;
+      case "assessments":
+        if (programme !== undefined && metaclass !== undefined) {
+          const itemSuffix =
+            assessmentId !== undefined ? `&item=${assessmentId}` : "";
+          navigateToHashRoute(
+            `/assessments/${programme}:${metaclass}${itemSuffix}`,
+          );
+          return;
+        }
+        if (assessmentId !== undefined) {
+          navigateToHashRoute(`/assessments/upcoming&item=${assessmentId}`);
+          return;
+        }
+        navigateToHashRoute("/assessments/upcoming");
+        return;
+      case "forums":
+      case "forum":
+        if (forumId !== undefined) {
+          navigateToHashRoute(`/forums/${forumId}`);
+          return;
+        }
+        break;
+      case "portals":
+      case "portal":
+        if (portalUuid) {
+          navigateToHashRoute(`/portals/${portalUuid}`);
+          return;
+        }
+        break;
+      case "goals":
+      case "goal":
+        navigateToHashRoute(year !== undefined ? `/goals/${year}` : "/goals");
+        return;
+      case "folio":
+      case "folios":
+        navigateToHashRoute("/folios/read");
+        return;
+      case "notices":
+      case "notice":
+        navigateToHashRoute("/notices");
+        return;
+      case "documents":
+      case "document":
+        navigateToHashRoute("/documents");
+        return;
+      case "reports":
+      case "report":
+        navigateToHashRoute("/reports");
+        return;
+      case "messages":
+      case "message":
+        // `/seqta/student/load/message/people` and related endpoints are
+        // only meaningful while SEQTA's message module is mounted. Use the
+        // same live hash navigation as the real message action instead of
+        // forcing a fresh bootstrap, which can drop back to dashboard for
+        // context-only endpoints.
+        void messageId; // noqa — preserved for future deep-select work
+        navigateInCurrentSeqtaApp("/messages");
+        return;
+      case "people":
+        if (route.includes("/load/message/people") || child === "people") {
+          navigateInCurrentSeqtaApp("/messages");
+          return;
+        }
+        break;
+      case "timetable":
+        navigateToHashRoute("/timetable");
+        return;
+    }
+
+    navigateToHashRoute(FALLBACK_ROUTE);
   }) as ActionHandler<any>,
 };

src/plugins/built-in/globalSearch/src/indexing/api.ts

@@ -0,0 +1,386 @@
+import { delay } from "@/seqta/utils/delay";
+
+/**
+ * Shared SEQTA HTTP layer used by every indexing job.
+ *
+ * - All requests are same-origin POSTs against `/seqta/student/...` with
+ *   `credentials: "include"` so they inherit the user's existing session.
+ * - Responses are parsed as JSON and lightly validated (status === "200" and
+ *   payload present, mirroring the SEQTA convention).
+ * - Failures are retried with exponential backoff up to a configurable limit.
+ * - A simple per-route concurrency / spacing limiter prevents heavy jobs (e.g.
+ *   per-subject course crawls) from hammering SEQTA.
+ */
+
+export interface SeqtaResponse<T = any> {
+  payload: T;
+  status: string;
+}
+
+export interface SeqtaFetchOptions {
+  /** Defaults to "POST". */
+  method?: "POST" | "GET";
+  /** Maximum number of retries for transient failures (default 2). */
+  retries?: number;
+  /** Initial backoff delay in ms (default 200). */
+  baseDelayMs?: number;
+  /** Hard cap on total request time in ms (default 20s). */
+  timeoutMs?: number;
+  /** AbortSignal for cancellation. */
+  signal?: AbortSignal;
+  /** Skip the routing limiter (rare; only for already-throttled callers). */
+  skipLimiter?: boolean;
+}
+
+const DEFAULT_RETRIES = 2;
+const DEFAULT_BASE_DELAY = 200;
+const DEFAULT_TIMEOUT = 20_000;
+
+/* ------------------------------------------------------------------ */
+/*                              limiter                               */
+/* ------------------------------------------------------------------ */
+
+/**
+ * Caps concurrent in-flight requests per normalized SEQTA route. Indexing
+ * jobs often fan out (e.g. one /load/courses per subject); we don't want them
+ * sending dozens of requests in parallel.
+ */
+class RouteLimiter {
+  private inFlight = new Map<string, number>();
+  private waiters = new Map<string, Array<() => void>>();
+  private readonly maxConcurrent: number;
+
+  constructor(maxConcurrent = 4) {
+    this.maxConcurrent = maxConcurrent;
+  }
+
+  async acquire(route: string): Promise<() => void> {
+    const current = this.inFlight.get(route) ?? 0;
+    if (current < this.maxConcurrent) {
+      this.inFlight.set(route, current + 1);
+      return () => this.release(route);
+    }
+
+    return new Promise((resolve) => {
+      const queue = this.waiters.get(route) ?? [];
+      queue.push(() => {
+        this.inFlight.set(route, (this.inFlight.get(route) ?? 0) + 1);
+        resolve(() => this.release(route));
+      });
+      this.waiters.set(route, queue);
+    });
+  }
+
+  private release(route: string) {
+    const next = (this.inFlight.get(route) ?? 1) - 1;
+    if (next <= 0) {
+      this.inFlight.delete(route);
+    } else {
+      this.inFlight.set(route, next);
+    }
+    const queue = this.waiters.get(route);
+    if (queue && queue.length > 0) {
+      const wake = queue.shift()!;
+      if (queue.length === 0) this.waiters.delete(route);
+      wake();
+    }
+  }
+}
+
+const routeLimiter = new RouteLimiter(4);
+
+/* ------------------------------------------------------------------ */
+/*                       route normalization                          */
+/* ------------------------------------------------------------------ */
+
+/**
+ * Strips the volatile anti-replay query token (e.g. `?mokx3qef`) so we can
+ * key caches and limiters off the canonical route.
+ */
+export function normalizeSeqtaPath(url: string): string {
+  try {
+    const parsed = new URL(url, location.origin);
+    // SEQTA appends a single random query token like `?mokx3qef`. Drop the
+    // entire query string so canonicalization is robust.
+    return parsed.pathname;
+  } catch {
+    // Fallback for already-relative URLs.
+    return url.split("?")[0];
+  }
+}
+
+/* ------------------------------------------------------------------ */
+/*                           sensitive routes                         */
+/* ------------------------------------------------------------------ */
+
+/**
+ * Routes whose responses must never be indexed because they contain
+ * credentials, secrets, JWTs, or arbitrary configuration blobs.
+ */
+const SENSITIVE_PATH_PATTERNS: RegExp[] = [
+  /\/seqta\/student\/login(\b|\/)/i,
+  /\/seqta\/student\/save\//i,
+  /\/seqta\/student\/load\/settings(\b|\/)/i,
+  /\/seqta\/student\/load\/prefs(\b|\/)/i,
+  /\/seqta\/student\/heartbeat(\b|\/)/i,
+  /\/seqta\/student\/storage(\b|\/)/i,
+  /\/seqta\/student\/themes\//i,
+  /\/seqta\/student\/branding\//i,
+  /\/seqta\/student\/releasealert\//i,
+  /\/seqta\/student\/files\/stream(\b|\/)/i,
+  /\/seqta\/student\/load\/file(\b|\/)/i,
+  /\/seqta\/ta\/masquerade(\b|\/)/i,
+];
+
+export function isSensitiveSeqtaPath(path: string): boolean {
+  const normalized = normalizeSeqtaPath(path);
+  return SENSITIVE_PATH_PATTERNS.some((re) => re.test(normalized));
+}
+
+/* ------------------------------------------------------------------ */
+/*                     student / user identity                        */
+/* ------------------------------------------------------------------ */
+
+interface SeqtaUserInfo {
+  id?: number;
+  personUUID?: string;
+  username?: string;
+  [key: string]: unknown;
+}
+
+let cachedUserInfo: SeqtaUserInfo | null = null;
+let inflightUserInfo: Promise<SeqtaUserInfo | null> | null = null;
+
+/**
+ * Resolves the current SEQTA user identity by re-using the same `login`
+ * handshake that the host page performs. This is the canonical way to
+ * discover the active student id and avoids the historical hard-coded
+ * `student: 69` placeholder that was incorrect on every real instance.
+ *
+ * Failures are intentionally NOT cached — a transient login glitch on the
+ * very first call must not poison the cache for the lifetime of the page,
+ * because every subsequent indexing pass that needs the student id (e.g.
+ * the assignments job) would skip silently.
+ */
+export async function getCurrentUserInfo(): Promise<SeqtaUserInfo | null> {
+  if (cachedUserInfo) return cachedUserInfo;
+  if (inflightUserInfo) return inflightUserInfo;
+
+  inflightUserInfo = (async () => {
+    try {
+      const res = await fetch(`${location.origin}/seqta/student/login`, {
+        method: "POST",
+        credentials: "include",
+        headers: { "Content-Type": "application/json; charset=utf-8" },
+        body: JSON.stringify({
+          mode: "normal",
+          query: null,
+          redirect_url: location.origin,
+        }),
+      });
+      if (!res.ok) return null;
+      const json = (await res.json()) as { payload?: SeqtaUserInfo };
+      const payload = json?.payload ?? null;
+      if (payload && typeof payload === "object") {
+        cachedUserInfo = payload;
+        return payload;
+      }
+      return null;
+    } catch (e) {
+      console.warn(
+        "[Global Search API] Failed to resolve current user info:",
+        e,
+      );
+      return null;
+    } finally {
+      inflightUserInfo = null;
+    }
+  })();
+
+  return inflightUserInfo;
+}
+
+/**
+ * Best-effort lookup of the active student id. Returns `undefined` when the
+ * value cannot be discovered (jobs should fall back gracefully rather than
+ * fabricating an id).
+ */
+export async function getCurrentStudentId(): Promise<number | undefined> {
+  const info = await getCurrentUserInfo();
+  const id = info?.id;
+  if (typeof id === "number" && Number.isFinite(id)) return id;
+  return undefined;
+}
+
+/* ------------------------------------------------------------------ */
+/*                            core fetch                              */
+/* ------------------------------------------------------------------ */
+
+class SeqtaApiError extends Error {
+  status: number;
+  route: string;
+  constructor(message: string, status: number, route: string) {
+    super(message);
+    this.name = "SeqtaApiError";
+    this.status = status;
+    this.route = route;
+  }
+}
+
+function isTransientError(err: unknown): boolean {
+  if (err instanceof SeqtaApiError) {
+    if (err.status === 0 || err.status >= 500) return true;
+    if (err.status === 429) return true;
+    return false;
+  }
+  if (err instanceof TypeError) return true;
+  if ((err as any)?.name === "AbortError") return false;
+  return true;
+}
+
+/**
+ * Sends a JSON POST against a SEQTA route and returns the parsed envelope.
+ *
+ * - Adds `credentials: "include"` so requests reuse the active session.
+ * - Sets `X-Requested-With: XMLHttpRequest` so SEQTA classifies the request
+ *   the same way as the first-party SPA (some routes 4xx without it).
+ * - Retries transient network/server errors with exponential backoff.
+ * - Validates that the response is JSON and has `status === "200"` (matches
+ *   the SEQTA convention; jobs that need raw payloads can pass `path` but
+ *   call `seqtaFetch` directly via the underlying API if they need to).
+ */
+export async function seqtaFetchJson<T = any>(
+  path: string,
+  body: Record<string, unknown> | undefined = {},
+  options: SeqtaFetchOptions = {},
+): Promise<SeqtaResponse<T>> {
+  const route = normalizeSeqtaPath(path);
+  const retries = Math.max(0, options.retries ?? DEFAULT_RETRIES);
+  const baseDelay = Math.max(50, options.baseDelayMs ?? DEFAULT_BASE_DELAY);
+  const timeoutMs = Math.max(1_000, options.timeoutMs ?? DEFAULT_TIMEOUT);
+
+  let release: (() => void) | null = null;
+  if (!options.skipLimiter) {
+    release = await routeLimiter.acquire(route);
+  }
+
+  try {
+    let attempt = 0;
+    let lastError: unknown = null;
+
+    while (attempt <= retries) {
+      const controller = new AbortController();
+      const timer = setTimeout(() => controller.abort(), timeoutMs);
+
+      const onAbort = () => controller.abort();
+      if (options.signal) {
+        if (options.signal.aborted) controller.abort();
+        else options.signal.addEventListener("abort", onAbort, { once: true });
+      }
+
+      try {
+        const res = await fetch(`${location.origin}${route}`, {
+          method: options.method ?? "POST",
+          credentials: "include",
+          headers: {
+            "Content-Type": "application/json; charset=utf-8",
+            "X-Requested-With": "XMLHttpRequest",
+            Accept: "text/javascript, text/html, application/xml, text/xml, */*",
+          },
+          body: body === undefined ? undefined : JSON.stringify(body),
+          signal: controller.signal,
+        });
+
+        if (!res.ok) {
+          throw new SeqtaApiError(
+            `HTTP ${res.status} ${res.statusText} for ${route}`,
+            res.status,
+            route,
+          );
+        }
+
+        const rawJson = (await res.json()) as unknown;
+        if (!rawJson || typeof rawJson !== "object") {
+          throw new SeqtaApiError(
+            `Invalid SEQTA response (not a JSON object) for ${route}`,
+            res.status,
+            route,
+          );
+        }
+
+        // SEQTA's "envelope" convention is `{ status, payload }`, but in
+        // practice some endpoints — notably `/seqta/student/load/subjects`
+        // and `/seqta/student/assessment/list/*` — occasionally return
+        // either a bare array or an envelope with a non-"200" status.
+        // Strict validation here was historically silently killing the
+        // assignments + courses indexing pipelines when those endpoints
+        // returned a quirky shape, so we normalize permissively and let
+        // callers handle missing/empty payloads.
+        let json: SeqtaResponse<T>;
+        if (Array.isArray(rawJson)) {
+          json = { payload: rawJson as unknown as T, status: "200" };
+        } else {
+          const obj = rawJson as Record<string, unknown>;
+          const hasEnvelopeKey = "payload" in obj || "status" in obj;
+          if (hasEnvelopeKey) {
+            json = {
+              payload: ("payload" in obj ? obj.payload : undefined) as T,
+              status:
+                typeof obj.status === "string"
+                  ? obj.status
+                  : typeof obj.status === "number"
+                    ? String(obj.status)
+                    : "200",
+            };
+          } else {
+            json = { payload: rawJson as unknown as T, status: "200" };
+          }
+        }
+
+        if (json.status && json.status !== "200") {
+          console.warn(
+            `[Global Search API] Non-200 SEQTA status "${json.status}" for ${route} — returning payload anyway`,
+          );
+        }
+
+        return json;
+      } catch (err) {
+        lastError = err;
+        if (!isTransientError(err) || attempt === retries) {
+          throw err;
+        }
+        const wait = Math.min(5_000, baseDelay * Math.pow(2, attempt));
+        await delay(wait);
+        attempt++;
+      } finally {
+        clearTimeout(timer);
+        if (options.signal) options.signal.removeEventListener("abort", onAbort);
+      }
+    }
+
+    throw lastError ?? new Error(`seqtaFetchJson exhausted retries for ${route}`);
+  } finally {
+    if (release) release();
+  }
+}
+
+/**
+ * Convenience helper: fetch and unwrap `.payload` directly. Returns `null`
+ * on failure rather than throwing, so jobs can use the value optionally.
+ */
+export async function seqtaFetchPayload<T = any>(
+  path: string,
+  body: Record<string, unknown> | undefined = {},
+  options: SeqtaFetchOptions = {},
+): Promise<T | null> {
+  try {
+    const res = await seqtaFetchJson<T>(path, body, options);
+    return res.payload ?? null;
+  } catch (e) {
+    console.warn(
+      `[Global Search API] Request to ${normalizeSeqtaPath(path)} failed:`,
+      e,
+    );
+    return null;
+  }
+}

src/plugins/built-in/globalSearch/src/indexing/extract.ts

@@ -0,0 +1,303 @@
+import { htmlToPlainText } from "./utils";
+import type { IndexItem } from "./types";
+
+/**
+ * Safe extraction helpers used by both active SEQTA jobs and the passive
+ * network observer.
+ *
+ * The goal is to take arbitrary SEQTA JSON / embedded HTML fragments and
+ * derive concise, redacted, search-friendly text without ever indexing
+ * obvious credentials, tokens, JWTs, or large binary blobs.
+ */
+
+/* ------------------------------------------------------------------ */
+/*                          sensitive keys                            */
+/* ------------------------------------------------------------------ */
+
+/**
+ * Field names whose values should never be indexed regardless of context.
+ * Matches SEQTA's frequently-used credential / config keys plus generic
+ * security-related names. Comparison is case-insensitive and matches both
+ * the full key and any sub-string fragments (so `client_secret`,
+ * `apiKey`, `dropboxKey` all hit).
+ */
+const SENSITIVE_KEY_FRAGMENTS: readonly string[] = [
+  "password",
+  "passwd",
+  "pwd",
+  "secret",
+  "token",
+  "jwt",
+  "session",
+  "cookie",
+  "auth",
+  "apikey",
+  "api_key",
+  "clientid",
+  "client_id",
+  "clientsecret",
+  "client_secret",
+  "credential",
+  "private",
+  "salt",
+  "hash",
+  "csrf",
+  "x-api",
+  "bearer",
+  "dropbox",
+  "oauth",
+  "signature",
+];
+
+export function isSensitiveKey(key: string): boolean {
+  if (!key) return false;
+  const lower = key.toLowerCase();
+  return SENSITIVE_KEY_FRAGMENTS.some((frag) => lower.includes(frag));
+}
+
+/**
+ * Returns true if the supplied scalar value looks credential-shaped: a long
+ * hex/base64-like blob that doesn't decode to readable text. This catches
+ * arbitrary tokens that don't have a clear field-name signal.
+ */
+export function looksLikeSecretValue(value: unknown): boolean {
+  if (typeof value !== "string") return false;
+  const trimmed = value.trim();
+  if (trimmed.length < 32) return false;
+
+  // Long contiguous base64 / hex with no whitespace and no humanish punctuation.
+  if (/\s/.test(trimmed)) return false;
+  if (/^[A-Za-z0-9+/=._-]{32,}$/.test(trimmed) && !/[.,!?]/.test(trimmed)) {
+    // Reject obvious URLs and UUIDs (they're useful and not secret).
+    if (/^https?:\/\//i.test(trimmed)) return false;
+    if (
+      /^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/i.test(
+        trimmed,
+      )
+    ) {
+      return false;
+    }
+    return true;
+  }
+
+  // JWT detection: three base64url segments separated by dots.
+  if (/^[A-Za-z0-9_-]+\.[A-Za-z0-9_-]+\.[A-Za-z0-9_-]+$/.test(trimmed)) {
+    return true;
+  }
+
+  return false;
+}
+
+/* ------------------------------------------------------------------ */
+/*                          text extraction                           */
+/* ------------------------------------------------------------------ */
+
+/**
+ * Recursively pulls human-readable text out of an arbitrary JSON value.
+ *
+ * - HTML strings are passed through `htmlToPlainText`.
+ * - Sensitive keys and secret-shaped values are skipped.
+ * - Long blobs are truncated to keep the index lean.
+ * - Arrays and objects are walked; depth is bounded to avoid pathological
+ *   structures.
+ */
+export interface ExtractTextOptions {
+  /** Hard cap on combined characters across the walk (default 4000). */
+  maxChars?: number;
+  /** Maximum recursion depth (default 6). */
+  maxDepth?: number;
+  /** Maximum array length to traverse (default 200). */
+  maxArrayItems?: number;
+  /** Skip individual string values longer than this (default 8000). */
+  maxStringLength?: number;
+}
+
+const DEFAULT_EXTRACT_OPTIONS: Required<ExtractTextOptions> = {
+  maxChars: 4000,
+  maxDepth: 6,
+  maxArrayItems: 200,
+  maxStringLength: 8000,
+};
+
+export function extractTextFromValue(
+  value: unknown,
+  options: ExtractTextOptions = {},
+): string {
+  const opts = { ...DEFAULT_EXTRACT_OPTIONS, ...options };
+  const parts: string[] = [];
+  let remaining = opts.maxChars;
+
+  const push = (text: string) => {
+    if (!text || remaining <= 0) return;
+    const trimmed = text.trim();
+    if (!trimmed) return;
+    const slice = trimmed.length > remaining ? trimmed.slice(0, remaining) : trimmed;
+    parts.push(slice);
+    remaining -= slice.length + 1;
+  };
+
+  const walk = (node: unknown, depth: number, parentKey: string | null) => {
+    if (remaining <= 0) return;
+    if (node === null || node === undefined) return;
+    if (parentKey && isSensitiveKey(parentKey)) return;
+
+    if (typeof node === "string") {
+      if (node.length > opts.maxStringLength) return;
+      if (looksLikeSecretValue(node)) return;
+      if (node.includes("<") && node.includes(">")) {
+        push(htmlToPlainText(node));
+      } else {
+        push(node);
+      }
+      return;
+    }
+
+    if (typeof node === "number" || typeof node === "boolean") {
+      // Numbers/booleans rarely contribute to search recall; skip to keep
+      // the index focused on text.
+      return;
+    }
+
+    if (depth >= opts.maxDepth) return;
+
+    if (Array.isArray(node)) {
+      const limit = Math.min(node.length, opts.maxArrayItems);
+      for (let i = 0; i < limit; i++) {
+        walk(node[i], depth + 1, parentKey);
+        if (remaining <= 0) return;
+      }
+      return;
+    }
+
+    if (typeof node === "object") {
+      for (const [key, child] of Object.entries(node as Record<string, unknown>)) {
+        if (remaining <= 0) return;
+        if (isSensitiveKey(key)) continue;
+        walk(child, depth + 1, key);
+      }
+    }
+  };
+
+  walk(value, 0, null);
+
+  return parts.join("\n").trim();
+}
+
+/* ------------------------------------------------------------------ */
+/*                         redacted clones                            */
+/* ------------------------------------------------------------------ */
+
+/**
+ * Returns a deep clone of `value` with sensitive keys/values stripped. The
+ * passive observer uses this when persisting metadata so we never store
+ * raw tokens or settings blobs in IndexedDB.
+ */
+export function redactSensitive<T>(value: T, depth = 0): T {
+  if (value === null || value === undefined) return value;
+  if (depth >= 8) return value;
+
+  if (Array.isArray(value)) {
+    return value
+      .slice(0, 200)
+      .map((v) => redactSensitive(v, depth + 1)) as unknown as T;
+  }
+  if (typeof value === "object") {
+    const out: Record<string, unknown> = {};
+    for (const [key, child] of Object.entries(value as Record<string, unknown>)) {
+      if (isSensitiveKey(key)) continue;
+      if (typeof child === "string" && looksLikeSecretValue(child)) continue;
+      out[key] = redactSensitive(child, depth + 1);
+    }
+    return out as T;
+  }
+  if (typeof value === "string" && looksLikeSecretValue(value)) {
+    return "" as unknown as T;
+  }
+  return value;
+}
+
+/* ------------------------------------------------------------------ */
+/*                       title / id heuristics                        */
+/* ------------------------------------------------------------------ */
+
+const TITLE_KEYS = [
+  "title",
+  "subject",
+  "name",
+  "label",
+  "heading",
+  "displayName",
+  "filename",
+  "code",
+];
+
+const ID_KEYS = ["id", "uuid", "messageID", "assessmentID", "notificationID"];
+
+/**
+ * Best-effort title extraction: returns the first sensible string-valued
+ * field commonly used by SEQTA payloads. Falls back to an empty string when
+ * none are present.
+ */
+export function pickTitle(node: unknown, fallback = ""): string {
+  if (!node || typeof node !== "object") return fallback;
+  const obj = node as Record<string, unknown>;
+  for (const key of TITLE_KEYS) {
+    const v = obj[key];
+    if (typeof v === "string" && v.trim()) return v.trim();
+  }
+  return fallback;
+}
+
+export function pickId(node: unknown, fallback = ""): string {
+  if (!node || typeof node !== "object") return fallback;
+  const obj = node as Record<string, unknown>;
+  for (const key of ID_KEYS) {
+    const v = obj[key];
+    if (typeof v === "string" && v.trim()) return v.trim();
+    if (typeof v === "number" && Number.isFinite(v)) return String(v);
+  }
+  return fallback;
+}
+
+/* ------------------------------------------------------------------ */
+/*                         IndexItem builders                         */
+/* ------------------------------------------------------------------ */
+
+/**
+ * Constructs an `IndexItem` from a raw entity, applying our standard
+ * extraction rules. Callers fill in the things that need domain knowledge
+ * (`category`, `actionId`, `metadata`, deep-link route hints) and we handle
+ * the boring text + redaction work.
+ */
+export function buildIndexItem(input: {
+  id: string;
+  text: string;
+  category: string;
+  rawForContent?: unknown;
+  contentOverride?: string;
+  metadata?: Record<string, unknown>;
+  actionId: string;
+  renderComponentId: string;
+  dateAdded?: number;
+  contentMaxChars?: number;
+}): IndexItem {
+  const content =
+    input.contentOverride !== undefined
+      ? input.contentOverride
+      : extractTextFromValue(input.rawForContent, {
+          maxChars: input.contentMaxChars ?? 1500,
+        });
+
+  const metadata = input.metadata ? redactSensitive(input.metadata) : {};
+
+  return {
+    id: input.id,
+    text: input.text,
+    category: input.category,
+    content,
+    dateAdded: input.dateAdded ?? Date.now(),
+    metadata,
+    actionId: input.actionId,
+    renderComponentId: input.renderComponentId,
+  };
+}

src/plugins/built-in/globalSearch/src/indexing/indexer.ts

@@ -1,10 +1,11 @@
-import { clear, get, getAll, put, remove } from "./db";
+import { clear, get, getAll, put, remove, resetDatabase } from "./db";
 import { jobs } from "./jobs";
 import { renderComponentMap } from "./renderComponents";
 import type { IndexItem, Job, JobContext } from "./types";
 import { VectorWorkerManager } from "./worker/vectorWorkerManager";
 import { loadDynamicItems } from "../utils/dynamicItems";
 import { getVectorizedItemIds } from "./utils";
+import { INDEX_SCHEMA_VERSION, SCHEMA_VERSION_KEY } from "./schemaVersion";
 
 const META_STORE = "meta";
 const LOCK_KEY = "bsq-indexer-lock";
@@ -12,6 +13,50 @@ const HEARTBEAT_INTERVAL = 10000;
 const LOCK_TIMEOUT = 20000;
 const LOCK_ACQUIRE_TIMEOUT = 5000;
 
+let schemaCheckPromise: Promise<void> | null = null;
+
+async function ensureSchemaCurrent(): Promise<void> {
+  if (schemaCheckPromise) return schemaCheckPromise;
+  schemaCheckPromise = (async () => {
+    let storedRaw: string | null = null;
+    try {
+      storedRaw = localStorage.getItem(SCHEMA_VERSION_KEY);
+    } catch {
+      return;
+    }
+    const stored = storedRaw ? parseInt(storedRaw, 10) : 0;
+    if (stored === INDEX_SCHEMA_VERSION) return;
+
+    console.warn(
+      `[Indexer] Schema version changed (${stored} -> ${INDEX_SCHEMA_VERSION}); resetting structured + vector indexes.`,
+    );
+
+    try {
+      await resetDatabase();
+    } catch (e) {
+      console.warn("[Indexer] Failed to reset structured database:", e);
+    }
+
+    try {
+      await new Promise<void>((resolve) => {
+        const req = indexedDB.deleteDatabase("embeddiaDB");
+        req.onsuccess = () => resolve();
+        req.onerror = () => resolve();
+        req.onblocked = () => resolve();
+      });
+    } catch (e) {
+      console.warn("[Indexer] Failed to reset embeddiaDB:", e);
+    }
+
+    try {
+      localStorage.setItem(SCHEMA_VERSION_KEY, String(INDEX_SCHEMA_VERSION));
+    } catch {
+      /* ignore */
+    }
+  })();
+  return schemaCheckPromise;
+}
+
 /* ─────────── Progress‑meta helpers ─────────── */
 async function loadProgress<T = any>(jobId: string): Promise<T | undefined> {
   const rec = await get(META_STORE, `progress:${jobId}`);
@@ -162,6 +207,8 @@ export async function loadAllStoredItems(): Promise<IndexItem[]> {
 }
 
 export async function runIndexing(): Promise<void> {
+  await ensureSchemaCurrent();
+
   if (!(await acquireLock())) {
     console.debug(
       "%c[Indexer] Could not acquire lock - another tab is indexing or this tab is already indexing",
@@ -178,8 +225,6 @@ export async function runIndexing(): Promise<void> {
   const totalSteps = jobIds.length + 1;
   dispatchProgress(completedJobs, totalSteps, true, "Starting jobs");
 
-  let hasStreamingJobs = false;
-
   for (const jobId of jobIds) {
     dispatchProgress(
       completedJobs,
@@ -255,10 +300,6 @@ export async function runIndexing(): Promise<void> {
       await setStoredItems(merged);
       await updateLastRunMeta(jobId);
 
-      if (jobId === 'messages' || jobId === 'notifications') {
-        hasStreamingJobs = true;
-      }
-
       console.debug(
         `%c[Indexer] ${job.label}: ${newItemsRaw.length} new items reported by run, ${merged.length} total items now in '${jobId}' store.`,
         "color: #00c46f",

src/plugins/built-in/globalSearch/src/indexing/jobs.ts

@@ -4,6 +4,14 @@ import { notificationsJob } from "./jobs/notifications";
 import { forumsJob } from "./jobs/forums";
 import { subjectsJob } from "./jobs/subjects";
 import { assignmentsJob } from "./jobs/assignments";
+import { coursesJob } from "./jobs/courses";
+import { noticesJob } from "./jobs/notices";
+import { documentsJob } from "./jobs/documents";
+import { folioJob } from "./jobs/folio";
+import { portalsJob } from "./jobs/portals";
+import { reportsJob } from "./jobs/reports";
+import { goalsJob } from "./jobs/goals";
+import { passiveJob } from "./jobs/passive";
 
 export const jobs: Record<string, Job> = {
   messages: messagesJob,
@@ -11,4 +19,12 @@ export const jobs: Record<string, Job> = {
   forums: forumsJob,
   subjects: subjectsJob,
   assignments: assignmentsJob,
+  courses: coursesJob,
+  notices: noticesJob,
+  documents: documentsJob,
+  folio: folioJob,
+  portals: portalsJob,
+  reports: reportsJob,
+  goals: goalsJob,
+  passive: passiveJob,
 };

src/plugins/built-in/globalSearch/src/indexing/jobs/courses.ts

@@ -0,0 +1,179 @@
+import type { IndexItem, Job } from "../types";
+import { seqtaFetchPayload } from "../api";
+import { buildIndexItem } from "../extract";
+import { htmlToPlainText } from "../utils";
+
+/**
+ * Indexes per-subject course content from `/seqta/student/load/courses`.
+ *
+ * The course payload contains the lesson grid in `w[][]` where each cell's
+ * `l` field is a (possibly empty) HTML snippet authored by teachers. We
+ * concatenate these into searchable text per course, plus the course title
+ * and code from `t` / `c`. Embedded files referenced via TED/SEQTA URLs are
+ * preserved as plain-text links so users can find them by URL fragment.
+ */
+
+interface SubjectsListPayload {
+  code: string;
+  description?: string;
+  active: number;
+  subjects: Array<{
+    code: string;
+    title?: string;
+    description?: string;
+    metaclass: number;
+    programme: number;
+  }>;
+}
+
+interface CoursePayload {
+  c?: string;
+  t?: string;
+  i?: number;
+  m?: number;
+  w?: Array<Array<{ l?: string; h?: string; t?: string; o?: string; i?: number }>>;
+  document?: string;
+}
+
+const fetchActiveSubjects = async (): Promise<
+  SubjectsListPayload["subjects"]
+> => {
+  const payload = await seqtaFetchPayload<SubjectsListPayload[]>(
+    "/seqta/student/load/subjects",
+    {},
+  );
+  if (!Array.isArray(payload)) return [];
+
+  const out: SubjectsListPayload["subjects"] = [];
+  for (const semester of payload) {
+    if (!semester || !Array.isArray(semester.subjects)) continue;
+    if (semester.active !== 1) continue;
+    for (const subject of semester.subjects) {
+      if (
+        subject &&
+        Number.isFinite(subject.programme) &&
+        Number.isFinite(subject.metaclass)
+      ) {
+        out.push(subject);
+      }
+    }
+  }
+  return out;
+};
+
+function flattenLessonHtml(payload: CoursePayload): string {
+  if (!Array.isArray(payload.w)) return "";
+  const fragments: string[] = [];
+  for (const row of payload.w) {
+    if (!Array.isArray(row)) continue;
+    for (const cell of row) {
+      if (!cell) continue;
+      if (typeof cell.l === "string" && cell.l.trim()) {
+        fragments.push(cell.l);
+      }
+      if (typeof cell.h === "string" && cell.h.trim()) {
+        fragments.push(cell.h);
+      }
+      if (typeof cell.t === "string" && cell.t.trim()) {
+        fragments.push(cell.t);
+      }
+      if (typeof cell.o === "string" && cell.o.trim()) {
+        fragments.push(cell.o);
+      }
+    }
+  }
+  if (fragments.length === 0) return "";
+  return htmlToPlainText(fragments.join("\n"));
+}
+
+export const coursesJob: Job = {
+  id: "courses",
+  label: "Courses",
+  renderComponentId: "course",
+  // Course content rarely changes minute-to-minute but does evolve per term.
+  // Refresh once per day (after pageLoad cool-down) to keep new lessons
+  // discoverable without hammering SEQTA.
+  frequency: { type: "expiry", afterMs: 1000 * 60 * 60 * 24 },
+
+  boostCriteria: (item, searchTerm) => {
+    if (!searchTerm) return -50;
+    let score = 0;
+    if (item.metadata?.subjectCode) score += 0.05;
+    if (item.metadata?.isActive) score += 0.02;
+    return score;
+  },
+
+  run: async (_ctx) => {
+    const subjects = await fetchActiveSubjects();
+    if (subjects.length === 0) {
+      console.debug("[Courses job] No active subjects discovered.");
+      return [];
+    }
+
+    const items: IndexItem[] = [];
+    const seenIds = new Set<string>();
+
+    // Sequential per-subject fetch keeps load on SEQTA bounded; the shared
+    // API layer also limits concurrency per route as a defense in depth.
+    for (const subject of subjects) {
+      const id = `course-${subject.programme}-${subject.metaclass}`;
+      if (seenIds.has(id)) continue;
+      seenIds.add(id);
+
+      const payload = await seqtaFetchPayload<CoursePayload>(
+        "/seqta/student/load/courses",
+        {
+          programme: String(subject.programme),
+          metaclass: String(subject.metaclass),
+        },
+      );
+
+      if (!payload) continue;
+
+      const title =
+        (typeof payload.t === "string" && payload.t.trim()) ||
+        subject.title ||
+        subject.description ||
+        subject.code ||
+        "Course";
+
+      const lessonText = flattenLessonHtml(payload);
+      const courseCode =
+        (typeof payload.c === "string" && payload.c.trim()) || subject.code;
+
+      const summary = [courseCode, lessonText]
+        .filter((s) => s && s.length > 0)
+        .join("\n")
+        .slice(0, 4000);
+
+      items.push(
+        buildIndexItem({
+          id,
+          text: title,
+          category: "courses",
+          contentOverride: summary || `Course content for ${title}`,
+          metadata: {
+            subjectCode: subject.code,
+            subjectName: subject.title ?? title,
+            programme: subject.programme,
+            metaclass: subject.metaclass,
+            courseCode,
+            isActive: true,
+            route: `/courses/${subject.programme}:${subject.metaclass}`,
+            entityType: "course",
+            icon: "\ueb4d",
+          },
+          actionId: "course",
+          renderComponentId: "course",
+        }),
+      );
+    }
+
+    console.debug(
+      `[Courses job] Indexed ${items.length} courses across ${subjects.length} subjects.`,
+    );
+    return items;
+  },
+
+  purge: (items) => items,
+};

src/plugins/built-in/globalSearch/src/indexing/jobs/documents.ts

@@ -0,0 +1,139 @@
+import type { IndexItem, Job } from "../types";
+import { seqtaFetchPayload } from "../api";
+
+/**
+ * Indexes file metadata from `/seqta/student/load/documents`.
+ *
+ * Each top-level entry is a category containing one or more documents
+ * (`docs[]`). We capture the human-readable title, filename, mimetype, and
+ * stable UUID/category for every doc, but never download or index the
+ * binary content itself - the document streaming endpoint uses one-time
+ * JWTs that are unsafe to persist or replay.
+ */
+
+interface DocumentEntry {
+  file?: number | string;
+  filename?: string;
+  size?: string | number;
+  context_uuid?: string;
+  mimetype?: string;
+  created_date?: string;
+  title?: string;
+  uuid?: string;
+  created_by?: string;
+}
+
+interface DocumentCategory {
+  id: number | string;
+  category: string;
+  colour?: string;
+  docs: DocumentEntry[];
+}
+
+function prettySize(size: string | number | undefined): string | null {
+  if (size === undefined || size === null) return null;
+  const bytes = typeof size === "string" ? parseInt(size, 10) : size;
+  if (!Number.isFinite(bytes) || bytes <= 0) return null;
+  const units = ["B", "KB", "MB", "GB"];
+  let value = bytes;
+  let i = 0;
+  while (value >= 1024 && i < units.length - 1) {
+    value /= 1024;
+    i++;
+  }
+  return `${value.toFixed(value < 10 && i > 0 ? 1 : 0)} ${units[i]}`;
+}
+
+function describeMime(mime: string | undefined): string | null {
+  if (!mime) return null;
+  if (mime.startsWith("application/pdf")) return "PDF";
+  if (mime.includes("officedocument.wordprocessingml")) return "Word";
+  if (mime.includes("officedocument.spreadsheetml")) return "Excel";
+  if (mime.includes("officedocument.presentationml")) return "PowerPoint";
+  if (mime.startsWith("image/")) return "Image";
+  if (mime.startsWith("video/")) return "Video";
+  if (mime.startsWith("audio/")) return "Audio";
+  return null;
+}
+
+export const documentsJob: Job = {
+  id: "documents",
+  label: "Documents",
+  renderComponentId: "document",
+  frequency: { type: "expiry", afterMs: 1000 * 60 * 60 * 12 }, // 12 hours
+
+  boostCriteria: (_item, searchTerm) => {
+    if (!searchTerm) return -20;
+    return 0;
+  },
+
+  run: async (_ctx) => {
+    const payload = await seqtaFetchPayload<DocumentCategory[] | null>(
+      "/seqta/student/load/documents",
+      {},
+    );
+    if (!Array.isArray(payload)) return [];
+
+    const items: IndexItem[] = [];
+    const seen = new Set<string>();
+
+    for (const category of payload) {
+      if (!category || !Array.isArray(category.docs)) continue;
+      for (const doc of category.docs) {
+        const uuid = doc.uuid || doc.context_uuid;
+        if (!uuid && !doc.file) continue;
+        const id = `document-${uuid ?? doc.file}`;
+        if (seen.has(id)) continue;
+        seen.add(id);
+
+        const title =
+          doc.title?.trim() ||
+          doc.filename?.trim() ||
+          `Document ${doc.file ?? uuid}`;
+
+        const sizeText = prettySize(doc.size);
+        const mimeLabel = describeMime(doc.mimetype);
+
+        const contentParts: string[] = [];
+        if (doc.filename && doc.filename !== title) contentParts.push(doc.filename);
+        if (category.category) contentParts.push(`Category: ${category.category}`);
+        if (mimeLabel) contentParts.push(mimeLabel);
+        if (sizeText) contentParts.push(sizeText);
+        if (doc.created_date) contentParts.push(`Added ${doc.created_date}`);
+
+        const dateAdded = doc.created_date
+          ? new Date(doc.created_date).getTime() || Date.now()
+          : Date.now();
+
+        items.push({
+          id,
+          text: title,
+          category: "documents",
+          content: contentParts.join(" \u2022 "),
+          dateAdded,
+          metadata: {
+            documentUuid: uuid,
+            fileId: doc.file,
+            filename: doc.filename,
+            mimetype: doc.mimetype,
+            sizeBytes:
+              typeof doc.size === "string" ? parseInt(doc.size, 10) : doc.size,
+            categoryId: category.id,
+            categoryName: category.category,
+            createdDate: doc.created_date,
+            entityType: "document",
+            route: "/documents",
+            icon: "\ueb6f",
+          },
+          actionId: "document",
+          renderComponentId: "document",
+        });
+      }
+    }
+
+    console.debug(`[Documents job] Indexed ${items.length} document entries.`);
+    return items;
+  },
+
+  purge: (items) => items,
+};

src/plugins/built-in/globalSearch/src/indexing/jobs/folio.ts

@@ -0,0 +1,134 @@
+import type { IndexItem, Job } from "../types";
+import { seqtaFetchPayload } from "../api";
+import { htmlToPlainText } from "../utils";
+import { delay } from "@/seqta/utils/delay";
+
+/**
+ * Indexes student folio entries from `/seqta/student/folio`.
+ *
+ * The list mode returns `{ me, list: [{ id, title, published, student }] }`,
+ * and the load mode returns the full body via `{ contents, files, ... }`.
+ * Folio bodies frequently contain `[[embed:raw|<html>]]` blocks which we
+ * normalize to plain text before indexing - the htmlToPlainText sanitizer
+ * never executes scripts because it parses into an inert document.
+ */
+
+interface FolioListPayload {
+  me?: string;
+  list?: Array<{
+    id: number | string;
+    title?: string;
+    published?: string;
+    student?: string;
+  }>;
+}
+
+interface FolioEntryPayload {
+  forum?: number;
+  contents?: string;
+  created?: string;
+  allow_comments?: boolean;
+  author?: { name?: string; year?: string; id?: number };
+  files?: unknown[];
+  id?: number | string;
+  published?: string;
+  title?: string;
+  updated?: string;
+}
+
+const PER_ITEM_DELAY_MS = 80;
+
+function stripEmbedRaw(text: string): string {
+  if (!text) return "";
+  return text.replace(/\[\[embed:raw\|([\s\S]*?)\]\]/g, (_match, inner) => {
+    return htmlToPlainText(typeof inner === "string" ? inner : "");
+  });
+}
+
+export const folioJob: Job = {
+  id: "folio",
+  label: "Folio",
+  renderComponentId: "folio",
+  frequency: { type: "expiry", afterMs: 1000 * 60 * 60 * 24 },
+
+  boostCriteria: (_item, searchTerm) => {
+    if (!searchTerm) return -30;
+    return 0;
+  },
+
+  run: async (ctx) => {
+    const stored = await ctx.getStoredItems("folio");
+    const existing = new Map(stored.map((i) => [i.id, i]));
+
+    const list = await seqtaFetchPayload<FolioListPayload | null>(
+      "/seqta/student/folio",
+      { mode: "list", page: 0, filters: {} },
+    );
+    if (!list || !Array.isArray(list.list)) return [];
+
+    const items: IndexItem[] = [];
+    for (const entry of list.list) {
+      if (!entry || entry.id === undefined) continue;
+      const id = `folio-${entry.id}`;
+      const dateAdded = entry.published
+        ? new Date(entry.published).getTime() || Date.now()
+        : Date.now();
+
+      // If we already have this folio and the title hasn't changed, reuse
+      // the stored content instead of paying for another /folio?mode=load.
+      const existingItem = existing.get(id);
+      const titleChanged = existingItem && existingItem.text !== (entry.title ?? "");
+      if (existingItem && !titleChanged) {
+        items.push({
+          ...existingItem,
+          dateAdded,
+        });
+        continue;
+      }
+
+      try {
+        const detail = await seqtaFetchPayload<FolioEntryPayload | null>(
+          "/seqta/student/folio",
+          { mode: "load", id: entry.id },
+        );
+        const rawContents = detail?.contents ?? "";
+        const flattened = stripEmbedRaw(rawContents);
+        const content = flattened.slice(0, 4000);
+
+        items.push({
+          id,
+          text: entry.title?.trim() || `Folio ${entry.id}`,
+          category: "folio",
+          content,
+          dateAdded,
+          metadata: {
+            folioId: entry.id,
+            student: list.me ?? entry.student,
+            publishedAt: entry.published,
+            updatedAt: detail?.updated,
+            createdAt: detail?.created,
+            authorName: detail?.author?.name,
+            authorId: detail?.author?.id,
+            forumId: detail?.forum,
+            allowComments: detail?.allow_comments,
+            fileCount: Array.isArray(detail?.files) ? detail!.files!.length : 0,
+            entityType: "folio",
+            route: "/folios/read",
+            icon: "\ueb16",
+          },
+          actionId: "folio",
+          renderComponentId: "folio",
+        });
+      } catch (e) {
+        console.warn(`[Folio job] Failed to load folio ${entry.id}:`, e);
+      }
+
+      await delay(PER_ITEM_DELAY_MS);
+    }
+
+    console.debug(`[Folio job] Indexed ${items.length} folio entries.`);
+    return items;
+  },
+
+  purge: (items) => items,
+};

src/plugins/built-in/globalSearch/src/indexing/jobs/goals.ts

@@ -0,0 +1,109 @@
+import type { IndexItem, Job } from "../types";
+import { seqtaFetchPayload } from "../api";
+import { extractTextFromValue } from "../extract";
+import { delay } from "@/seqta/utils/delay";
+
+/**
+ * Indexes student goals from `/seqta/student/load/goals`.
+ *
+ * The endpoint exposes `mode: "years"` which returns the list of available
+ * years and `mode: "list"` (per-year) which returns the actual goals. We
+ * gracefully degrade if the school has goals disabled (the years payload
+ * is empty in that case).
+ */
+
+interface GoalEntry {
+  id?: number | string;
+  uuid?: string;
+  title?: string;
+  description?: string;
+  status?: string;
+  year?: number | string;
+  created?: string;
+  updated?: string;
+}
+
+const PER_YEAR_DELAY_MS = 80;
+
+export const goalsJob: Job = {
+  id: "goals",
+  label: "Goals",
+  renderComponentId: "goal",
+  frequency: { type: "expiry", afterMs: 1000 * 60 * 60 * 24 * 3 }, // every 3 days
+
+  boostCriteria: (_item, searchTerm) => {
+    if (!searchTerm) return -40;
+    return 0;
+  },
+
+  run: async (_ctx) => {
+    const years = await seqtaFetchPayload<Array<string | number> | null>(
+      "/seqta/student/load/goals",
+      { mode: "years" },
+    );
+    if (!Array.isArray(years) || years.length === 0) {
+      console.debug("[Goals job] No goal years available; skipping.");
+      return [];
+    }
+
+    const items: IndexItem[] = [];
+    const seen = new Set<string>();
+
+    for (const year of years) {
+      try {
+        const yearGoals = await seqtaFetchPayload<GoalEntry[] | null>(
+          "/seqta/student/load/goals",
+          { mode: "list", year },
+        );
+        if (!Array.isArray(yearGoals)) continue;
+
+        for (const goal of yearGoals) {
+          if (!goal) continue;
+          const stableId = goal.uuid ?? goal.id;
+          if (stableId === undefined || stableId === null) continue;
+          const id = `goal-${stableId}`;
+          if (seen.has(id)) continue;
+          seen.add(id);
+
+          const title =
+            goal.title?.trim() || goal.description?.slice(0, 80) || `Goal ${stableId}`;
+          const dateAdded = goal.updated || goal.created
+            ? new Date(goal.updated ?? goal.created!).getTime() || Date.now()
+            : Date.now();
+
+          items.push({
+            id,
+            text: title,
+            category: "goals",
+            content: extractTextFromValue(
+              { description: goal.description, status: goal.status },
+              { maxChars: 1000 },
+            ),
+            dateAdded,
+            metadata: {
+              goalId: goal.id,
+              goalUuid: goal.uuid,
+              status: goal.status,
+              year: goal.year ?? year,
+              createdAt: goal.created,
+              updatedAt: goal.updated,
+              entityType: "goal",
+              route: `/goals/${year}`,
+              icon: "\uea15",
+            },
+            actionId: "goal",
+            renderComponentId: "goal",
+          });
+        }
+      } catch (e) {
+        console.warn(`[Goals job] Failed to fetch goals for year ${year}:`, e);
+      }
+      await delay(PER_YEAR_DELAY_MS);
+    }
+
+    console.debug(`[Goals job] Indexed ${items.length} goal entries.`);
+    return items;
+  },
+
+  purge: (items) => items,
+};

src/plugins/built-in/globalSearch/src/indexing/jobs/notices.ts

@@ -0,0 +1,218 @@
+import type { IndexItem, Job } from "../types";
+import { seqtaFetchPayload } from "../api";
+import { htmlToPlainText } from "../utils";
+import { delay } from "@/seqta/utils/delay";
+
+/**
+ * Indexes daily notices from `/seqta/student/load/notices`.
+ *
+ * SEQTA returns notices keyed by date, so we sweep a sliding window
+ * (default: 14 days back) the first time we run, then incrementally pull
+ * the most recent days on subsequent runs. Sensitive routes are excluded
+ * because notices are surfaced for the active student already.
+ */
+
+interface NoticeRecord {
+  id?: number | string;
+  title?: string;
+  contents?: string;
+  staff?: string;
+  staff_id?: number;
+  date?: string;
+  label?: number;
+  label_title?: string;
+  colour?: string;
+}
+
+interface NoticesProgress {
+  earliestDate: string | null;
+  lastSweepBackTo: string | null;
+}
+
+const SWEEP_DAYS = 14;
+const MAX_HISTORY_DAYS = 365;
+const FETCH_DELAY_MS = 60;
+
+function formatYmd(date: Date): string {
+  const y = date.getFullYear();
+  const m = (date.getMonth() + 1).toString().padStart(2, "0");
+  const d = date.getDate().toString().padStart(2, "0");
+  return `${y}-${m}-${d}`;
+}
+
+function parseYmd(value: string | null | undefined): Date | null {
+  if (!value) return null;
+  const match = /^(\d{4})-(\d{2})-(\d{2})$/.exec(value);
+  if (!match) return null;
+  const [, y, m, d] = match;
+  return new Date(Number(y), Number(m) - 1, Number(d));
+}
+
+const fetchNoticesForDate = async (date: string): Promise<NoticeRecord[]> => {
+  const payload = await seqtaFetchPayload<NoticeRecord[] | { notices?: NoticeRecord[] } | null>(
+    "/seqta/student/load/notices",
+    { date },
+  );
+  if (!payload) return [];
+  if (Array.isArray(payload)) return payload;
+  if (Array.isArray((payload as any).notices)) return (payload as any).notices;
+  return [];
+};
+
+const fetchLabelLookup = async (): Promise<Map<number, string>> => {
+  const payload = await seqtaFetchPayload<
+    Array<{ id: number; title?: string }>
+  >("/seqta/student/load/notices", { mode: "labels" });
+  const map = new Map<number, string>();
+  if (Array.isArray(payload)) {
+    for (const entry of payload) {
+      if (entry && typeof entry.id === "number" && entry.title) {
+        map.set(entry.id, entry.title);
+      }
+    }
+  }
+  return map;
+};
+
+export const noticesJob: Job = {
+  id: "notices",
+  label: "Notices",
+  renderComponentId: "notice",
+  frequency: { type: "expiry", afterMs: 1000 * 60 * 60 * 6 }, // 6 hours
+
+  boostCriteria: (item, searchTerm) => {
+    if (!searchTerm) return -10;
+    let score = 0;
+    const ts = item.metadata?.timestamp;
+    if (typeof ts === "string") {
+      const ageDays =
+        (Date.now() - new Date(ts).getTime()) / (1000 * 60 * 60 * 24);
+      if (ageDays >= 0 && ageDays <= 7) score += 0.05;
+    }
+    return score;
+  },
+
+  run: async (ctx) => {
+    const stored = await ctx.getStoredItems("notices");
+    const existingIds = new Set(stored.map((i) => i.id));
+    const progress = (await ctx.getProgress<NoticesProgress>()) ?? {
+      earliestDate: null,
+      lastSweepBackTo: null,
+    };
+
+    const labelLookup = await fetchLabelLookup();
+
+    const today = new Date();
+    today.setHours(0, 0, 0, 0);
+
+    // Sweep window: always the most recent SWEEP_DAYS, plus extend further
+    // back the first time we run until we hit MAX_HISTORY_DAYS.
+    const earliestEverIso = formatYmd(
+      new Date(today.getTime() - MAX_HISTORY_DAYS * 86_400_000),
+    );
+
+    const dates: string[] = [];
+    for (let offset = 0; offset < SWEEP_DAYS; offset++) {
+      const day = new Date(today.getTime() - offset * 86_400_000);
+      dates.push(formatYmd(day));
+    }
+    if (
+      !progress.lastSweepBackTo ||
+      progress.lastSweepBackTo > earliestEverIso
+    ) {
+      // Walk backwards in batches of ~30 days per run so we don't blow up
+      // a single indexing pass.
+      const startBack = parseYmd(progress.lastSweepBackTo) ?? today;
+      const targetBack = new Date(startBack.getTime() - 30 * 86_400_000);
+      const minBack = parseYmd(earliestEverIso) ?? targetBack;
+      const stopBack = targetBack < minBack ? minBack : targetBack;
+      for (
+        let cursor = new Date(startBack.getTime() - SWEEP_DAYS * 86_400_000);
+        cursor >= stopBack;
+        cursor = new Date(cursor.getTime() - 86_400_000)
+      ) {
+        dates.push(formatYmd(cursor));
+      }
+      progress.lastSweepBackTo = formatYmd(stopBack);
+    }
+
+    const items: IndexItem[] = [];
+    const seen = new Set<string>();
+
+    for (const date of dates) {
+      try {
+        const notices = await fetchNoticesForDate(date);
+        for (const notice of notices) {
+          if (!notice || (notice.id === undefined && !notice.title)) continue;
+          const id = `notice-${date}-${notice.id ?? notice.title}`;
+          if (seen.has(id)) continue;
+          seen.add(id);
+
+          const labelTitle =
+            notice.label_title ??
+            (typeof notice.label === "number"
+              ? labelLookup.get(notice.label) ?? null
+              : null);
+
+          const bodyText = notice.contents
+            ? htmlToPlainText(notice.contents)
+            : "";
+
+          items.push({
+            id,
+            text: notice.title?.trim() || `Notice ${notice.id ?? date}`,
+            category: "notices",
+            content: bodyText.slice(0, 4000),
+            dateAdded: new Date(date).getTime(),
+            metadata: {
+              noticeId: notice.id,
+              date,
+              author: notice.staff,
+              authorId: notice.staff_id,
+              label: labelTitle,
+              labelId: notice.label,
+              colour: notice.colour,
+              timestamp: date,
+              entityType: "notice",
+              route: "/notices",
+              icon: "\ueb24",
+            },
+            actionId: "notice",
+            renderComponentId: "notice",
+          });
+        }
+      } catch (e) {
+        console.warn(`[Notices job] Failed to fetch notices for ${date}:`, e);
+      }
+      await delay(FETCH_DELAY_MS);
+    }
+
+    if (items.length > 0) {
+      const dateStrings = items
+        .map((i) => i.metadata?.date as string | undefined)
+        .filter((d): d is string => !!d);
+      if (dateStrings.length > 0) {
+        const earliest = dateStrings.sort()[0];
+        if (
+          !progress.earliestDate ||
+          earliest < progress.earliestDate
+        ) {
+          progress.earliestDate = earliest;
+        }
+      }
+    }
+
+    await ctx.setProgress(progress);
+
+    const newCount = items.filter((i) => !existingIds.has(i.id)).length;
+    console.debug(
+      `[Notices job] Indexed ${items.length} notices across ${dates.length} dates (${newCount} new).`,
+    );
+    return items;
+  },
+
+  purge: (items) => {
+    const oneYearAgo = Date.now() - 365 * 24 * 60 * 60 * 1000;
+    return items.filter((i) => i.dateAdded >= oneYearAgo);
+  },
+};

src/plugins/built-in/globalSearch/src/indexing/jobs/passive.ts

@@ -0,0 +1,49 @@
+import type { Job } from "../types";
+
+/**
+ * Stub job for the passive-observer store.
+ *
+ * The passive observer (see `passiveObserver.ts`) writes captured items
+ * directly into IndexedDB via `getAll`/`put`. We still register a job here
+ * so the indexer:
+ *  - Creates the `passive` object store on first use.
+ *  - Picks up the right `renderComponentId` when materializing in-memory
+ *    items in `loadAllStoredItems()`.
+ *  - Applies a deterministic boost / purge policy to passive results.
+ *
+ * `run()` is a no-op: the passive observer has its own write path so it
+ * works whether or not an active indexing pass is running.
+ */
+export const passiveJob: Job = {
+  id: "passive",
+  label: "Recently viewed",
+  renderComponentId: "passive",
+  // Run frequently so any newly captured items are merged into the
+  // dynamic-items cache on the next indexing tick. The actual capture is
+  // continuous; this is only the synchronization cadence.
+  frequency: { type: "interval", ms: 1000 * 60 * 5 },
+
+  boostCriteria: (item, searchTerm) => {
+    // Passive items are noisier than curated ones, so penalize them
+    // slightly when there's no query and only modestly help on matches.
+    if (!searchTerm) return -60;
+    let score = 0;
+    if (item.metadata?.entityType) score += 0.02;
+    return score;
+  },
+
+  run: async () => {
+    return [];
+  },
+
+  purge: (items) => {
+    // Keep the most recent ~500 passive entries and anything newer than
+    // 30 days. This caps storage growth from heavy browsing sessions.
+    const cutoff = Date.now() - 30 * 24 * 60 * 60 * 1000;
+    const recent = items
+      .filter((i) => i.dateAdded >= cutoff)
+      .sort((a, b) => b.dateAdded - a.dateAdded)
+      .slice(0, 500);
+    return recent;
+  },
+};

src/plugins/built-in/globalSearch/src/indexing/jobs/portals.ts

@@ -0,0 +1,90 @@
+import type { IndexItem, Job } from "../types";
+import { seqtaFetchPayload } from "../api";
+
+/**
+ * Indexes the user's external portal entries from `/seqta/student/load/portals`.
+ *
+ * Portals are user-facing tiles linking to third-party tools (Mathletics,
+ * Seesaw, Google Classroom, ...). We index their labels and external URLs
+ * so users can jump to them via the global search palette without scrolling
+ * the dashboard.
+ */
+
+interface PortalPayload {
+  id: number | string;
+  label?: string;
+  url?: string;
+  uuid?: string;
+  icon?: string;
+  priority?: number;
+  is_power_portal?: boolean;
+  contents?: string;
+  inherit_styles?: boolean;
+}
+
+function normalizePortalUrl(raw: string | undefined): string | undefined {
+  if (!raw) return undefined;
+  const trimmed = raw.trim();
+  if (!trimmed) return undefined;
+  if (/^https?:\/\//i.test(trimmed)) return trimmed;
+  return `https://${trimmed.replace(/^\/+/, "")}`;
+}
+
+export const portalsJob: Job = {
+  id: "portals",
+  label: "Portals",
+  renderComponentId: "portal",
+  frequency: { type: "expiry", afterMs: 1000 * 60 * 60 * 24 * 7 }, // weekly
+
+  boostCriteria: (_item, searchTerm) => {
+    if (!searchTerm) return -50;
+    return 0;
+  },
+
+  run: async (_ctx) => {
+    const payload = await seqtaFetchPayload<PortalPayload[] | null>(
+      "/seqta/student/load/portals",
+      {},
+    );
+    if (!Array.isArray(payload)) return [];
+
+    const items: IndexItem[] = [];
+    const seen = new Set<string>();
+
+    for (const portal of payload) {
+      if (!portal || (portal.id === undefined && !portal.uuid)) continue;
+      const id = `portal-${portal.uuid ?? portal.id}`;
+      if (seen.has(id)) continue;
+      seen.add(id);
+
+      const url = normalizePortalUrl(portal.url);
+      const label = portal.label?.trim() || `Portal ${portal.id}`;
+      const contentParts: string[] = [];
+      if (url) contentParts.push(url);
+      if (portal.is_power_portal) contentParts.push("Power Portal");
+
+      items.push({
+        id,
+        text: label,
+        category: "portals",
+        content: contentParts.join(" \u2022 "),
+        dateAdded: Date.now(),
+        metadata: {
+          portalId: portal.id,
+          portalUuid: portal.uuid,
+          url,
+          isPowerPortal: !!portal.is_power_portal,
+          entityType: "portal",
+          icon: "\ueb01",
+        },
+        actionId: "portal",
+        renderComponentId: "portal",
+      });
+    }
+
+    console.debug(`[Portals job] Indexed ${items.length} portal entries.`);
+    return items;
+  },
+
+  purge: (items) => items,
+};

src/plugins/built-in/globalSearch/src/indexing/jobs/reports.ts

@@ -0,0 +1,97 @@
+import type { IndexItem, Job } from "../types";
+import { seqtaFetchPayload } from "../api";
+
+/**
+ * Indexes report metadata from `/seqta/student/load/reports`.
+ *
+ * Reports are PDFs gated behind SEQTA's authenticated download endpoint, so
+ * we only index the human-readable metadata (year, term, title, file UUID)
+ * and a stable hash route so the search palette can deep-link straight
+ * into the reports page.
+ */
+
+interface ReportEntry {
+  id?: number | string;
+  uuid?: string;
+  title?: string;
+  description?: string;
+  date_published?: string;
+  date_created?: string;
+  year?: number | string;
+  term?: number | string;
+  metaclass?: number;
+  programme?: number;
+  filename?: string;
+}
+
+export const reportsJob: Job = {
+  id: "reports",
+  label: "Reports",
+  renderComponentId: "report",
+  frequency: { type: "expiry", afterMs: 1000 * 60 * 60 * 24 }, // daily
+
+  boostCriteria: (_item, searchTerm) => {
+    if (!searchTerm) return -25;
+    return 0;
+  },
+
+  run: async (_ctx) => {
+    const payload = await seqtaFetchPayload<ReportEntry[] | null>(
+      "/seqta/student/load/reports",
+      {},
+    );
+    if (!Array.isArray(payload)) return [];
+
+    const items: IndexItem[] = [];
+    const seen = new Set<string>();
+
+    for (const report of payload) {
+      if (!report) continue;
+      const stableId = report.uuid ?? report.id;
+      if (stableId === undefined || stableId === null) continue;
+      const id = `report-${stableId}`;
+      if (seen.has(id)) continue;
+      seen.add(id);
+
+      const title = report.title?.trim() || `Report ${stableId}`;
+      const dateAdded = report.date_published
+        ? new Date(report.date_published).getTime() || Date.now()
+        : Date.now();
+
+      const contentParts: string[] = [];
+      if (report.description) contentParts.push(report.description);
+      if (report.year) contentParts.push(`Year ${report.year}`);
+      if (report.term) contentParts.push(`Term ${report.term}`);
+      if (report.date_published) contentParts.push(report.date_published);
+
+      items.push({
+        id,
+        text: title,
+        category: "reports",
+        content: contentParts.join(" \u2022 "),
+        dateAdded,
+        metadata: {
+          reportId: report.id,
+          reportUuid: report.uuid,
+          year: report.year,
+          term: report.term,
+          metaclass: report.metaclass,
+          programme: report.programme,
+          publishedAt: report.date_published,
+          createdAt: report.date_created,
+          filename: report.filename,
+          entityType: "report",
+          route: "/reports",
+          icon: "\ueb70",
+        },
+        actionId: "report",
+        renderComponentId: "report",
+      });
+    }
+
+    console.debug(`[Reports job] Indexed ${items.length} reports.`);
+    return items;
+  },
+
+  purge: (items) => items,
+};

src/plugins/built-in/globalSearch/src/indexing/passiveObserver.ts

@@ -0,0 +1,632 @@
+import type { IndexItem } from "./types";
+import { put, getAll } from "./db";
+import {
+  buildIndexItem,
+  extractTextFromValue,
+  pickId,
+  pickTitle,
+} from "./extract";
+import { isSensitiveSeqtaPath, normalizeSeqtaPath } from "./api";
+import { loadAllStoredItems } from "./indexer";
+import { loadDynamicItems } from "../utils/dynamicItems";
+import { renderComponentMap } from "./renderComponents";
+import { jobs } from "./jobs";
+
+/**
+ * Passive network observer.
+ *
+ * Wraps the page's `fetch` (and best-effort `XMLHttpRequest`) so that any
+ * successful same-origin SEQTA JSON response observed while the user
+ * browses is opportunistically distilled into IndexItems and persisted to
+ * the `passive` object store.
+ *
+ * Hard guarantees:
+ *  - Only same-origin requests under `/seqta/student/` are considered.
+ *  - The shared sensitive-route denylist (login, save/*, settings, prefs,
+ *    heartbeat, branding, themes, file streams, masquerade, ...) is checked
+ *    before any persistence.
+ *  - Response bodies are read via `Response.clone()` so we never consume the
+ *    body the host page intends to use.
+ *  - Sensitive keys/values are stripped via `redactSensitive` before the
+ *    item is stored.
+ *  - Binary file contents are never indexed (we only work on JSON responses
+ *    served as `text/json` / `application/json`).
+ */
+
+const STORE_ID = "passive";
+const FLUSH_DEBOUNCE_MS = 1500;
+const MAX_ITEMS_PER_RESPONSE = 50;
+const MAX_PER_RESPONSE_TEXT_CHARS = 1500;
+
+let installed = false;
+let pendingFlush: ReturnType<typeof setTimeout> | null = null;
+let pendingDirty = false;
+
+export function isPassiveObserverInstalled(): boolean {
+  return installed;
+}
+
+/* ------------------------------------------------------------------ */
+/*                          eligibility checks                        */
+/* ------------------------------------------------------------------ */
+
+function isSameOriginSeqtaUrl(url: string): boolean {
+  try {
+    const parsed = new URL(url, location.origin);
+    if (parsed.origin !== location.origin) return false;
+    return parsed.pathname.startsWith("/seqta/student/");
+  } catch {
+    return false;
+  }
+}
+
+function looksLikeJsonContentType(contentType: string | null): boolean {
+  if (!contentType) return false;
+  return /json/i.test(contentType);
+}
+
+/* ------------------------------------------------------------------ */
+/*                        item synthesis                              */
+/* ------------------------------------------------------------------ */
+
+interface CapturedContext {
+  route: string;
+  requestBody: unknown;
+  observedAt: number;
+}
+
+function categoryFromRoute(route: string): string {
+  // /seqta/student/load/courses -> courses
+  // /seqta/student/load/message -> message
+  const tail = route.replace(/^\/seqta\/student\//, "").split("/").filter(Boolean);
+  if (tail.length === 0) return "passive";
+  // message/people is a support endpoint that backs the messages compose UI.
+  // We treat it as a low-priority `messages-support` record rather than a
+  // standalone "people" category so it never competes with real assessments
+  // / messages in the result list.
+  if (route.includes("/load/message/people")) return "messages-support";
+  return tail[tail.length - 1].toLowerCase();
+}
+
+/**
+ * `/seqta/student/load/message/people` returns the contact picker dataset
+ * used by the messages compose view. We only want to surface entries that
+ * actually carry a human display name — the rest is structural noise that
+ * historically caused raw API paths to appear as titles.
+ */
+function isPeopleEntityWorthIndexing(entity: unknown): boolean {
+  if (!entity || typeof entity !== "object") return false;
+  const obj = entity as Record<string, unknown>;
+  const first = stringField(obj, [
+    "preferredName",
+    "preferred",
+    "firstname",
+    "firstName",
+    "first_name",
+    "given",
+    "givenName",
+  ]);
+  const last = stringField(obj, [
+    "surname",
+    "lastname",
+    "lastName",
+    "last_name",
+    "familyName",
+  ]);
+  const display = stringField(obj, ["displayName", "name", "fullName"]);
+  return Boolean((first && last) || display);
+}
+
+function sourcePageForRoute(route: string): string | undefined {
+  if (route.includes("/load/message/people")) return "/messages";
+  if (route.includes("/load/message")) return "/messages";
+  if (route.includes("/load/messages")) return "/messages";
+  if (route.includes("/load/courses")) return "/courses";
+  if (route.includes("/load/assessments")) return "/assessments/upcoming";
+  if (route.includes("/load/notices")) return "/notices";
+  if (route.includes("/load/documents")) return "/documents";
+  if (route.includes("/folio")) return "/folios/read";
+  if (route.includes("/load/forums")) return "/forums";
+  if (route.includes("/load/goals")) return "/goals";
+  if (route.includes("/load/reports")) return "/reports";
+  if (route.includes("/load/portals")) return "/dashboard";
+  return undefined;
+}
+
+/** Programme + metaclass for `/load/courses` POST body or embedded course JSON. */
+function extractProgrammeMetaclass(
+  requestBody: unknown,
+  entity: unknown,
+): { programme: number; metaclass: number } | null {
+  const coerce = (value: unknown): number | undefined => {
+    if (typeof value === "number" && Number.isFinite(value)) return value;
+    if (typeof value === "string") {
+      const t = value.trim();
+      if (!t) return undefined;
+      const n = Number(t);
+      return Number.isFinite(n) ? n : undefined;
+    }
+    return undefined;
+  };
+
+  const read = (
+    src: Record<string, unknown> | null,
+  ): { programme: number; metaclass: number } | null => {
+    if (!src) return null;
+    const programme = coerce(
+      src.programme ?? src.programmeId ?? src.programmeID,
+    );
+    const metaclass = coerce(
+      src.metaclass ?? src.metaclassId ?? src.metaclassID ?? src.subjectId,
+    );
+    if (programme !== undefined && metaclass !== undefined) {
+      return { programme, metaclass };
+    }
+    return null;
+  };
+
+  if (requestBody && typeof requestBody === "object" && !Array.isArray(requestBody)) {
+    const r = read(requestBody as Record<string, unknown>);
+    if (r) return r;
+  }
+  if (entity && typeof entity === "object" && !Array.isArray(entity)) {
+    const r = read(entity as Record<string, unknown>);
+    if (r) return r;
+  }
+  return null;
+}
+
+function entitiesFromPayload(payload: unknown): unknown[] {
+  if (Array.isArray(payload)) return payload;
+  if (payload && typeof payload === "object") {
+    const obj = payload as Record<string, unknown>;
+    // SEQTA frequently nests arrays as `payload.list`, `.messages`,
+    // `.items`, `.tasks`, etc. Pull the first array-shaped child as our
+    // best guess; if none exists, fall back to the object itself so we
+    // still index a single entry.
+    for (const key of [
+      "list",
+      "items",
+      "messages",
+      "tasks",
+      "pending",
+      "forums",
+      "docs",
+    ]) {
+      const value = obj[key];
+      if (Array.isArray(value)) return value;
+    }
+    return [payload];
+  }
+  return [];
+}
+
+/**
+ * Whitelist of entity-shaped fields we hoist into item metadata so the
+ * `passive` action handler can deep-link into the right SEQTA SPA route.
+ * These mirror what the active jobs already store (see `courses.ts`,
+ * `portals.ts`, etc.) so the action only has to consult one source.
+ */
+const DEEP_LINK_FIELDS = [
+  "programme",
+  "programmeId",
+  "programmeID",
+  "metaclass",
+  "metaclassId",
+  "metaclassID",
+  "year",
+  "uuid",
+  "portalUuid",
+  "forum",
+  "forumId",
+  "assessmentId",
+  "assessmentID",
+  "messageId",
+] as const;
+
+function pickDeepLinkHints(
+  entity: unknown,
+): Record<string, string | number> {
+  if (!entity || typeof entity !== "object") return {};
+  const src = entity as Record<string, unknown>;
+  const out: Record<string, string | number> = {};
+  for (const key of DEEP_LINK_FIELDS) {
+    const value = src[key];
+    if (typeof value === "number" && Number.isFinite(value)) {
+      out[key] = value;
+    } else if (typeof value === "string" && value) {
+      out[key] = value;
+    }
+  }
+  return out;
+}
+
+function stringField(
+  entity: Record<string, unknown>,
+  keys: readonly string[],
+): string | undefined {
+  for (const key of keys) {
+    const value = entity[key];
+    if (typeof value === "string" && value.trim()) return value.trim();
+  }
+  return undefined;
+}
+
+function titleFromEndpoint(
+  route: string,
+  entity: unknown,
+  extractedContent: string,
+  fallback: string,
+): string {
+  if (route.includes("/load/message/people") && entity && typeof entity === "object") {
+    const obj = entity as Record<string, unknown>;
+    const first = stringField(obj, [
+      "preferredName",
+      "preferred",
+      "firstname",
+      "firstName",
+      "first_name",
+      "given",
+      "givenName",
+    ]);
+    const last = stringField(obj, [
+      "surname",
+      "lastname",
+      "lastName",
+      "last_name",
+      "familyName",
+    ]);
+    const full = [first, last].filter(Boolean).join(" ").trim();
+    if (full) return full.slice(0, 200);
+  }
+
+  const picked = pickTitle(entity, "");
+  if (picked) return picked.slice(0, 200);
+
+  // Last resort: show a human-readable content preview instead of a raw API
+  // path like `/seqta/student/load/message/people#20`.
+  const firstLine = extractedContent
+    .split(/\r?\n/)
+    .map((line) => line.trim())
+    .find(Boolean);
+  return (firstLine || fallback).slice(0, 200);
+}
+
+function synthesizeItems(
+  ctx: CapturedContext,
+  payload: unknown,
+): IndexItem[] {
+  const entities = entitiesFromPayload(payload);
+  if (entities.length === 0) return [];
+
+  const category = categoryFromRoute(ctx.route);
+  const now = ctx.observedAt;
+  const out: IndexItem[] = [];
+
+  const isPeopleSupport = ctx.route.includes("/load/message/people");
+
+  const limit = Math.min(entities.length, MAX_ITEMS_PER_RESPONSE);
+  for (let i = 0; i < limit; i++) {
+    const entity = entities[i];
+    if (!entity || (typeof entity !== "object" && typeof entity !== "string")) {
+      continue;
+    }
+
+    // For the messages compose-people endpoint, skip records that don't
+    // carry a real human name. We never want raw entries like
+    // `/seqta/student/load/message/people#20` becoming titles, and we
+    // explicitly route the rest to /messages so they're treated as support
+    // records, not standalone "people" results.
+    if (isPeopleSupport && !isPeopleEntityWorthIndexing(entity)) {
+      continue;
+    }
+
+    const fallbackId = `${ctx.route}#${i}`;
+    const entityId = pickId(entity, fallbackId);
+    const stableId = `passive-${ctx.route.replace(/\//g, "_")}-${entityId}`;
+
+    const content = extractTextFromValue(entity, {
+      maxChars: MAX_PER_RESPONSE_TEXT_CHARS,
+    });
+    const title = titleFromEndpoint(ctx.route, entity, content, fallbackId);
+    if (!content && (!title || title === fallbackId)) {
+      // Skip records that produced neither title nor content; they are
+      // structurally noise (e.g. tiny acknowledgement payloads).
+      continue;
+    }
+
+    const deepLinkHints = pickDeepLinkHints(entity);
+    const sourcePage = sourcePageForRoute(ctx.route);
+    const coursePm = ctx.route.includes("/load/courses")
+      ? extractProgrammeMetaclass(ctx.requestBody, entity)
+      : null;
+
+    out.push(
+      buildIndexItem({
+        id: stableId,
+        text: title,
+        category,
+        contentOverride: content,
+        metadata: {
+          route: ctx.route,
+          source: "passive",
+          observedAt: new Date(now).toISOString(),
+          entityType: category,
+          entityId,
+          icon: "\ueb71",
+          sourcePage,
+          // Mark message/people as a low-priority support record so the
+          // search ranker can deprioritize it relative to real messages,
+          // assessments, courses, etc.
+          ...(isPeopleSupport ? { supportRecord: true, priority: "low" } : {}),
+          ...deepLinkHints,
+          ...(coursePm
+            ? { programme: coursePm.programme, metaclass: coursePm.metaclass }
+            : {}),
+        },
+        actionId: "passive",
+        renderComponentId: "passive",
+        dateAdded: now,
+      }),
+    );
+  }
+
+  return out;
+}
+
+/* ------------------------------------------------------------------ */
+/*                          persistence                               */
+/* ------------------------------------------------------------------ */
+
+async function persistItems(items: IndexItem[]): Promise<void> {
+  if (items.length === 0) return;
+
+  // Dedupe against existing entries. We replace on collision so the latest
+  // observation wins (e.g. if a message changes title).
+  for (const item of items) {
+    try {
+      await put(STORE_ID, item, item.id);
+    } catch (e) {
+      console.warn(
+        `[Passive Observer] Failed to persist item ${item.id}:`,
+        e,
+      );
+    }
+  }
+
+  pendingDirty = true;
+  scheduleFlush();
+}
+
+function scheduleFlush() {
+  if (pendingFlush) return;
+  pendingFlush = setTimeout(() => {
+    pendingFlush = null;
+    if (!pendingDirty) return;
+    pendingDirty = false;
+    void flushDynamicItems();
+  }, FLUSH_DEBOUNCE_MS);
+}
+
+async function flushDynamicItems(): Promise<void> {
+  try {
+    const all = await loadAllStoredItems();
+    const decorated = all.map((item) => {
+      try {
+        const jobDef =
+          jobs[item.category] ||
+          Object.values(jobs).find((j) => j.id === item.category) ||
+          jobs[item.renderComponentId];
+        let renderComponent = item.renderComponent;
+        if (jobDef) {
+          renderComponent =
+            renderComponentMap[jobDef.renderComponentId] || renderComponent;
+        } else if (renderComponentMap[item.renderComponentId]) {
+          renderComponent = renderComponentMap[item.renderComponentId];
+        }
+        try {
+          const cloned = JSON.parse(JSON.stringify(item));
+          cloned.renderComponent = renderComponent;
+          return cloned;
+        } catch {
+          return { ...item, renderComponent };
+        }
+      } catch {
+        return item;
+      }
+    });
+    loadDynamicItems(decorated);
+    window.dispatchEvent(
+      new CustomEvent("dynamic-items-updated", {
+        detail: {
+          incremental: true,
+          jobId: STORE_ID,
+          streaming: false,
+        },
+      }),
+    );
+  } catch (e) {
+    console.warn("[Passive Observer] Failed to refresh dynamic items:", e);
+  }
+}
+
+/* ------------------------------------------------------------------ */
+/*                          fetch hook                                */
+/* ------------------------------------------------------------------ */
+
+async function consumeResponse(
+  response: Response,
+  url: string,
+  requestBody: unknown,
+): Promise<void> {
+  if (!response.ok) return;
+
+  const route = normalizeSeqtaPath(url);
+  if (isSensitiveSeqtaPath(route)) return;
+
+  const contentType = response.headers.get("content-type");
+  if (!looksLikeJsonContentType(contentType)) return;
+
+  let body: any;
+  try {
+    body = await response.clone().json();
+  } catch {
+    return;
+  }
+
+  if (!body || typeof body !== "object") return;
+  if (body.status && body.status !== "200") return;
+
+  const payload = body.payload;
+  if (payload === undefined || payload === null) return;
+
+  const items = synthesizeItems(
+    {
+      route,
+      requestBody,
+      observedAt: Date.now(),
+    },
+    payload,
+  );
+
+  if (items.length > 0) {
+    await persistItems(items);
+  }
+}
+
+function tryParseJson(value: unknown): unknown {
+  if (typeof value !== "string") return value;
+  try {
+    return JSON.parse(value);
+  } catch {
+    return value;
+  }
+}
+
+/**
+ * Installs the passive observer once. Subsequent calls are no-ops.
+ *
+ * Designed to be called from the global-search plugin bootstrap after
+ * `mountSearchBar` succeeds so the observer is only active when the
+ * plugin itself is enabled.
+ */
+export function installPassiveObserver(): void {
+  if (installed) return;
+  if (typeof window === "undefined" || typeof window.fetch !== "function") {
+    return;
+  }
+  installed = true;
+
+  const originalFetch = window.fetch.bind(window);
+  window.fetch = async function patchedFetch(
+    input: RequestInfo | URL,
+    init?: RequestInit,
+  ): Promise<Response> {
+    const response = await originalFetch(input, init);
+
+    try {
+      const url =
+        typeof input === "string"
+          ? input
+          : input instanceof URL
+            ? input.toString()
+            : input.url;
+      if (isSameOriginSeqtaUrl(url)) {
+        const body = init?.body;
+        const parsed =
+          body && typeof body === "string"
+            ? tryParseJson(body)
+            : undefined;
+        // Fire-and-forget: never block the host page on indexing work.
+        void consumeResponse(response, url, parsed);
+      }
+    } catch (e) {
+      // Never let observer errors bubble up to the host page.
+      console.debug("[Passive Observer] fetch hook error:", e);
+    }
+
+    return response;
+  };
+
+  // Best-effort XHR hook for the rare callers that bypass fetch.
+  const ProtoXhr = (window as any).XMLHttpRequest?.prototype;
+  if (ProtoXhr) {
+    const originalOpen = ProtoXhr.open;
+    const originalSend = ProtoXhr.send;
+    ProtoXhr.open = function patchedOpen(
+      this: XMLHttpRequest,
+      method: string,
+      url: string,
+      ...rest: any[]
+    ) {
+      try {
+        (this as any).__bsplusUrl = url;
+        (this as any).__bsplusMethod = method;
+      } catch {
+        /* ignore */
+      }
+      return originalOpen.call(this, method, url, ...rest);
+    };
+    ProtoXhr.send = function patchedSend(
+      this: XMLHttpRequest,
+      body?: any,
+    ) {
+      try {
+        const url = (this as any).__bsplusUrl as string | undefined;
+        if (url && isSameOriginSeqtaUrl(url)) {
+          const parsed =
+            typeof body === "string" ? tryParseJson(body) : undefined;
+          this.addEventListener("load", () => {
+            try {
+              if (this.status < 200 || this.status >= 300) return;
+              const ct = this.getResponseHeader("content-type");
+              if (!looksLikeJsonContentType(ct)) return;
+              const route = normalizeSeqtaPath(url);
+              if (isSensitiveSeqtaPath(route)) return;
+              let json: any;
+              try {
+                json = JSON.parse(this.responseText);
+              } catch {
+                return;
+              }
+              if (!json || typeof json !== "object") return;
+              if (json.status && json.status !== "200") return;
+              const payload = json.payload;
+              if (payload === undefined || payload === null) return;
+              const items = synthesizeItems(
+                {
+                  route,
+                  requestBody: parsed,
+                  observedAt: Date.now(),
+                },
+                payload,
+              );
+              if (items.length > 0) {
+                void persistItems(items);
+              }
+            } catch (e) {
+              console.debug("[Passive Observer] xhr load error:", e);
+            }
+          });
+        }
+      } catch {
+        /* ignore */
+      }
+      return originalSend.call(this, body);
+    };
+  }
+
+  console.debug("[Passive Observer] Installed.");
+}
+
+/**
+ * Returns currently-stored passive items. Mainly used for diagnostics from
+ * `window.globalSearchDebug`.
+ */
+export async function getStoredPassiveItems(): Promise<IndexItem[]> {
+  try {
+    return (await getAll(STORE_ID)) as IndexItem[];
+  } catch {
+    return [];
+  }
+}

src/plugins/built-in/globalSearch/src/indexing/renderComponents.ts

@@ -2,10 +2,23 @@ import type { SvelteComponent } from "svelte";
 import AssessmentItem from "../components/items/AssessmentItem.svelte";
 import ForumItem from "../components/items/ForumItem.svelte";
 import SubjectItem from "../components/items/SubjectItem.svelte";
+import GenericItem from "../components/items/GenericItem.svelte";
 
 export const renderComponentMap: Record<string, typeof SvelteComponent> = {
   assessment: AssessmentItem as unknown as typeof SvelteComponent,
   message: AssessmentItem as unknown as typeof SvelteComponent,
   forum: ForumItem as unknown as typeof SvelteComponent,
   subject: SubjectItem as unknown as typeof SvelteComponent,
+  // New categories share a generic, category-aware row component to keep
+  // the palette consistent without bespoke layouts for every job. The
+  // component reads `item.metadata.icon` and the `category` to pick a
+  // sensible default visual treatment.
+  course: GenericItem as unknown as typeof SvelteComponent,
+  notice: GenericItem as unknown as typeof SvelteComponent,
+  document: GenericItem as unknown as typeof SvelteComponent,
+  folio: GenericItem as unknown as typeof SvelteComponent,
+  portal: GenericItem as unknown as typeof SvelteComponent,
+  report: GenericItem as unknown as typeof SvelteComponent,
+  goal: GenericItem as unknown as typeof SvelteComponent,
+  passive: GenericItem as unknown as typeof SvelteComponent,
 };

src/plugins/built-in/globalSearch/src/indexing/resetIndexes.ts

@@ -0,0 +1,112 @@
+import { SCHEMA_VERSION_KEY } from "./schemaVersion";
+
+/**
+ * Hard-reset of all global-search persistence.
+ *
+ * This module is intentionally dependency-free (no imports from `db.ts`,
+ * the worker manager, embeddia, or any heavy bundle) so it can be
+ * statically imported from:
+ *
+ *  - The always-loaded plugin shell (`lazy.ts`) for the manual
+ *    "Reset Index" settings button. Statically importing means the button
+ *    keeps working across extension updates — there's no chunk hash to
+ *    chase via dynamic import, which previously produced
+ *    `Failed to fetch dynamically imported module: .../assets/<chunk>.js`
+ *    when an older settings page tried to load a chunk that the new build
+ *    had already replaced.
+ *
+ *  - The version-check path (`utils/versionCheck.ts`) for the auto-reset
+ *    that fires whenever the extension's manifest version changes.
+ *
+ * The function:
+ *   1. Notifies in-process modules to drop in-memory caches and any open
+ *      IndexedDB connections via custom DOM events (best effort).
+ *   2. Deletes the structured `betterseqta-index` and the vector
+ *      `embeddiaDB` databases.
+ *   3. Clears version-tracking localStorage keys so the next indexing
+ *      pass treats the world as fresh.
+ *
+ * It never throws on partial failure: each step is wrapped in try/catch
+ * so a stuck connection on one DB doesn't block the other.
+ */
+
+const STRUCTURED_DB = "betterseqta-index";
+const VECTOR_DB = "embeddiaDB";
+const STRUCTURED_VERSION_KEY = "betterseqta-index-version";
+
+function deleteIndexedDb(name: string): Promise<void> {
+  return new Promise((resolve) => {
+    let resolved = false;
+    const finish = () => {
+      if (resolved) return;
+      resolved = true;
+      resolve();
+    };
+
+    let req: IDBOpenDBRequest;
+    try {
+      req = indexedDB.deleteDatabase(name);
+    } catch (e) {
+      console.warn(`[Reset] Could not start delete of ${name}:`, e);
+      finish();
+      return;
+    }
+
+    req.onsuccess = () => finish();
+    req.onerror = () => {
+      console.warn(`[Reset] Error deleting ${name}:`, req.error);
+      finish();
+    };
+    req.onblocked = () => {
+      // Connections are still open in another tab. Wait briefly, retry,
+      // then resolve regardless so we never hang the caller forever.
+      console.warn(
+        `[Reset] Delete of ${name} blocked; will retry then resolve.`,
+      );
+      setTimeout(() => {
+        try {
+          const retry = indexedDB.deleteDatabase(name);
+          retry.onsuccess = () => finish();
+          retry.onerror = () => finish();
+          retry.onblocked = () => finish();
+        } catch {
+          finish();
+        }
+      }, 600);
+    };
+  });
+}
+
+export async function resetSearchIndexes(): Promise<void> {
+  try {
+    if (typeof window !== "undefined") {
+      window.dispatchEvent(
+        new CustomEvent("betterseqta-clear-search-cache"),
+      );
+      window.dispatchEvent(
+        new CustomEvent("betterseqta-clear-embedding-cache"),
+      );
+      window.dispatchEvent(
+        new CustomEvent("betterseqta-reset-search-index"),
+      );
+    }
+  } catch {
+    /* ignore — events are best-effort */
+  }
+
+  // Give listeners a tick to close any open IDB connections; otherwise
+  // the delete request below comes back with `onblocked`.
+  await new Promise<void>((resolve) => setTimeout(resolve, 150));
+
+  await Promise.allSettled([
+    deleteIndexedDb(STRUCTURED_DB),
+    deleteIndexedDb(VECTOR_DB),
+  ]);
+
+  try {
+    localStorage.removeItem(STRUCTURED_VERSION_KEY);
+    localStorage.removeItem(SCHEMA_VERSION_KEY);
+  } catch {
+    /* ignore */
+  }
+}

src/plugins/built-in/globalSearch/src/indexing/schemaVersion.ts

@@ -0,0 +1,16 @@
+/**
+ * Index schema version. Bump whenever the IndexItem shape, category set,
+ * or text construction changes in a way that should invalidate previously
+ * stored items (and their embeddings).
+ *
+ * On mismatch, both the structured IndexedDB store and the embeddiaDB are
+ * wiped before the next indexing pass so we don't serve stale results.
+ *
+ * Kept in its own file (with no imports) so very lightweight callers — the
+ * always-loaded plugin shell in `lazy.ts`, the version-check path — can
+ * pull it in without bringing the heavy indexer/worker bundle along.
+ */
+export const INDEX_SCHEMA_VERSION = 6;
+
+/** Key used to track the schema version a previous run wrote out. */
+export const SCHEMA_VERSION_KEY = "bsq-index-schema-version";

src/plugins/built-in/globalSearch/src/indexing/selfTests.ts

@@ -0,0 +1,328 @@
+import {
+  isSensitiveKey,
+  looksLikeSecretValue,
+  redactSensitive,
+  extractTextFromValue,
+  pickTitle,
+  pickId,
+  buildIndexItem,
+} from "./extract";
+import { isSensitiveSeqtaPath, normalizeSeqtaPath } from "./api";
+import {
+  coursesPayload,
+  documentsPayload,
+  folioEntryPayload,
+  noticesPayload,
+  portalsPayload,
+  settingsPayload,
+  subjectsListPayload,
+} from "./__fixtures__/seqtaResponses";
+
+/**
+ * Lightweight in-process self-tests for the global-search overhaul.
+ *
+ * The repository does not (yet) ship with a test runner, so we instead
+ * expose a deterministic suite of assertions over the pure helpers that
+ * back active jobs and the passive observer. This is intentionally
+ * dependency-free so it can run inside the extension page (`window.
+ * globalSearchDebug.runSelfTests()`) and from any future Vitest harness
+ * without modification.
+ */
+
+interface TestCase {
+  name: string;
+  run: () => void | Promise<void>;
+}
+
+class AssertionError extends Error {
+  constructor(message: string) {
+    super(message);
+    this.name = "AssertionError";
+  }
+}
+
+function assert(condition: unknown, message: string): asserts condition {
+  if (!condition) throw new AssertionError(message);
+}
+
+function assertEqual<T>(actual: T, expected: T, label: string) {
+  if (actual !== expected) {
+    throw new AssertionError(
+      `${label}: expected ${JSON.stringify(expected)} but got ${JSON.stringify(actual)}`,
+    );
+  }
+}
+
+function assertContains(haystack: string, needle: string, label: string) {
+  if (!haystack.includes(needle)) {
+    throw new AssertionError(
+      `${label}: expected "${haystack}" to contain "${needle}"`,
+    );
+  }
+}
+
+function assertNotContains(haystack: string, needle: string, label: string) {
+  if (haystack.includes(needle)) {
+    throw new AssertionError(
+      `${label}: expected "${haystack}" NOT to contain "${needle}"`,
+    );
+  }
+}
+
+const cases: TestCase[] = [
+  {
+    name: "normalizeSeqtaPath strips query tokens",
+    run: () => {
+      assertEqual(
+        normalizeSeqtaPath("/seqta/student/load/messages?mokx3qef"),
+        "/seqta/student/load/messages",
+        "trailing token",
+      );
+      assertEqual(
+        normalizeSeqtaPath(
+          "https://learn.example.com/seqta/student/load/courses?abc123",
+        ),
+        "/seqta/student/load/courses",
+        "absolute URL",
+      );
+    },
+  },
+  {
+    name: "isSensitiveSeqtaPath catches credential routes",
+    run: () => {
+      assert(
+        isSensitiveSeqtaPath("/seqta/student/login?xyz"),
+        "login is sensitive",
+      );
+      assert(
+        isSensitiveSeqtaPath("/seqta/student/save/message"),
+        "save/* is sensitive",
+      );
+      assert(
+        isSensitiveSeqtaPath("/seqta/student/load/settings"),
+        "settings is sensitive",
+      );
+      assert(
+        isSensitiveSeqtaPath("/seqta/student/load/prefs?z=1"),
+        "prefs is sensitive",
+      );
+      assert(
+        isSensitiveSeqtaPath("/seqta/ta/masquerade"),
+        "masquerade is sensitive",
+      );
+      assert(
+        !isSensitiveSeqtaPath("/seqta/student/load/messages"),
+        "messages is NOT sensitive",
+      );
+      assert(
+        !isSensitiveSeqtaPath("/seqta/student/load/courses"),
+        "courses is NOT sensitive",
+      );
+    },
+  },
+  {
+    name: "isSensitiveKey covers the credential vocabulary",
+    run: () => {
+      for (const key of [
+        "password",
+        "Password",
+        "client_secret",
+        "apiKey",
+        "X-API-Token",
+        "jwtSession",
+        "oauth_signature",
+      ]) {
+        assert(isSensitiveKey(key), `expected ${key} to be sensitive`);
+      }
+      for (const key of ["title", "subject", "uuid", "metaclass"]) {
+        assert(!isSensitiveKey(key), `expected ${key} to be safe`);
+      }
+    },
+  },
+  {
+    name: "looksLikeSecretValue catches token-shaped strings",
+    run: () => {
+      assert(
+        looksLikeSecretValue(
+          "eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJzdWIiOiIxMjMifQ.abc123def456",
+        ),
+        "JWT looks secret",
+      );
+      assert(
+        looksLikeSecretValue("a".repeat(40) + "b".repeat(40)),
+        "long base64-ish string looks secret",
+      );
+      assert(
+        !looksLikeSecretValue("Hello world"),
+        "short readable text is safe",
+      );
+      assert(
+        !looksLikeSecretValue("https://example.com/foo/bar"),
+        "URLs are not secrets",
+      );
+      assert(
+        !looksLikeSecretValue("3162189c-2052-4f83-ad83-a66c57460ea2"),
+        "UUIDs are useful and not secret",
+      );
+    },
+  },
+  {
+    name: "redactSensitive scrubs settings payloads",
+    run: () => {
+      const cleaned = redactSensitive(settingsPayload);
+      const json = JSON.stringify(cleaned);
+      assertNotContains(json, "global.dropbox.api.key", "dropbox key dropped");
+      assertNotContains(json, "xxx-do-not-index", "secret value dropped");
+    },
+  },
+  {
+    name: "extractTextFromValue distills HTML and skips secrets",
+    run: () => {
+      const text = extractTextFromValue({
+        title: "Hello",
+        body: "<p>Some <strong>HTML</strong> body.</p>",
+        password: "should-not-appear",
+        nested: { token: "leak-me-please" },
+      });
+      assertContains(text, "Hello", "title preserved");
+      assertContains(text, "HTML body", "html flattened");
+      assertNotContains(text, "should-not-appear", "password redacted");
+      assertNotContains(text, "leak-me-please", "nested token redacted");
+    },
+  },
+  {
+    name: "pickTitle / pickId prefer common SEQTA fields",
+    run: () => {
+      assertEqual(
+        pickTitle({ title: "Hello", name: "Other" }),
+        "Hello",
+        "title wins over name",
+      );
+      assertEqual(
+        pickTitle({ filename: "doc.pdf" }),
+        "doc.pdf",
+        "filename fallback",
+      );
+      assertEqual(pickId({ id: 42 }), "42", "numeric id stringified");
+      assertEqual(pickId({ uuid: "abc" }), "abc", "uuid id");
+    },
+  },
+  {
+    name: "buildIndexItem produces redacted, well-formed records",
+    run: () => {
+      const item = buildIndexItem({
+        id: "x-1",
+        text: "Test",
+        category: "passive",
+        rawForContent: {
+          title: "Test",
+          body: "<p>Hello</p>",
+          token: "should-be-stripped",
+        },
+        metadata: { route: "/seqta/student/load/whatever", apiKey: "drop" },
+        actionId: "passive",
+        renderComponentId: "passive",
+      });
+      assertEqual(item.id, "x-1", "id propagated");
+      assertContains(item.content, "Hello", "html distilled");
+      assertNotContains(item.content, "should-be-stripped", "token stripped");
+      assert(
+        !("apiKey" in (item.metadata as Record<string, unknown>)),
+        "apiKey metadata stripped",
+      );
+      assertEqual(item.category, "passive", "category passes through");
+    },
+  },
+  {
+    name: "courses fixture flattens lesson HTML",
+    run: () => {
+      // Verify that the structural shape we depend on still matches.
+      assert(Array.isArray(coursesPayload.w), "lesson grid present");
+      const lessonHtml = (coursesPayload.w[0]?.[1] as { l?: string })?.l ?? "";
+      assertContains(lessonHtml, "ed.ted.com", "lesson html link present");
+    },
+  },
+  {
+    name: "subjects fixture exposes programme/metaclass",
+    run: () => {
+      const subject = subjectsListPayload[0]?.subjects[0];
+      assert(subject, "fixture has at least one subject");
+      assert(
+        Number.isFinite(subject!.programme) &&
+          Number.isFinite(subject!.metaclass),
+        "programme & metaclass numeric",
+      );
+    },
+  },
+  {
+    name: "documents fixture exposes uuid + filename",
+    run: () => {
+      const doc = documentsPayload[0]?.docs[0];
+      assert(doc?.uuid && doc?.filename, "uuid + filename present");
+    },
+  },
+  {
+    name: "notices fixture is HTML-bearing",
+    run: () => {
+      assertContains(
+        noticesPayload[0]?.contents ?? "",
+        "<p>",
+        "notice html present",
+      );
+    },
+  },
+  {
+    name: "portals fixture has external url",
+    run: () => {
+      assert(portalsPayload[0]?.url?.includes("mathletics"), "portal url");
+    },
+  },
+  {
+    name: "folio entry contents passes html-flattening",
+    run: () => {
+      const distilled = extractTextFromValue(folioEntryPayload, {
+        maxChars: 4000,
+      });
+      assertContains(distilled, "reflection", "folio body extracted");
+    },
+  },
+];
+
+export interface SelfTestReport {
+  passed: number;
+  failed: number;
+  failures: Array<{ name: string; error: string }>;
+}
+
+/**
+ * Runs every assertion case and resolves with a summary. Never throws.
+ *
+ * Designed to be invoked from `window.globalSearchDebug.runSelfTests()`
+ * by maintainers who want to validate the indexing pipeline against a
+ * real SEQTA tab.
+ */
+export async function runGlobalSearchSelfTests(): Promise<SelfTestReport> {
+  const report: SelfTestReport = { passed: 0, failed: 0, failures: [] };
+  for (const test of cases) {
+    try {
+      await test.run();
+      report.passed++;
+    } catch (e) {
+      report.failed++;
+      const error =
+        e instanceof Error ? `${e.name}: ${e.message}` : String(e);
+      report.failures.push({ name: test.name, error });
+    }
+  }
+  if (report.failed > 0) {
+    console.warn(
+      `[Global Search Self-Tests] ${report.failed} failed / ${report.passed} passed`,
+      report.failures,
+    );
+  } else {
+    console.info(
+      `[Global Search Self-Tests] All ${report.passed} cases passed`,
+    );
+  }
+  return report;
+}

src/plugins/built-in/globalSearch/src/indexing/worker/vectorWorkerManager.ts

@@ -19,6 +19,8 @@ export class VectorWorkerManager {
   private initializationMutex = false;
   private idleTimer: NodeJS.Timeout | null = null;
   private unloadTimer: NodeJS.Timeout | null = null;
+  /** Non-streaming `process` jobs must not hit the idle shutdown mid-flight. */
+  private vectorizationLockCount = 0;
 
   private streamingSession: {
     isActive: boolean;
@@ -92,6 +94,12 @@ export class VectorWorkerManager {
             break;
 
           case "progress":
+            if (
+              data.status === "processing" ||
+              data.status === "started"
+            ) {
+              this.bumpActivityDuringVectorization();
+            }
             if (this.progressCallback) {
               this.progressCallback(data);
 
@@ -120,6 +128,7 @@ export class VectorWorkerManager {
             break;
 
           case "streamingProgress":
+            this.bumpActivityDuringVectorization();
             if (this.progressCallback && this.streamingSession?.isActive) {
               const { processed } = data;
               this.progressCallback({
@@ -150,6 +159,7 @@ export class VectorWorkerManager {
     this.readyPromise = null;
     this.progressCallback = null;
     this.initializationMutex = false;
+    this.vectorizationLockCount = 0;
     this.clearIdleTimer();
     this.clearUnloadTimer();
     if (this.streamingSession?.isActive) {
@@ -158,15 +168,27 @@ export class VectorWorkerManager {
   }
 
   private startIdleTimer() {
+    if (this.vectorizationLockCount > 0 || this.streamingSession?.isActive) {
+      return;
+    }
     this.clearIdleTimer();
     this.idleTimer = setTimeout(() => {
-      if (!this.streamingSession?.isActive && this.isInitialized) {
+      if (this.vectorizationLockCount > 0) return;
+      if (this.streamingSession?.isActive) return;
+      if (!this.isInitialized) return;
       console.debug("[VectorWorker] Auto-shutting down due to 2 minutes of inactivity");
       this.resetWorkerState();
-      }
     }, 120000); // 2 minutes
   }
 
+  /** Extends idle deadline while embeddings run; cheap if no idle timer is scheduled. */
+  private bumpActivityDuringVectorization() {
+    if (this.vectorizationLockCount > 0 || this.streamingSession?.isActive) {
+      this.clearIdleTimer();
+    }
+    this.updateActivity();
+  }
+
   private clearIdleTimer() {
     if (this.idleTimer) {
       clearTimeout(this.idleTimer);
@@ -184,6 +206,7 @@ export class VectorWorkerManager {
   private scheduleUnload(delay: number = 10000) {
     this.clearUnloadTimer();
     this.unloadTimer = setTimeout(() => {
+      if (this.vectorizationLockCount > 0) return;
       if (!this.streamingSession?.isActive && this.isInitialized) {
         console.debug("[VectorWorker] Auto-unloading after processing complete");
         this.resetWorkerState();
@@ -193,6 +216,9 @@ export class VectorWorkerManager {
 
   private updateActivity() {
     this.clearUnloadTimer();
+    if (this.vectorizationLockCount > 0 || this.streamingSession?.isActive) {
+      return;
+    }
     this.startIdleTimer();
   }
 
@@ -298,8 +324,31 @@ export class VectorWorkerManager {
       return;
     }
 
-    this.progressCallback = onProgress || null;
+    // Wait until the worker reports a terminal status. Previously this method
-    this.updateActivity();
+    // returned as soon as the job was queued, so indexers.ts continued into
+    // stopHeartbeat/loadAll/loadDynamicItems on the main thread while
+    // vectorization was still running — blocking indexing-progress handlers
+    // and freezing the chip on “Vectorization in progress”.
+    this.vectorizationLockCount++;
+    this.clearIdleTimer();
+    this.clearUnloadTimer();
+
+    try {
+      await new Promise<void>((resolve) => {
+        let settled = false;
+        const wrap: ProgressCallback = (data) => {
+          onProgress?.(data);
+          if (
+            !settled &&
+            (data.status === "complete" ||
+              data.status === "error" ||
+              data.status === "cancelled")
+          ) {
+            settled = true;
+            resolve();
+          }
+        };
+        this.progressCallback = wrap;
 
         console.debug(
           `Sending ${uniqueItems.length} unique items to worker for processing.`,
@@ -309,6 +358,16 @@ export class VectorWorkerManager {
           type: "process",
           data: { items: uniqueItems },
         });
+      });
+    } finally {
+      this.vectorizationLockCount = Math.max(0, this.vectorizationLockCount - 1);
+      if (
+        this.vectorizationLockCount === 0 &&
+        !this.streamingSession?.isActive
+      ) {
+        this.startIdleTimer();
+      }
+    }
   }
 
   async startStreamingSession(

src/plugins/built-in/globalSearch/src/search/dedupeIndexItems.ts

@@ -0,0 +1,151 @@
+import type { CombinedResult } from "../core/types";
+import type { IndexItem } from "../indexing/types";
+
+function toFiniteNumber(value: unknown): number | undefined {
+  if (typeof value === "number" && Number.isFinite(value)) return value;
+  if (typeof value === "string") {
+    const t = value.trim();
+    if (!t) return undefined;
+    const n = Number(t);
+    return Number.isFinite(n) ? n : undefined;
+  }
+  return undefined;
+}
+
+/** Same SPA destination as handlers for `course` / `subjectcourse` / passive `courses`. */
+function shouldDedupeAsSameCourseSPA(item: IndexItem): boolean {
+  if (item.actionId === "subjectassessment") return false;
+  if (item.metadata?.type === "assessments") return false;
+
+  if (item.renderComponentId === "course") return true;
+  if (item.actionId === "course") return true;
+  if (item.actionId === "subjectcourse") return true;
+  if (
+    item.actionId === "passive" &&
+    item.metadata?.sourcePage === "/courses"
+  ) {
+    return true;
+  }
+  return false;
+}
+
+export function courseDestinationKey(item: IndexItem): string | undefined {
+  if (!shouldDedupeAsSameCourseSPA(item)) return undefined;
+  const md = item.metadata ?? {};
+  const programme = toFiniteNumber(
+    md.programme ?? md.programmeId ?? md.programmeID,
+  );
+  const metaclass = toFiniteNumber(
+    md.metaclass ?? md.metaclassId ?? md.metaclassID ?? md.subjectId,
+  );
+  if (programme === undefined || metaclass === undefined) return undefined;
+  return `course:${programme}:${metaclass}`;
+}
+
+function isPassiveLike(item: IndexItem): boolean {
+  return (
+    item.actionId === "passive" || item.metadata?.source === "passive"
+  );
+}
+
+function pickBetterCourseNavDuplicate(a: IndexItem, b: IndexItem): IndexItem {
+  const aP = isPassiveLike(a);
+  const bP = isPassiveLike(b);
+  if (aP && !bP) return b;
+  if (!aP && bP) return a;
+  // Prefer curated job row (courses store) vs other categories
+  if (a.category === "courses" && b.category !== "courses") return a;
+  if (b.category === "courses" && a.category !== "courses") return b;
+  if (a.renderComponentId === "course" && b.renderComponentId !== "course")
+    return a;
+  if (b.renderComponentId === "course" && a.renderComponentId !== "course")
+    return b;
+  const ad = typeof a.dateAdded === "number" ? a.dateAdded : 0;
+  const bd = typeof b.dateAdded === "number" ? b.dateAdded : 0;
+  return ad >= bd ? a : b;
+}
+
+/**
+ * Collapses multiple index rows that open the same course hash route
+ * (e.g. `course` job + passive `/load/courses` capture) so search shows one hit.
+ */
+export function dedupeIndexItemsForSearch(items: IndexItem[]): IndexItem[] {
+  const winners = new Map<string, IndexItem>();
+
+  for (const item of items) {
+    const key = courseDestinationKey(item);
+    if (!key) continue;
+    const prev = winners.get(key);
+    winners.set(
+      key,
+      prev ? pickBetterCourseNavDuplicate(prev, item) : item,
+    );
+  }
+
+  const seenCanon = new Set<string>();
+  const out: IndexItem[] = [];
+
+  for (const item of items) {
+    const key = courseDestinationKey(item);
+    if (!key) {
+      out.push(item);
+      continue;
+    }
+    if (seenCanon.has(key)) continue;
+    seenCanon.add(key);
+    out.push(winners.get(key)!);
+  }
+
+  return out;
+}
+
+function dynamicCourseKey(row: CombinedResult): string | undefined {
+  if (row.type !== "dynamic") return undefined;
+  return courseDestinationKey(row.item as IndexItem);
+}
+
+/**
+ * Final pass after hybrid expansion: vector-only recall can still surface a
+ * second row for the same `/courses/P:M` SPA route using a stale passive id.
+ */
+export function dedupeCombinedResultsByCourseNav(
+  results: CombinedResult[],
+): CombinedResult[] {
+  const best = new Map<string, CombinedResult>();
+
+  for (const r of results) {
+    const key = dynamicCourseKey(r);
+    if (!key) continue;
+    const prev = best.get(key);
+    if (!prev) {
+      best.set(key, r);
+      continue;
+    }
+    const aItem = prev.item as IndexItem;
+    const bItem = r.item as IndexItem;
+    const winnerItem = pickBetterCourseNavDuplicate(aItem, bItem);
+    const envelope = winnerItem.id === aItem.id ? prev : r;
+    best.set(key, {
+      ...envelope,
+      score: Math.max(prev.score, r.score),
+      id: winnerItem.id,
+      item: winnerItem,
+    });
+  }
+
+  const seenCanon = new Set<string>();
+  const out: CombinedResult[] = [];
+
+  for (const r of results) {
+    const key = dynamicCourseKey(r);
+    if (!key) {
+      out.push(r);
+      continue;
+    }
+    if (seenCanon.has(key)) continue;
+    seenCanon.add(key);
+    out.push(best.get(key)!);
+  }
+
+  return out;
+}

src/plugins/built-in/globalSearch/src/search/hybridSearch.ts

@@ -2,6 +2,32 @@ import type { IndexItem } from "../indexing/types";
 import type { CombinedResult } from "../core/types";
 import { searchVectors, type VectorSearchResult } from "./vector/vectorSearch";
 import { jobs } from "../indexing/jobs";
+import {
+  getLexicalMatchQuality,
+  isStrongLexicalMatch,
+  STRONG_LEXICAL_THRESHOLD,
+} from "./lexicalMatch";
+
+function isIndexItem(item: CombinedResult["item"]): item is IndexItem {
+  return (item as IndexItem).dateAdded !== undefined;
+}
+
+/**
+ * Heuristic for "this query is still too short / too sparse for vector
+ * recall to be reliable". When true we should not promote vector-only
+ * results above lexical ones.
+ *
+ * Note: this is intentionally distinct from the absolute >2 character cut-off
+ * used for `hybridSearch`. Vector recall on 3-7 character single-token
+ * queries is noisy enough that we should keep lexical results dominant.
+ */
+function isWeakSemanticQuery(trimmedQuery: string): boolean {
+  if (trimmedQuery.length < 8) return true;
+  const meaningfulTokens = trimmedQuery
+    .split(/\s+/)
+    .filter((t) => t.length >= 3);
+  return meaningfulTokens.length < 2;
+}
 
 /**
  * Hybrid Search Implementation
@@ -36,14 +62,6 @@ const DEFAULT_OPTIONS: Required<HybridSearchOptions> = {
   recencyWeight: 0.1,
 };
 
-/**
- * Normalizes a score to 0-1 range
- */
-function normalizeScore(score: number, min: number, max: number): number {
-  if (max === min) return 0.5;
-  return Math.max(0, Math.min(1, (score - min) / (max - min)));
-}
-
 /**
  * Calculates recency boost based on item age
  */
@@ -55,28 +73,56 @@ function calculateRecencyBoost(item: IndexItem, now: number): number {
 }
 
 /**
- * Calculates popularity boost (can be extended with click tracking, etc.)
+ * Category-aware popularity / structure boost.
+ *
+ * High-confidence curated content (assignments, courses, subjects, forums)
+ * sits above noisier sources (notices, documents) which sit above the
+ * passive store. This keeps the most actionable hits at the top while
+ * still surfacing wide-recall semantic matches when relevant.
  */
 function calculatePopularityBoost(item: IndexItem): number {
-  // For now, boost based on category and metadata
   let boost = 0;
 
-  // Boost assignments/assessments
+  switch (item.category) {
-  if (item.category === "assignments") {
+    case "assignments":
-    boost += 0.1;
+      boost += 0.12;
+      break;
+    case "subjects":
+    case "courses":
+      boost += 0.08;
+      break;
+    case "forums":
+    case "messages":
+      boost += 0.06;
+      break;
+    case "notices":
+    case "folio":
+    case "reports":
+    case "goals":
+      boost += 0.04;
+      break;
+    case "documents":
+      boost += 0.03;
+      break;
+    case "portals":
+      boost += 0.02;
+      break;
+    case "passive":
+      boost -= 0.1;
+      break;
+    case "messages-support":
+      boost -= 0.18;
+      break;
   }
 
-  // Boost upcoming items
+  if (item.metadata?.isUpcoming) boost += 0.12;
-  if (item.metadata?.isUpcoming) {
+  if (item.metadata?.subjectCode) boost += 0.04;
-    boost += 0.15;
+  if (item.metadata?.entityType === "course") boost += 0.02;
-  }
+  if (item.metadata?.source === "passive") boost -= 0.08;
+  if (item.metadata?.supportRecord) boost -= 0.12;
+  if (item.metadata?.priority === "low") boost -= 0.05;
 
-  // Boost items with subject codes (more structured)
+  return Math.max(-0.2, Math.min(boost, 0.3));
-  if (item.metadata?.subjectCode) {
-    boost += 0.05;
-  }
-  
-  return Math.min(boost, 0.3); // Cap at 0.3
 }
 
 /**
@@ -98,10 +144,6 @@ export async function hybridSearch(
   // Limit BM25 results to top K
   const topBm25Results = bm25Results.slice(0, opts.bm25TopK);
 
-  // Get vector search results for reranking
-  // We'll search the full index and then filter to our BM25 results
-  let vectorResults: VectorSearchResult[] = [];
-  
   if (trimmedQuery.length > 2) {
     try {
       // Get more vector results than BM25 results to ensure coverage
@@ -121,59 +163,59 @@ export async function hybridSearch(
       // Now rerank BM25 results with vector scores
       const now = Date.now();
       
-      const rerankedResults = topBm25Results.map(result => {
+      const rerankedResults: CombinedResult[] = topBm25Results.map(result => {
         const item = result.item;
 
-        // Normalize BM25 score to 0-1
+        // Static command items don't have dateAdded/metadata/category to score
-        // Fuse.js scores: lower is better (0 = perfect match)
+        // against — pass them through untouched so palette commands still
-        // We need to invert: higher score = better match
+        // surface correctly.
-        // Result.score is typically 0-100, where higher = better
+        if (!isIndexItem(item)) {
-        // So we normalize it to 0-1
+          return result;
+        }
+
+        // Normalize BM25 score to 0-1.
+        // Result.score is typically 0-100, where higher = better, so we
+        // clamp into the 0..1 range.
         const normalizedBm25Score = Math.max(0, Math.min(1, result.score / 100));
 
-        // Get vector similarity (0-1, already normalized)
+        // Get vector similarity (0-1, already normalized). If item wasn't in
-        // If item wasn't in vector results, use a default low score
+        // vector results, use a default mid-low score.
-        const vectorSimilarity = vectorMap.get(item.id) || 0.3; // Default to 0.3 if not found
+        const vectorSimilarity = vectorMap.get(item.id) || 0.3;
 
-        // Calculate recency boost (0-1 range)
         const recencyBoost = opts.recencyBoost
           ? calculateRecencyBoost(item, now) * opts.recencyWeight
           : 0;
 
-        // Calculate popularity boost (0-1 range)
         const popularityBoost = calculatePopularityBoost(item);
 
-        // Apply job-specific boost if available
         const job = jobs[item.category];
         let jobBoost = 0;
         if (job && typeof job.boostCriteria === 'function') {
           const boost = job.boostCriteria(item, trimmedQuery);
           if (boost) {
-            jobBoost = boost / 100; // Normalize boost to 0-1
+            jobBoost = boost / 100;
           }
         }
 
-        // Combine scores using weighted average
+        // Lexical guardrail: title matches must outweigh fuzzy vector/content
-        // BM25 and vector are weighted, boosts are additive
+        // overlap so exact titles lead the list.
+        const lexicalQuality = getLexicalMatchQuality(item, trimmedQuery);
+        let lexicalBonus = lexicalQuality > 0 ? lexicalQuality / 80 : 0;
+        if (lexicalQuality >= 12) lexicalBonus += 0.42;
+        else if (lexicalQuality >= 10) lexicalBonus += 0.24;
+        else if (lexicalQuality >= 8) lexicalBonus += 0.14;
+
         const hybridScore =
           (normalizedBm25Score * opts.bm25Weight) +
           (vectorSimilarity * opts.vectorWeight) +
           recencyBoost +
           popularityBoost +
-          jobBoost;
+          jobBoost +
+          lexicalBonus;
 
         return {
           ...result,
-          score: hybridScore * 100, // Scale back to 0-100 for consistency
+          score: hybridScore * 100,
-          // Store component scores for debugging (optional, can be removed in production)
-          _hybridScores: {
-            bm25: normalizedBm25Score,
-            vector: vectorSimilarity,
-            recency: recencyBoost,
-            popularity: popularityBoost,
-            jobBoost: jobBoost,
-            final: hybridScore,
-          },
         };
       });
       
@@ -200,7 +242,7 @@ export async function hybridSearch(
 export async function hybridSearchWithExpansion(
   bm25Results: CombinedResult[],
   query: string,
-  allItems: IndexItem[],
+  _allItems: IndexItem[],
   options: HybridSearchOptions = {},
 ): Promise<CombinedResult[]> {
   const opts = { ...DEFAULT_OPTIONS, ...options };
@@ -214,6 +256,13 @@ export async function hybridSearchWithExpansion(
     return rerankedBm25;
   }
 
+  // For short / single-token queries vector expansion brings in too much
+  // noise (and is the main reason results "flicker" between adjacent
+  // keystrokes). Keep semantic recall for longer queries.
+  if (isWeakSemanticQuery(trimmedQuery)) {
+    return rerankedBm25.slice(0, opts.finalLimit);
+  }
+
   // Get vector search results
   let vectorResults: VectorSearchResult[] = [];
   try {
@@ -229,8 +278,26 @@ export async function hybridSearchWithExpansion(
 
   const now = Date.now();
 
+  // Compute the floor at which a vector-only result is allowed to enter the
+  // ranking. Strong lexical matches in the BM25 list set this floor — a
+  // vector-only result must beat the lowest strong lexical match's score by
+  // a margin to displace it.
+  let strongLexicalFloor = -Infinity;
+  for (const r of rerankedBm25) {
+    if (isIndexItem(r.item) && isStrongLexicalMatch(r.item, trimmedQuery)) {
+      if (r.score > strongLexicalFloor) {
+        strongLexicalFloor = r.score;
+      }
+    }
+  }
+  // Vector-only results may sit at most at this score:
+  const vectorOnlyCeiling = strongLexicalFloor === -Infinity
+    ? Infinity
+    : strongLexicalFloor - 1;
+
   vectorResults.forEach(v => {
-    if (!bm25Ids.has(v.object.id)) {
+    if (bm25Ids.has(v.object.id)) return;
+
     // This is a semantic match that BM25 missed
     const item = v.object;
 
@@ -240,8 +307,24 @@ export async function hybridSearchWithExpansion(
       : 0;
     const popularityBoost = calculatePopularityBoost(item);
 
+    // Penalize vector-only matches that have no lexical content overlap.
+    // Vector recall on its own is fuzzy — without lexical confirmation we
+    // should rank these below curated keyword hits.
+    const lexicalQuality = getLexicalMatchQuality(item, trimmedQuery);
+    let vectorOnlyPenalty = 0;
+    if (lexicalQuality === 0) {
+      vectorOnlyPenalty -= 0.18;
+    }
+
+    // Passive captures without lexical confirmation are demoted further —
+    // they're often raw API records that should never lead the result list.
+    if (item.category === "passive" && lexicalQuality < STRONG_LEXICAL_THRESHOLD) {
+      vectorOnlyPenalty -= 0.12;
+    }
+
     // Vector-only results get lower base score but high vector similarity
-      const vectorScore = v.similarity * opts.vectorWeight + recencyBoost + popularityBoost;
+    const vectorScore =
+      v.similarity * opts.vectorWeight + recencyBoost + popularityBoost + vectorOnlyPenalty;
 
     // Apply job-specific boost if available
     const job = jobs[item.category];
@@ -253,20 +336,15 @@ export async function hybridSearchWithExpansion(
       }
     }
 
+    let finalScore = (vectorScore + jobBoost) * 100;
+    if (finalScore > vectorOnlyCeiling) finalScore = vectorOnlyCeiling;
+
     vectorOnlyResults.push({
       id: item.id,
       type: "dynamic" as const,
-        score: (vectorScore + jobBoost) * 100,
+      score: finalScore,
       item,
-        _hybridScores: {
-          bm25: 0,
-          vector: v.similarity,
-          recency: recencyBoost,
-          popularity: popularityBoost,
-          final: vectorScore + jobBoost,
-        },
     });
-    }
   });
 
   // Combine reranked BM25 results with vector-only results

src/plugins/built-in/globalSearch/src/search/lexicalMatch.ts

@@ -0,0 +1,118 @@
+import type { IndexItem } from "../indexing/types";
+
+/**
+ * Maximum bonus a strong lexical title match can contribute on top of the
+ * underlying Fuse / hybrid score. Tuned to outweigh small vector reranking
+ * deltas so a true assessment-title match cannot be displaced by a vector
+ * neighbour as the user types one more character.
+ */
+export const LEXICAL_TITLE_BONUS = 12;
+
+/**
+ * Threshold at or above which a result counts as a "strong lexical match".
+ * Strong matches must always be surfaced and protected from vector reranking
+ * displacing them.
+ */
+export const STRONG_LEXICAL_THRESHOLD = 6;
+
+const WORD_SPLIT_RE = /\s+/;
+const NON_WORD_RE = /[^a-z0-9]+/gi;
+
+function normalize(value: string | undefined | null): string {
+  if (!value) return "";
+  return String(value).toLowerCase().trim();
+}
+
+function tokens(value: string): string[] {
+  return normalize(value)
+    .split(WORD_SPLIT_RE)
+    .map((t) => t.replace(NON_WORD_RE, ""))
+    .filter(Boolean);
+}
+
+/**
+ * Score how strongly the query lexically matches the title-like fields of an
+ * IndexItem. Return value is a non-negative number — 0 means no useful match.
+ *
+ * Tiers (roughly):
+ *   ~12  exact title equality
+ *   ~10  title starts with full query string
+ *   ~8   title contains full query string, on a word boundary
+ *   ~7   ordered token-prefix match (e.g. `world w` vs `World War 2 Essay`)
+ *   ~5   subject / metadata title contains query
+ *   ~3   any token in title starts with query
+ *   ~2   substring anywhere in title
+ *   0    no lexical signal
+ *
+ * The function is intentionally cheap (string ops only, no regex compilation
+ * per call beyond the constants above) because it is called for every item in
+ * the candidate pool.
+ */
+export function getLexicalMatchQuality(item: IndexItem, query: string): number {
+  const q = normalize(query);
+  if (!q) return 0;
+
+  const title = normalize(item.text);
+  if (!title) return 0;
+
+  if (title === q) return 12;
+  if (title.startsWith(q + " ") || title.startsWith(q)) return 10;
+
+  const queryTokens = tokens(q);
+  const titleTokens = tokens(title);
+
+  if (queryTokens.length > 0 && titleTokens.length >= queryTokens.length) {
+    let bestStreakStart = -1;
+    for (let i = 0; i <= titleTokens.length - queryTokens.length; i++) {
+      let ok = true;
+      for (let j = 0; j < queryTokens.length; j++) {
+        const tt = titleTokens[i + j];
+        const qt = queryTokens[j];
+        const isLast = j === queryTokens.length - 1;
+        if (isLast) {
+          if (!tt.startsWith(qt)) {
+            ok = false;
+            break;
+          }
+        } else {
+          if (tt !== qt) {
+            ok = false;
+            break;
+          }
+        }
+      }
+      if (ok) {
+        bestStreakStart = i;
+        break;
+      }
+    }
+    if (bestStreakStart === 0) return 9;
+    if (bestStreakStart > 0) return 7;
+  }
+
+  if (title.includes(" " + q) || title.includes(q + " ")) return 8;
+
+  // Token starts-with anywhere
+  for (const t of titleTokens) {
+    if (t.startsWith(q)) return 3;
+  }
+
+  // Subject / curated metadata title
+  const md = (item.metadata ?? {}) as Record<string, unknown>;
+  const subjectName = normalize(
+    typeof md.subjectName === "string" ? md.subjectName : "",
+  );
+  const subjectCode = normalize(
+    typeof md.subjectCode === "string" ? md.subjectCode : "",
+  );
+  if (subjectName && (subjectName === q || subjectName.startsWith(q))) return 5;
+  if (subjectCode && (subjectCode === q || subjectCode.startsWith(q))) return 5;
+
+  if (title.includes(q)) return 2;
+
+  return 0;
+}
+
+export function isStrongLexicalMatch(item: IndexItem, query: string): boolean {
+  return getLexicalMatchQuality(item, query) >= STRONG_LEXICAL_THRESHOLD;
+}

src/plugins/built-in/globalSearch/src/search/searchUtils.ts

@@ -3,10 +3,64 @@ import { getStaticCommands, type StaticCommandItem } from "../core/commands";
 import { getDynamicItems } from "../utils/dynamicItems";
 import type { CombinedResult } from "../core/types";
 import type { IndexItem } from "../indexing/types";
-import { searchVectors } from "./vector/vectorSearch";
+import { dedupeCombinedResultsByCourseNav, dedupeIndexItemsForSearch } from "./dedupeIndexItems";
-import type { VectorSearchResult } from "./vector/vectorTypes";
-import { jobs } from "../indexing/jobs";
 import { hybridSearchWithExpansion } from "./hybridSearch";
+import {
+  getLexicalMatchQuality,
+  isStrongLexicalMatch,
+  STRONG_LEXICAL_THRESHOLD,
+} from "./lexicalMatch";
+
+/** Same normalization as lexical matching (trim + lowercase). */
+function normSearchKey(s: string): string {
+  return s.trim().toLowerCase();
+}
+
+/**
+ * Exact title tiers so palette navigation (e.g. "Home", "Assessments") always
+ * wins over hybrid-scored body matches. Higher = sort earlier.
+ */
+function exactTitleSortTier(r: CombinedResult, queryNorm: string): number {
+  if (!queryNorm) return 0;
+  if (r.type === "command") {
+    const cmd = r.item as StaticCommandItem;
+    if (normSearchKey(cmd.text) !== queryNorm) return 0;
+    return cmd.category === "navigation" ? 3 : 2;
+  }
+  const ix = r.item as IndexItem;
+  if (normSearchKey(ix.text) === queryNorm) return 1;
+  return 0;
+}
+
+function compareCombinedSearchResults(
+  a: CombinedResult,
+  b: CombinedResult,
+  queryNorm: string,
+): number {
+  const tierDiff = exactTitleSortTier(b, queryNorm) - exactTitleSortTier(a, queryNorm);
+  if (tierDiff !== 0) return tierDiff;
+
+  if (a.type === "command" && b.type === "dynamic") {
+    return b.score - a.score - 10;
+  }
+  if (a.type === "dynamic" && b.type === "command") {
+    return b.score - a.score + 10;
+  }
+  return b.score - a.score;
+}
+
+function syntheticIndexFromCommand(cmd: StaticCommandItem): IndexItem {
+  return {
+    id: cmd.id,
+    text: cmd.text,
+    category: cmd.category,
+    content: "",
+    dateAdded: 0,
+    metadata: {},
+    actionId: "",
+    renderComponentId: "",
+  };
+}
 
 // Search result cache for better performance
 const searchCache = new Map<string, { results: CombinedResult[]; timestamp: number }>();
@@ -25,8 +79,10 @@ function setCachedResults(query: string, results: CombinedResult[]) {
   // Limit cache size
   if (searchCache.size >= MAX_CACHE_SIZE) {
     const firstKey = searchCache.keys().next().value;
+    if (firstKey !== undefined) {
       searchCache.delete(firstKey);
     }
+  }
   searchCache.set(query, { results, timestamp: Date.now() });
 }
 
@@ -46,8 +102,9 @@ if (typeof window !== 'undefined') {
 }
 
 export function createSearchIndexes() {
+  clearSearchCache();
   const commands = getStaticCommands();
-  const dynamicItems = getDynamicItems();
+  const dynamicItems = dedupeIndexItemsForSearch(getDynamicItems());
 
   // Optimized command search options
   const commandOptions = {
@@ -61,23 +118,40 @@ export function createSearchIndexes() {
     findAllMatches: false, // Performance optimization
   };
 
-  // Optimized dynamic content search options
+  // Optimized dynamic content search options.
+  // The expanded corpus mixes structured entities (assessments, subjects)
+  // with free-form text (course content, notices, folio bodies, passive
+  // captures) so we list a broad set of metadata keys while keeping titles
+  // dominant in the ranking.
+  // NOTE: metadata.route is intentionally excluded. Raw API paths like
+  // `/seqta/student/load/message/people` should never influence ranking — they
+  // historically caused passive-capture support records to bubble up above
+  // real assessments when the user typed substrings that happened to appear in
+  // the path.
   const dynamicOptions = {
     keys: [
-      { name: "text", weight: 3 }, // Increased weight for title matches
+      { name: "text", weight: 3 }, // Title is king
       { name: "content", weight: 1 },
-      { name: "category", weight: 0.5 }, // Lower weight for category
+      { name: "category", weight: 0.4 },
-      { name: "metadata.subjectName", weight: 1.5 }, // Boost subject name matches
+      { name: "metadata.subjectName", weight: 1.6 },
-      { name: "metadata.subjectCode", weight: 1.5 }, // Boost subject code matches
+      { name: "metadata.subjectCode", weight: 1.6 },
+      { name: "metadata.subject", weight: 1.4 },
+      { name: "metadata.courseCode", weight: 1.2 },
+      { name: "metadata.filename", weight: 1.2 },
+      { name: "metadata.author", weight: 0.8 },
+      { name: "metadata.authorName", weight: 0.8 },
+      { name: "metadata.label", weight: 0.6 },
+      { name: "metadata.categoryName", weight: 0.6 },
+      { name: "metadata.entityType", weight: 0.4 },
     ],
     includeScore: true,
     includeMatches: true,
-    threshold: 0.5, // More permissive for better partial word matching (increased from 0.4)
+    threshold: 0.5,
-    minMatchCharLength: 2, // Minimum 2 characters for Fuse.js matches (substring fallback handles shorter queries)
+    minMatchCharLength: 2,
-    distance: 100, // Increased to allow matches across longer strings
+    distance: 100,
     useExtendedSearch: true,
-    ignoreLocation: true, // Allow matches anywhere in the string for better partial word matching
+    ignoreLocation: true,
-    findAllMatches: true, // Enable to find all matches for better partial word support
+    findAllMatches: true,
     shouldSort: true,
   };
 
@@ -117,7 +191,19 @@ export function searchCommands(
   return searchResults.map((result: FuseResult<StaticCommandItem>) => {
     const item = result.item;
     const fuseScore = 15 * (1 - (result.score || 0.5));
-    const score = fuseScore + (item.priority ?? 0);
+    let score = fuseScore + (item.priority ?? 0);
+
+    // Static palette titles share the same lexical tiers as index titles, but
+    // Fuse scores are tiny versus hybrid dynamic scores — scale title matches
+    // up so "Assessments" / prefix matches stay competitive with body hits.
+    const titleLex = getLexicalMatchQuality(syntheticIndexFromCommand(item), query);
+    if (titleLex >= 12) score += 240;
+    else if (titleLex >= 10) score += 195;
+    else if (titleLex >= 9) score += 165;
+    else if (titleLex >= 8) score += 140;
+    else if (titleLex >= 7) score += 120;
+    else if (titleLex >= 6) score += 100;
+    else if (titleLex > 0) score += titleLex * 14;
 
     return {
       id: item.id,
@@ -197,15 +283,24 @@ export function searchDynamicItems(
     const recencyBoost = sortByRecent ? 1 / (ageInDays + 1) : 0;
     score += recencyBoost;
 
-    // Boost for exact text matches (especially at the start)
+    // Lexical title bonus — sticky across adjacent keystrokes so a strong
-    const textLower = item.text.toLowerCase();
+    // title prefix match like `world wa` doesn't disappear from the top once
-    if (textLower.startsWith(queryLower)) {
+    // vector reranking kicks in.
-      score += 5; // Strong boost for prefix matches
+    const lexicalQuality = getLexicalMatchQuality(item, queryLower);
-    } else if (textLower.includes(queryLower)) {
+    if (lexicalQuality > 0) {
-      score += 2; // Boost for substring matches
+      score += lexicalQuality;
+      // Curated-content boost: assessments and assignments with a strong
+      // title match should be elevated further, since they are the items
+      // users are most often hunting for.
+      if (
+        lexicalQuality >= STRONG_LEXICAL_THRESHOLD &&
+        (item.category === "assignments" || item.category === "assessments")
+      ) {
+        score += 4;
+      }
     }
 
-    // Boost for category matches
+    // Category match (small nudge)
     if (item.category.toLowerCase().includes(queryLower)) {
       score += 1;
     }
@@ -221,17 +316,12 @@ export function searchDynamicItems(
 
   // Add additional matches from simple substring search
   additionalMatches.forEach((item) => {
-    // Check if already in results
     if (!results.find(r => r.id === item.id)) {
-      const textLower = item.text.toLowerCase();
       let score = 5; // Base score for substring matches
 
-      // Boost for prefix matches
+      const lexicalQuality = getLexicalMatchQuality(item, queryLower);
-      if (textLower.startsWith(queryLower)) {
+      score += lexicalQuality;
-        score += 5;
-      }
 
-      // Recency boost
       const ageInDays = (now - item.dateAdded) / (1000 * 60 * 60 * 24);
       const recencyBoost = sortByRecent ? 1 / (ageInDays + 1) : 0;
       score += recencyBoost;
@@ -241,6 +331,7 @@ export function searchDynamicItems(
         type: "dynamic" as const,
         score,
         item,
+        matches: undefined,
       });
     }
   });
@@ -249,6 +340,7 @@ export function searchDynamicItems(
   return results.sort((a, b) => b.score - a.score).slice(0, limit);
 }
 
+
 export async function performSearch(
   query: string,
   commandsFuse: Fuse<StaticCommandItem>,
@@ -286,12 +378,37 @@ export async function performSearch(
       sortByRecent,
     );
 
+    // Step 2b: Always include strong lexical title matches, even if Fuse
+    // missed them with the current threshold. This is the safety net that
+    // stops `world wa` from dropping a `World War 2 Essay` assessment that
+    // `world w` happily showed.
+    const allItems = Array.from(dynamicIdToItemMap.values());
+    const seen = new Set(bm25Results.map((r) => r.id));
+    const lexicalAdds: CombinedResult[] = [];
+    for (const item of allItems) {
+      if (seen.has(item.id)) continue;
+      if (!isStrongLexicalMatch(item, trimmedQuery)) continue;
+      const quality = getLexicalMatchQuality(item, trimmedQuery);
+      let score = 6 + quality;
+      if (item.category === "assignments" || item.category === "assessments") {
+        score += 4;
+      }
+      lexicalAdds.push({
+        id: item.id,
+        type: "dynamic" as const,
+        score,
+        item,
+        matches: undefined,
+      });
+    }
+    if (lexicalAdds.length > 0) {
+      bm25Results.push(...lexicalAdds);
+      bm25Results.sort((a, b) => b.score - a.score);
+    }
+
     // Step 3: Apply hybrid search (BM25 + Vector reranking + boosting)
     if (trimmedQuery.length > 2 && bm25Results.length > 0) {
       try {
-        // Get all items for expansion
-        const allItems = Array.from(dynamicIdToItemMap.values());
-        
         // Apply hybrid search with expansion
         dynamicResults = await hybridSearchWithExpansion(
           bm25Results,
@@ -320,22 +437,19 @@ export async function performSearch(
   // Step 4: Combine command and dynamic results
   const allResults = [...commandResults, ...dynamicResults];
 
-  // Sort by score (commands typically have higher priority)
+  allResults.sort((a, b) =>
-  allResults.sort((a, b) => {
+    compareCombinedSearchResults(a, b, trimmedQuery),
-    // Commands always come first if scores are similar
+  );
-    if (a.type === "command" && b.type === "dynamic") {
+
-      return b.score - a.score - 10; // Commands get +10 boost
+  const dedupedResults = dedupeCombinedResultsByCourseNav(allResults);
-    }
+  dedupedResults.sort((a, b) =>
-    if (a.type === "dynamic" && b.type === "command") {
+    compareCombinedSearchResults(a, b, trimmedQuery),
-      return b.score - a.score + 10; // Commands get +10 boost
+  );
-    }
-    return b.score - a.score;
-  });
 
   // Cache results for queries longer than 2 chars
   if (trimmedQuery.length > 2) {
-    setCachedResults(trimmedQuery, allResults);
+    setCachedResults(trimmedQuery, dedupedResults);
   }
 
-  return allResults;
+  return dedupedResults;
 }

src/plugins/built-in/globalSearch/src/search/vector/vectorSearch.ts

@@ -40,7 +40,6 @@ export interface VectorSearchResult extends SearchResult {
 
 // Cache for query embeddings to avoid recomputing
 const embeddingCache = new Map<string, number[]>();
-const EMBEDDING_CACHE_TTL = 1000 * 60 * 30; // 30 minutes
 const MAX_EMBEDDING_CACHE_SIZE = 50;
 
 function getCachedEmbedding(query: string): number[] | null {
@@ -55,8 +54,10 @@ function setCachedEmbedding(query: string, embedding: number[]) {
   // Limit cache size
   if (embeddingCache.size >= MAX_EMBEDDING_CACHE_SIZE) {
     const firstKey = embeddingCache.keys().next().value;
+    if (firstKey !== undefined) {
       embeddingCache.delete(firstKey);
     }
+  }
   embeddingCache.set(query, embedding);
 }
 

src/plugins/built-in/globalSearch/src/utils/versionCheck.ts

@@ -1,4 +1,5 @@
 import browser from "webextension-polyfill";
+import { resetSearchIndexes } from "../indexing/resetIndexes";
 
 const VERSION_STORAGE_KEY = "betterseqta-global-search-version";
 const VERSION_CACHE_KEY = "betterseqta-global-search-cache-version";
@@ -40,32 +41,51 @@ export function storeVersion(version: string): void {
 }
 
 /**
- * Checks if the extension has been updated and clears caches if needed
+ * Checks if the extension has been updated and clears caches + resets the
- * Returns true if an update was detected
+ * search index if needed.
+ *
+ * The reset is intentionally aggressive: every manifest version bump
+ * triggers a full IndexedDB wipe so changes to indexer extraction logic,
+ * job sets, or item shape can never serve stale results from an older
+ * build. The next indexing pass will repopulate from scratch in the
+ * background. Re-population is bounded by the per-job rate limits in
+ * `api.ts` so it can't hammer SEQTA after an update.
+ *
+ * Returns true if an update was detected.
  */
 export async function checkAndHandleUpdate(): Promise<boolean> {
   const currentVersion = getCurrentVersion();
   const storedVersion = getStoredVersion();
 
-  // If no stored version, this is first run - store current version
+  // First run: just remember the version, don't reset (the user likely
+  // just installed the extension; the index is already empty).
   if (!storedVersion) {
-    console.debug(`[Version Check] First run detected, storing version ${currentVersion}`);
+    console.debug(
+      `[Version Check] First run detected, storing version ${currentVersion}`,
+    );
     storeVersion(currentVersion);
     return false;
   }
 
-  // If versions match, no update
   if (storedVersion === currentVersion) {
     return false;
   }
 
-  // Version mismatch detected - extension was updated
+  console.log(
-  console.log(`[Version Check] Extension updated from ${storedVersion} to ${currentVersion}, clearing caches...`);
+    `[Version Check] Extension updated from ${storedVersion} to ${currentVersion}, resetting search index...`,
+  );
 
-  // Clear all caches
   await clearAllCaches();
 
-  // Store new version
+  try {
+    await resetSearchIndexes();
+    console.log(
+      "[Version Check] Search index reset; next indexing pass will repopulate from scratch.",
+    );
+  } catch (e) {
+    console.warn("[Version Check] resetSearchIndexes failed:", e);
+  }
+
   storeVersion(currentVersion);
 
   return true;

src/seqta/utils/Openers/OpenWhatsNewPopup.ts

@@ -33,6 +33,14 @@ export function OpenWhatsNewPopup(onDismissed?: () => void) {
 
   const text = stringToHTML(/* html */ `
     <div class="whatsnewTextContainer" style="height: 50%;overflow-y: auto;">
+
+      <h1>3.6.6 – Global Search improvements!</h1>
+      <li>Tuned hybrid search and indexing reliability.</li>
+      <li>Clearer progress UI and green “Done!” when a pass finishes.</li>
+      <li>Merged duplicate course hits that opened the same page.</li>
+      <li>Reset Index reminds you to reload the tab to rebuild.</li>
+      <li>Index now captures all content accross the site, not just assessments and courses.</li>
+
       <h1>3.6.5 - Theme of the Month, custom message folders & assessment weighting overrides</h1>
       <li>Added Theme of the Month — a monthly featured theme popup with a link to view it in the theme store.</li>
       <li>Added custom message folders for organising direct DM's with drag to reorder.</li>