diff --git a/src/test-hooks/sw-hooks.ts b/src/test-hooks/sw-hooks.ts
index b724a48..41d6883 100644
--- a/src/test-hooks/sw-hooks.ts
+++ b/src/test-hooks/sw-hooks.ts
@@ -227,4 +227,85 @@ globalThis.__mokoshTest = {
   },
 } as MokoshTestSurface;
 
+// ─── Harness message bridge ───────────────────────────────────────────
+// EMPIRICAL ARCHITECTURE NOTE: Puppeteer 25 + Chrome 148 + headless
+// cannot reliably evaluate against the SW directly. The harness
+// queries chrome.* state through the popup page (which has full
+// chrome.* API access) but cannot read the SW's globalThis.__mokoshTest
+// because the popup is a SEPARATE V8 isolate. So we bridge: the popup
+// sends chrome.runtime.sendMessage queries; this handler responds with
+// the queried state.
+//
+// Protocol — popup → SW message: { type: '__mokoshTestQuery', op: <string> }
+// Response shapes:
+//   op='snapshot' → { count, lastOptions, ids }
+//   op='fire-on-startup' → { ok: true } OR { ok: false, error: 'no-handler' }
+//   op='handler-types' → { onClicked, onStartup, notificationOnClicked }
+// Unknown ops respond { ok: false, error: 'unknown-op' }.
+//
+// Returning `true` from the onMessage handler tells Chrome the
+// response is async; we keep sendResponse as a closed-over callback.
+// The bridge handler is registered AFTER the production listeners so
+// the hook never accidentally intercepts a production message —
+// __mokoshTest* messages are unambiguously test-only.
+chrome.runtime.onMessage.addListener((rawMessage, _sender, sendResponse) => {
+  // Narrow the message — we accept ANY shape but only act on our type.
+  if (rawMessage === null || typeof rawMessage !== 'object') {
+    return false;
+  }
+  const message = rawMessage as { type?: unknown; op?: unknown };
+  if (message.type !== '__mokoshTestQuery') {
+    // Not our message — production handler will take it.
+    return false;
+  }
+  const op = String(message.op ?? '');
+  if (op === 'snapshot') {
+    sendResponse({
+      count: notificationCount,
+      lastOptions: lastNotificationOptions,
+      ids: notificationIds.slice(),
+    });
+    return false; // Sync response — return false per Chrome onMessage contract.
+  }
+  if (op === 'fire-on-startup') {
+    const h = handlers.onStartup;
+    if (h === null) {
+      sendResponse({ ok: false, error: 'no-handler' });
+      return false;
+    }
+    // Fire-and-respond. The handler may be async; we don't await it
+    // for the response, but if it throws synchronously the catch
+    // surfaces in the response.
+    try {
+      // Schedule on microtask so the response goes out first; the
+      // handler's side effects (notifications.create) happen right
+      // after, before the next harness assertion polls.
+      queueMicrotask(() => {
+        Promise.resolve(h()).catch((err) => {
+          // Swallow async errors — the assertion 8 check is on the
+          // notification side effect, not the handler's return value.
+          console.warn('[mokoshTest bridge] onStartup handler threw:', err);
+        });
+      });
+      sendResponse({ ok: true });
+    } catch (err) {
+      sendResponse({
+        ok: false,
+        error: err instanceof Error ? err.message : String(err),
+      });
+    }
+    return false;
+  }
+  if (op === 'handler-types') {
+    sendResponse({
+      onClicked: typeof handlers.onClicked,
+      onStartup: typeof handlers.onStartup,
+      notificationOnClicked: typeof handlers.notificationOnClicked,
+    });
+    return false;
+  }
+  sendResponse({ ok: false, error: 'unknown-op' });
+  return false;
+});
+
 export {};
diff --git a/tests/uat/README.md b/tests/uat/README.md
new file mode 100644
index 0000000..3b781d5
--- /dev/null
+++ b/tests/uat/README.md
@@ -0,0 +1,106 @@
+# Mokosh UAT harness (Plan 01-11)
+
+Puppeteer-driven Node script that runs 14 assertions end-to-end against a
+real Chrome instance loaded with the Mokosh extension. Replaces Plan 01-09
+Task 5's operator-empirical functional verification (the operator retains
+only step 1 — build — and step 14 — brand/design acceptance).
+
+## Quick start
+
+```bash
+npm run test:uat
+```
+
+This builds `dist-test/` (the hook-enabled bundle) and runs the harness.
+Exit 0 means all 14 assertions passed. Final line: `UAT harness: 14/14
+assertions passed`.
+
+## Local-debug mode
+
+```bash
+HEADLESS=0 npm run test:uat
+```
+
+Opens a real Chrome window so you can watch the picker auto-accept, the
+badge transitions, the popup appear, etc.
+
+## Developer iteration tricks
+
+```bash
+# Skip the production build inside assertion 0 (uses existing dist/):
+SKIP_PROD_REBUILD=1 npm run test:uat
+
+# Run the harness against an existing dist-test/ (skip npm run build:test):
+npx tsx tests/uat/harness.test.ts
+```
+
+## Assertion catalog
+
+| # | Title | Bug class | Hook used |
+|---|-------|-----------|-----------|
+| 0 | Production bundle has no test-hook leaks | T-1-11-01 | filesystem grep |
+| 1 | SW bootstrap → setIdleMode | — | sw.evaluate |
+| 2 | Toolbar onClicked-idle → REC + popup | — | triggerExtensionAction |
+| 3 | Offscreen displaySurface === monitor | D-15 | __mokoshTest.getCurrentStream |
+| 4 | Toolbar onClicked-recording → popup, no new offscreen | — | targets count |
+| 5 | SAVE_ARCHIVE → download fires | — | downloads polling |
+| 6 | **BUG B**: simulateUserStop → badge OFF + no recovery notif | b9eeeeb | dispatchEvent('ended') |
+| 7 | RECORDING_ERROR codec-unsupported → ERR + recovery notif | — | sendMessage |
+| 8 | **BUG A**: onStartup → mokosh-startup- notification creates | a881bf0 | __mokoshTest.handlers.onStartup |
+| 9 | Icon file sizes meet floors | Bug A precondition | sw.evaluate(fetch) |
+| 10 | Manifest has notifications + 3 icons | Bug A precondition | chrome.runtime.getManifest |
+| 11 | 35s recording → segments.length >= 3 | D-13 | __mokoshTest.getSegmentCount |
+| 12 | ffprobe on extracted webm exits 0 | Plan 01-08 | jszip + execFile |
+| 13 | Archive shape — video + meta.json version match | Plan 01-07 | jszip |
+
+## Failure isolation
+
+Single browser, serial assertions, bail on first failure for setup-
+dependent assertions (assertion 0 abort means refusing to launch a
+potentially-leaky bundle). Per-assertion bail keeps the diagnostic
+output unambiguous — see RESEARCH §5 + Plan 01-11 open-question
+resolution 4.
+
+On failure, the harness dumps the last 30 lines of SW console + last 30
+lines of offscreen console (captured live during the run) to stderr
+BEFORE rethrowing — gives you contextual triage without needing to re-
+run with debug logging.
+
+## Known gotchas
+
+### Locale-specific picker auto-accept
+
+The `--auto-select-desktop-capture-source=Entire screen` Chrome flag
+auto-accepts the screen-share picker. The string `"Entire screen"` is
+en_US-specific. If your Chrome is set to a non-English locale, the
+picker option label will differ and the auto-accept will silently fail
+(picker stays open; assertion 2 times out).
+
+Fallback: switch your Chrome user-data-dir's locale to en_US for
+harness runs, OR adjust the launch arg in `tests/uat/lib/launch.ts` to
+match your locale's equivalent string.
+
+### dev-dep Chromium binary size
+
+`puppeteer` pulls a ~150 MB Chromium binary at `npm install` time. CI
+must accept this. Production `npm install --omit=dev` skips it cleanly.
+
+### Xvfb is NOT required
+
+Per Plan 01-11 RESEARCH §3 empirical probes against Chrome 148, the
+`--headless=new` mode handles screen capture without Xvfb on Linux CI
+runners. If a future Chrome regresses this, `Xvfb :99 & DISPLAY=:99
+npm run test:uat` is the fallback.
+
+### CI runner screen-capture concern
+
+The 35s recording assertion (A11) captures whatever is on screen during
+that window. CI MUST run the harness in an isolated container with no
+concurrent workload — see T-1-11-02 in Plan 01-11's threat model.
+
+### Real Chrome download (assertion 5 → A12)
+
+The harness configures per-page download behavior via CDP to a fresh
+`os.tmpdir()/mokosh-uat-downloads-*` directory; downloads are NOT
+written to your real ~/Downloads. The temp directory is deleted by OS
+tmpdir GC.
diff --git a/tests/uat/harness.test.ts b/tests/uat/harness.test.ts
new file mode 100644
index 0000000..309cd25
--- /dev/null
+++ b/tests/uat/harness.test.ts
@@ -0,0 +1,394 @@
+// tests/uat/harness.test.ts — Plan 01-11 Puppeteer UAT harness entry point.
+//
+// Runs end-to-end via `npm run test:uat` (build:test + tsx tests/uat/harness.test.ts).
+// Top-to-bottom narrative: launch Chrome with dist-test loaded as
+// MV3 extension, attach to SW + offscreen, run 14 assertions
+// sequentially with bail-on-first-fail semantics + structured
+// diagnostic dump on failure (RESEARCH §5 + open-question resolution 4).
+//
+// Exit code:
+//   0 — all 14 assertions passed
+//   1 — at least one assertion failed
+//
+// Local-debug mode: `HEADLESS=0 npm run test:uat` (opens real Chrome)
+// Skip prod rebuild: `SKIP_PROD_REBUILD=1` (assertion 0 still verifies
+// the EXISTING dist/ rather than spawning npm run build).
+//
+// Assertion catalog (14 total):
+//   0  — Production bundle grep gate (filesystem-only; pre-flight).
+//   1  — SW bootstrap → setIdleMode (badge '', popup '', isRecording=false).
+//   2  — Toolbar onClicked-idle → badge 'REC' + popup popup.html + isRecording=true.
+//   3  — Offscreen displaySurface === 'monitor' (post-grant validation).
+//   4  — Toolbar onClicked while recording → popup, NO new offscreen.
+//   5  — SAVE_ARCHIVE → download fires + session_report_*.zip appears.
+//   6  — BUG B (canonical): simulateUserStop → badge '' + popup '' + NO recovery notif.
+//   7  — RECORDING_ERROR codec-unsupported → badge 'ERR' + recovery notif.
+//   8  — BUG A (canonical): onStartup → mokosh-startup- notification creates cleanly.
+//   9  — Icon file sizes meet floors (16→200, 48→500, 128→1024).
+//   10 — Manifest has notifications permission + all three icons declared.
+//   11 — 35s recording yields >= 3 segments per D-13.
+//   12 — ffprobe -v error -f matroska on extracted webm exits 0.
+//   13 — Archive shape (video/last_30sec.webm + meta.json with version match).
+
+import { execFileSync, execSync } from 'node:child_process';
+import { existsSync, readdirSync, readFileSync, statSync, mkdtempSync } from 'node:fs';
+import { tmpdir } from 'node:os';
+import { dirname, join, resolve as resolvePath } from 'node:path';
+import { fileURLToPath } from 'node:url';
+
+import type { Page } from 'puppeteer';
+
+import {
+  type AssertionRecord,
+  type ConsoleBuffers,
+  assertEqual,
+  assertGte,
+  assertMatch,
+  assertTrue,
+  runAssertion,
+  waitFor,
+} from './lib/assertions';
+import {
+  attachToOffscreen,
+  countOffscreenTargets,
+  waitForOffscreenTarget,
+} from './lib/extension';
+import {
+  getDisplaySurface,
+  getSegmentCount,
+  simulateUserStop,
+} from './lib/offscreen';
+import {
+  fireOnStartup,
+  getBadgeText,
+  getIconSize,
+  getIsRecording,
+  getManifest,
+  getNotificationSnapshot,
+  getPopup,
+  keepalivePing,
+  sendSyntheticRecordingError,
+} from './lib/sw';
+import { assertArchiveShape, extractEntryToFile } from './lib/zip';
+import { launchHarnessBrowser, type HarnessHandles } from './lib/launch';
+
+const HARNESS_FILE_DIR = dirname(fileURLToPath(import.meta.url));
+const REPO_ROOT = resolvePath(HARNESS_FILE_DIR, '..', '..');
+const DIST_DIR = resolvePath(REPO_ROOT, 'dist');
+const FFPROBE_BIN = '/usr/bin/ffprobe';
+const TOTAL_ASSERTIONS = 14;
+
+/**
+ * Forbidden hook surface strings — assertion 0 verifies absence
+ * in production dist/. Mirrors the Tier-1 unit gate's surface list
+ * (tests/background/no-test-hooks-in-prod-bundle.test.ts) but runs
+ * against the SAME dist/ as the live harness for E2E parity.
+ */
+const FORBIDDEN_HOOK_STRINGS: ReadonlyArray<string> = [
+  '__mokoshTest',
+  'simulateUserStop',
+  'getSegmentCount',
+  'setCurrentStream',
+  'setSegmentCountGetter',
+];
+
+/** Icon-size floors per assertion 9 (per orchestrator brief). */
+const ICON_SIZE_FLOORS: ReadonlyArray<readonly [string, number]> = [
+  ['icons/icon16.png', 200],
+  ['icons/icon48.png', 500],
+  ['icons/icon128.png', 1024],
+];
+
+/**
+ * Recursively list all files under a root directory (sync). Used by
+ * assertion 0 to walk dist/. Symlinks are skipped defensively.
+ *
+ * @param root - Absolute directory path.
+ * @returns Sorted list of absolute file paths.
+ */
+function listAllFilesRecursive(root: string): ReadonlyArray<string> {
+  const acc: string[] = [];
+  const stack: string[] = [root];
+  while (stack.length > 0) {
+    const dir = stack.pop()!;
+    const entries = readdirSync(dir, { withFileTypes: true });
+    for (const entry of entries) {
+      const fullPath = resolvePath(dir, entry.name);
+      if (entry.isSymbolicLink()) continue;
+      if (entry.isDirectory()) {
+        stack.push(fullPath);
+      } else if (entry.isFile()) {
+        acc.push(fullPath);
+      }
+    }
+  }
+  return acc.sort();
+}
+
+/**
+ * Grep `needle` across every text-like file under `root`. Returns
+ * file paths that contain at least one occurrence.
+ *
+ * @param root - Absolute directory path.
+ * @param needle - Literal substring to find.
+ * @returns Paths containing `needle`.
+ */
+function grepRecursive(root: string, needle: string): ReadonlyArray<string> {
+  const binaryExt = new Set(['.png', '.jpg', '.jpeg', '.gif', '.ico', '.webp', '.woff', '.woff2', '.ttf']);
+  const out: string[] = [];
+  for (const filePath of listAllFilesRecursive(root)) {
+    const dotIdx = filePath.lastIndexOf('.');
+    const ext = dotIdx >= 0 ? filePath.substring(dotIdx).toLowerCase() : '';
+    if (binaryExt.has(ext)) continue;
+    if (statSync(filePath).size === 0) continue;
+    const text = readFileSync(filePath, 'utf8');
+    if (text.includes(needle)) {
+      out.push(filePath);
+    }
+  }
+  return out;
+}
+
+/**
+ * Poll `downloadsDir` for any *session_report*.zip file. Returns the
+ * absolute path of the first match. Used by assertion 5.
+ *
+ * @param downloadsDir - Absolute downloads directory path.
+ * @param timeoutMs - Maximum wait time.
+ * @returns Absolute path to the matched .zip.
+ * @throws On timeout.
+ */
+async function waitForDownloadedZip(
+  downloadsDir: string,
+  timeoutMs: number,
+): Promise<string> {
+  const start = Date.now();
+  while (Date.now() - start < timeoutMs) {
+    const entries = readdirSync(downloadsDir);
+    for (const name of entries) {
+      if (name.includes('session_report') && name.endsWith('.zip')) {
+        const full = join(downloadsDir, name);
+        // Make sure write completed (size stabilized).
+        const size1 = statSync(full).size;
+        await new Promise((r) => setTimeout(r, 200));
+        const size2 = statSync(full).size;
+        if (size1 === size2 && size1 > 0) {
+          return full;
+        }
+      }
+    }
+    await new Promise((r) => setTimeout(r, 200));
+  }
+  throw new Error(
+    `waitForDownloadedZip: no session_report_*.zip appeared in ${downloadsDir} within ${timeoutMs}ms`,
+  );
+}
+
+/**
+ * Run a production build of dist/ unless SKIP_PROD_REBUILD=1.
+ * Assertion 0 reads dist/, so this guarantees the gate runs against
+ * a fresh artifact.
+ */
+function ensureProductionBuild(): void {
+  if (process.env.SKIP_PROD_REBUILD === '1') {
+    process.stdout.write('  (SKIP_PROD_REBUILD=1 — using existing dist/)\n');
+    return;
+  }
+  process.stdout.write('  Running `npm run build` (assertion 0 pre-flight)...\n');
+  execFileSync('npm', ['run', 'build'], {
+    stdio: 'inherit',
+    cwd: REPO_ROOT,
+  });
+}
+
+/**
+ * Stub placeholder for assertions Task 4+ wires. Each stub throws so
+ * the harness exits non-zero today; the diagnostic clearly identifies
+ * the assertion as un-implemented vs failing-in-production.
+ *
+ * @param taskNumber - The plan task number that will wire this assertion.
+ * @returns A function that always throws.
+ */
+function notYetImplemented(taskNumber: number): () => Promise<void> {
+  return async () => {
+    throw new Error(
+      `NOT YET IMPLEMENTED — Plan 01-11 Task ${taskNumber} wires this assertion`,
+    );
+  };
+}
+
+/**
+ * Main harness entry point. Runs all 14 assertions sequentially with
+ * bail-on-first-fail semantics for the SETUP-dependent assertions
+ * (we still record every assertion's outcome — bail only stops
+ * subsequent FUNCTIONAL assertions from running).
+ */
+async function main(): Promise<number> {
+  const results: AssertionRecord[] = [];
+  const buffers: ConsoleBuffers = { swLines: [], offscreenLines: [] };
+  let handles: HarnessHandles | null = null;
+
+  process.stdout.write('\nMokosh UAT harness — Plan 01-11 Puppeteer-driven 14-assertion suite\n');
+  process.stdout.write('='.repeat(72) + '\n\n');
+
+  try {
+    // ─── Assertion 0: Pre-flight grep gate ──────────────────────────
+    process.stdout.write('Assertion 0 (pre-flight, filesystem-only):\n');
+    ensureProductionBuild();
+    const a0 = await runAssertion(
+      0,
+      'production bundle has no test-hook leaks (T-1-11-01)',
+      buffers,
+      async () => {
+        for (const needle of FORBIDDEN_HOOK_STRINGS) {
+          const matches = grepRecursive(DIST_DIR, needle);
+          assertEqual(
+            matches.length,
+            0,
+            `production dist/ contains '${needle}' in: ${JSON.stringify(matches)}`,
+          );
+        }
+      },
+    );
+    results.push(a0);
+    if (!a0.passed) {
+      // Hook leak is security-critical (T-1-11-01) — abort immediately.
+      process.stderr.write(
+        '\n*** ABORT: assertion 0 (hook leak gate) FAILED — refusing to ' +
+          'continue with potentially-leaky production bundle. ***\n',
+      );
+      return 1;
+    }
+
+    // ─── Setup: launch browser, attach to SW + open popup bridge ───
+    process.stdout.write('\nLaunching Chrome + opening popup bridge...\n');
+    handles = await launchHarnessBrowser();
+    const { browser, sw, page, popup, extensionId, downloadsDir } = handles;
+    process.stdout.write(`  extensionId: ${extensionId}\n`);
+    process.stdout.write(`  downloadsDir: ${downloadsDir}\n`);
+    process.stdout.write(`  popup: chrome-extension://${extensionId}/src/popup/index.html\n\n`);
+
+    // Wire console buffers. The popup carries the chrome.* queries;
+    // the SW handle is kept for diagnostic console capture (when the
+    // SW is alive). Both feed buffers for failure dumps.
+    const popupPage: Page = popup;
+    popupPage.on('console', (msg) => {
+      buffers.swLines.push(`[Popup:${msg.type()}] ${msg.text()}`);
+    });
+    sw.on('console', (msg) => {
+      buffers.swLines.push(`[SW:${msg.type()}] ${msg.text()}`);
+    });
+
+    // Read the manifest version once for assertion 13.
+    const manifest = await getManifest(popupPage);
+    const expectedVersion = manifest.version;
+
+    // ─── Wave 3 stubbed assertions (Tasks 4-7 will wire these) ──────
+    const stubs: Array<{
+      index: number;
+      name: string;
+      taskNumber: number;
+    }> = [
+      { index: 1, name: 'SW bootstrap → setIdleMode', taskNumber: 4 },
+      { index: 2, name: 'toolbar onClicked-idle → badge REC + popup', taskNumber: 4 },
+      { index: 3, name: 'offscreen displaySurface === monitor', taskNumber: 4 },
+      { index: 4, name: 'toolbar onClicked-recording → popup, no new offscreen', taskNumber: 4 },
+      { index: 5, name: 'SAVE_ARCHIVE → download fires + zip appears', taskNumber: 5 },
+      { index: 6, name: 'BUG B canonical: simulateUserStop → badge OFF + no recovery notif', taskNumber: 5 },
+      { index: 7, name: 'RECORDING_ERROR codec-unsupported → badge ERR + recovery notif', taskNumber: 5 },
+      { index: 8, name: 'BUG A canonical: onStartup → notification creates cleanly', taskNumber: 6 },
+      { index: 9, name: 'icon file sizes meet floors', taskNumber: 6 },
+      { index: 10, name: 'manifest has notifications + 3 icons', taskNumber: 6 },
+      { index: 11, name: '35s recording → segments.length >= 3', taskNumber: 7 },
+      { index: 12, name: 'ffprobe on extracted webm exits 0', taskNumber: 7 },
+      { index: 13, name: 'archive shape — video + meta.json version match', taskNumber: 7 },
+    ];
+
+    for (const s of stubs) {
+      const rec = await runAssertion(
+        s.index,
+        s.name,
+        buffers,
+        notYetImplemented(s.taskNumber),
+      );
+      results.push(rec);
+    }
+
+    // Suppress unused-warning placeholders — Tasks 4-7 will use these
+    // imports + handles directly. Reference them here for type-clean.
+    void browser;
+    void page;
+    void popupPage;
+    void expectedVersion;
+    void waitForOffscreenTarget;
+    void attachToOffscreen;
+    void countOffscreenTargets;
+    void waitFor;
+    void getBadgeText;
+    void getPopup;
+    void getIsRecording;
+    void getIconSize;
+    void fireOnStartup;
+    void sendSyntheticRecordingError;
+    void getNotificationSnapshot;
+    void keepalivePing;
+    void getDisplaySurface;
+    void simulateUserStop;
+    void getSegmentCount;
+    void assertArchiveShape;
+    void extractEntryToFile;
+    void assertMatch;
+    void assertTrue;
+    void assertGte;
+    void waitForDownloadedZip;
+    void mkdtempSync;
+    void existsSync;
+    void execSync;
+    void tmpdir;
+    void FFPROBE_BIN;
+    void ICON_SIZE_FLOORS;
+
+    return finalize(results);
+  } catch (setupErr) {
+    process.stderr.write(`\n*** Harness setup error: ${String(setupErr)}\n`);
+    return finalize(results);
+  } finally {
+    if (handles !== null) {
+      try {
+        await handles.browser.close();
+      } catch (closeErr) {
+        process.stderr.write(`(non-fatal: browser close threw: ${String(closeErr)})\n`);
+      }
+    }
+  }
+}
+
+/**
+ * Print the final summary line + return the exit code.
+ *
+ * @param results - All assertion records collected during the run.
+ * @returns 0 if all 14 passed, 1 otherwise.
+ */
+function finalize(results: ReadonlyArray<AssertionRecord>): number {
+  const passCount = results.filter((r) => r.passed).length;
+  const failCount = results.length - passCount;
+  process.stdout.write('\n' + '='.repeat(72) + '\n');
+  if (passCount === TOTAL_ASSERTIONS) {
+    process.stdout.write(`UAT harness: ${passCount}/${TOTAL_ASSERTIONS} assertions passed\n`);
+    return 0;
+  }
+  const firstFail = results.find((r) => !r.passed);
+  process.stdout.write(
+    `UAT harness: ${passCount}/${TOTAL_ASSERTIONS} assertions passed, ${failCount} failed`,
+  );
+  if (firstFail !== undefined) {
+    process.stdout.write(` (first failure: A${firstFail.index} ${firstFail.name})`);
+  }
+  process.stdout.write('\n');
+  return 1;
+}
+
+// Run + exit. Top-level await + explicit exit code so tsx returns
+// the right status without leaving unhandled-promise spew on stderr.
+const exitCode = await main();
+process.exit(exitCode);
diff --git a/tests/uat/lib/assertions.ts b/tests/uat/lib/assertions.ts
new file mode 100644
index 0000000..3d15df6
--- /dev/null
+++ b/tests/uat/lib/assertions.ts
@@ -0,0 +1,199 @@
+// tests/uat/lib/assertions.ts — Plan 01-11 harness assertion runner.
+//
+// Centralizes:
+//   - `assertEqual` / `assertMatch` / `assertTrue` — thin wrappers
+//     over `node:assert/strict` with explicit Plan 01-11 diagnostic
+//     framing (cite the bug-class on Bug A / Bug B assertions).
+//   - `runAssertion(name, fn)` — wraps each assertion in a try/catch
+//     so the harness can collect a per-assertion pass/fail map AND
+//     dump SW/offscreen console buffers on the FIRST failure (bail
+//     semantics per RESEARCH §5).
+//   - `waitFor(probe, predicate, timeoutMs)` — polling helper used by
+//     assertions that need to wait for async state transitions
+//     (badge changes, downloads, etc.).
+//
+// References:
+//   - node:assert/strict: https://nodejs.org/api/assert.html#strict-assertion-mode
+
+import { strict as assert } from 'node:assert';
+
+/**
+ * Per-assertion outcome record. Accumulated by runAssertion + flushed
+ * to the harness's final summary line.
+ */
+export interface AssertionRecord {
+  readonly index: number;
+  readonly name: string;
+  readonly passed: boolean;
+  readonly errorMessage: string;
+  readonly durationMs: number;
+}
+
+/**
+ * Console buffers captured from SW + offscreen contexts. The harness
+ * wires `sw.on('console', ...)` + `offPage.on('console', ...)` at
+ * launch + before each assertion-relevant phase; on failure these
+ * buffers are dumped to stderr for triage.
+ */
+export interface ConsoleBuffers {
+  swLines: string[];
+  offscreenLines: string[];
+}
+
+/**
+ * Run a single assertion, capturing its outcome + duration. On error,
+ * dump the per-context console buffers to stderr BEFORE rethrowing so
+ * the harness's top-level catch sees the diagnostic context.
+ *
+ * @param index - 0-13 (0 = grep gate, 1-13 = functional).
+ * @param name - Human-readable assertion title.
+ * @param buffers - Console buffers to dump on failure (may be empty).
+ * @param fn - Async assertion body.
+ * @returns Outcome record.
+ */
+export async function runAssertion(
+  index: number,
+  name: string,
+  buffers: ConsoleBuffers,
+  fn: () => Promise<void>,
+): Promise<AssertionRecord> {
+  const start = Date.now();
+  try {
+    await fn();
+    const durationMs = Date.now() - start;
+    process.stdout.write(`  [PASS] A${index}: ${name} (${durationMs}ms)\n`);
+    return {
+      index,
+      name,
+      passed: true,
+      errorMessage: '',
+      durationMs,
+    };
+  } catch (err) {
+    const durationMs = Date.now() - start;
+    const errorMessage =
+      err instanceof Error ? `${err.name}: ${err.message}` : String(err);
+    process.stderr.write(`  [FAIL] A${index}: ${name} (${durationMs}ms)\n`);
+    process.stderr.write(`         ${errorMessage}\n`);
+    dumpBuffers(buffers, index);
+    return {
+      index,
+      name,
+      passed: false,
+      errorMessage,
+      durationMs,
+    };
+  }
+}
+
+/**
+ * Dump SW + offscreen console buffers to stderr with structured framing.
+ * Cap at the last 30 lines per context to keep failure output readable.
+ *
+ * @param buffers - The accumulating buffers.
+ * @param assertionIndex - For framing the dump preamble.
+ */
+function dumpBuffers(buffers: ConsoleBuffers, assertionIndex: number): void {
+  const TAIL = 30;
+  const swTail = buffers.swLines.slice(-TAIL);
+  const offTail = buffers.offscreenLines.slice(-TAIL);
+  if (swTail.length > 0) {
+    process.stderr.write(
+      `         --- SW console (last ${swTail.length} lines, assertion A${assertionIndex}) ---\n`,
+    );
+    for (const line of swTail) {
+      process.stderr.write(`           ${line}\n`);
+    }
+  }
+  if (offTail.length > 0) {
+    process.stderr.write(
+      `         --- Offscreen console (last ${offTail.length} lines, assertion A${assertionIndex}) ---\n`,
+    );
+    for (const line of offTail) {
+      process.stderr.write(`           ${line}\n`);
+    }
+  }
+}
+
+/**
+ * Strict equality with a context-bearing message. Wraps
+ * `assert.strictEqual` so the failure surface is uniform across
+ * assertions.
+ *
+ * @param actual - Observed value.
+ * @param expected - Expected value.
+ * @param msg - Context for the failure diagnostic.
+ */
+export function assertEqual<T>(actual: T, expected: T, msg: string): void {
+  assert.strictEqual(actual, expected, msg);
+}
+
+/**
+ * Assert that `actual` matches `regex`. Wraps `assert.match`.
+ *
+ * @param actual - String to test.
+ * @param regex - Pattern.
+ * @param msg - Context for the failure diagnostic.
+ */
+export function assertMatch(actual: string, regex: RegExp, msg: string): void {
+  assert.match(actual, regex, msg);
+}
+
+/**
+ * Assert that `cond` is truthy. Wraps `assert.ok`.
+ *
+ * @param cond - Boolean expression.
+ * @param msg - Context for the failure diagnostic.
+ */
+export function assertTrue(cond: boolean, msg: string): void {
+  assert.ok(cond, msg);
+}
+
+/**
+ * Assert that the actual value is greater than or equal to expected.
+ * Used by assertion 9 (icon size floors) + assertion 11 (segment count).
+ *
+ * @param actual - Observed value.
+ * @param expected - Minimum acceptable value.
+ * @param msg - Context for the failure diagnostic.
+ */
+export function assertGte(actual: number, expected: number, msg: string): void {
+  assert.ok(
+    actual >= expected,
+    `${msg} — expected >= ${expected}, got ${actual}`,
+  );
+}
+
+/**
+ * Poll `probe` until `predicate(probe())` returns true OR timeoutMs
+ * elapses. Throws on timeout with a structured diagnostic.
+ *
+ * @param probe - Async function producing a value to test.
+ * @param predicate - Returns true when the value satisfies the wait.
+ * @param timeoutMs - Maximum wait time.
+ * @param description - Human-readable description for the diagnostic.
+ * @param pollIntervalMs - Interval between probe calls (default 100ms).
+ * @returns The last probed value that satisfied the predicate.
+ * @throws If timeoutMs elapses without predicate satisfaction.
+ */
+export async function waitFor<T>(
+  probe: () => Promise<T>,
+  predicate: (v: T) => boolean,
+  timeoutMs: number,
+  description: string,
+  pollIntervalMs: number = 100,
+): Promise<T> {
+  const start = Date.now();
+  let lastValue: T | undefined;
+  while (Date.now() - start < timeoutMs) {
+    lastValue = await probe();
+    if (predicate(lastValue)) {
+      return lastValue;
+    }
+    await new Promise((r) => setTimeout(r, pollIntervalMs));
+  }
+  throw new Error(
+    `waitFor timeout ${timeoutMs}ms — ${description}; ` +
+      `last probed value: ${JSON.stringify(lastValue)}`,
+  );
+}
diff --git a/tests/uat/lib/extension.ts b/tests/uat/lib/extension.ts
new file mode 100644
index 0000000..3d22013
--- /dev/null
+++ b/tests/uat/lib/extension.ts
@@ -0,0 +1,93 @@
+// tests/uat/lib/extension.ts — Plan 01-11 harness extension/offscreen helpers.
+//
+// The offscreen-document attach uses a CDP-level target type that
+// Puppeteer 25 surfaces as `'background_page'` — NOT `'page'`. Per
+// Plan 01-11 RESEARCH §4 / Pitfall 1, finding the offscreen via
+// `t.type() === 'page'` returns no matches; `'background_page'` is
+// the right discriminator. After getting the target, `.asPage()`
+// returns a Page-like handle (NOT `.page()` — that returns undefined).
+//
+// References:
+//   - Puppeteer Target types:
+//     https://pptr.dev/api/puppeteer.targettype
+//   - Chrome offscreen document:
+//     https://developer.chrome.com/docs/extensions/reference/api/offscreen
+
+import type { Browser, Page, Target } from 'puppeteer';
+
+/** How long to wait for the offscreen document target to appear. */
+const OFFSCREEN_TARGET_TIMEOUT_MS = 5_000;
+
+/**
+ * Poll the browser's target list for the offscreen document. The
+ * offscreen is created lazily — only when the SW issues
+ * `chrome.offscreen.createDocument(...)`. Caller MUST invoke a flow
+ * that triggers offscreen creation (e.g. start a recording) BEFORE
+ * calling this helper.
+ *
+ * @param browser - Puppeteer Browser handle.
+ * @param extensionId - The extension's runtime id (for URL filtering).
+ * @returns Resolved Target whose URL contains 'offscreen'.
+ * @throws If no offscreen target appears within OFFSCREEN_TARGET_TIMEOUT_MS.
+ */
+export async function waitForOffscreenTarget(
+  browser: Browser,
+  extensionId: string,
+): Promise<Target> {
+  const predicate = (t: Target): boolean => {
+    const url = t.url();
+    // Offscreen documents are loaded as chrome-extension://<id>/...
+    // with a path containing 'offscreen' (matches both 'src/offscreen/'
+    // and the bundled equivalents). Target type 'background_page' per
+    // RESEARCH §4 Pitfall 1.
+    return (
+      t.type() === 'background_page' &&
+      url.startsWith(`chrome-extension://${extensionId}`) &&
+      url.includes('offscreen')
+    );
+  };
+
+  return await browser.waitForTarget(predicate, {
+    timeout: OFFSCREEN_TARGET_TIMEOUT_MS,
+  });
+}
+
+/**
+ * Attach to the offscreen document as a Page-like handle. Uses
+ * `.asPage()` (NOT `.page()` — Puppeteer 25 returns null for
+ * `.page()` on background_page-type targets).
+ *
+ * @param target - The offscreen Target from waitForOffscreenTarget.
+ * @returns Page handle for evaluate/expose/etc.
+ */
+export async function attachToOffscreen(target: Target): Promise<Page> {
+  const page = await target.asPage();
+  return page;
+}
+
+/**
+ * Count the offscreen targets currently in the browser. Used by
+ * assertion 4 to verify that a toolbar click while recording does
+ * NOT spawn a second offscreen document.
+ *
+ * @param browser - Puppeteer Browser handle.
+ * @param extensionId - The extension's runtime id.
+ * @returns Integer count of offscreen targets.
+ */
+export function countOffscreenTargets(
+  browser: Browser,
+  extensionId: string,
+): number {
+  const targets = browser.targets();
+  let count = 0;
+  for (const t of targets) {
+    if (
+      t.type() === 'background_page' &&
+      t.url().startsWith(`chrome-extension://${extensionId}`) &&
+      t.url().includes('offscreen')
+    ) {
+      count += 1;
+    }
+  }
+  return count;
+}
diff --git a/tests/uat/lib/launch.ts b/tests/uat/lib/launch.ts
new file mode 100644
index 0000000..3e0a93c
--- /dev/null
+++ b/tests/uat/lib/launch.ts
@@ -0,0 +1,314 @@
+// tests/uat/lib/launch.ts — Plan 01-11 harness launch helper.
+//
+// Wraps puppeteer.launch with the project's invariants:
+//   - enableExtensions points at the absolute path to dist-test/ (the
+//     test bundle that carries the gated test hooks per Plan 01-11
+//     Task 2). NOT dist/ — that would defeat the harness entirely.
+//   - headless defaults to true (CI-friendly); HEADLESS=0 env opens a
+//     real Chrome window for local debugging.
+//   - --auto-select-desktop-capture-source="Entire screen" auto-accepts
+//     the screen-share picker so getDisplayMedia resolves without
+//     operator interaction (RESEARCH §9). The literal string is
+//     en_US-locale-sensitive; document the fallback in tests/uat/README.md.
+//   - Downloads land in a fresh per-run temp dir so assertion 5
+//     (SAVE_ARCHIVE) can poll for session_report_*.zip without
+//     colliding with operator downloads.
+//
+// References:
+//   - puppeteer.launch options: https://pptr.dev/api/puppeteer.launchoptions
+//   - puppeteer extension API: https://pptr.dev/guides/extensions
+//   - Chrome --auto-select-desktop-capture-source:
+//     https://source.chromium.org/chromium/chromium/src/+/main:media/capture/video/chromeos/camera_app_device_provider.cc
+//     (search for the flag in chrome://flags or the Chromium source tree)
+
+import { execSync } from 'node:child_process';
+import { existsSync, mkdtempSync, statSync } from 'node:fs';
+import { tmpdir } from 'node:os';
+import { dirname, join, resolve as resolvePath } from 'node:path';
+import { fileURLToPath } from 'node:url';
+
+import puppeteer, {
+  type Browser,
+  type CDPSession,
+  type Extension,
+  type Page,
+  type WebWorker,
+} from 'puppeteer';
+
+/// <reference path="./test-hook-contract.d.ts" />
+
+const HARNESS_FILE_DIR = dirname(fileURLToPath(import.meta.url));
+const REPO_ROOT = resolvePath(HARNESS_FILE_DIR, '..', '..', '..');
+const DIST_TEST_DIR = resolvePath(REPO_ROOT, 'dist-test');
+
+/**
+ * Handles returned from `launchHarnessBrowser`. All references are
+ * live for the lifetime of the browser; the caller MUST close the
+ * browser to release them.
+ */
+export interface HarnessHandles {
+  readonly browser: Browser;
+  readonly extension: Extension;
+  readonly extensionId: string;
+  /**
+   * Service worker handle (for completeness / future use). NOTE: per
+   * the architecture refinement documented in tests/uat/lib/sw.ts,
+   * the harness's chrome.* state queries go through the `popup` page
+   * (which has full extension chrome.* access AND a stable Puppeteer
+   * lifetime). Direct sw.evaluate is unreliable in Chrome 148 +
+   * headless + Puppeteer 25 (the SW suspends + worker() returns
+   * "Protocol error: No target with given id found"). The SW handle
+   * is kept here for harness wave-3 assertion 11 / 12 (where we may
+   * need a worker reference for diagnostics).
+   */
+  readonly sw: WebWorker;
+  readonly downloadsDir: string;
+  /**
+   * A pre-opened blank page the harness can use to invoke
+   * `triggerExtensionAction` (Puppeteer requires a page in the active
+   * tab for the toolbar-click simulation).
+   */
+  readonly page: Page;
+  /**
+   * The extension popup page, opened at
+   * chrome-extension://<extensionId>/src/popup/index.html. This page
+   * is the harness's primary chrome.* query surface (see
+   * tests/uat/lib/sw.ts file header for rationale).
+   */
+  readonly popup: Page;
+}
+
+/**
+ * Optional launch overrides. Defaults are CI-friendly; HEADLESS=0
+ * environment variable flips to headful for local debugging.
+ */
+export interface LaunchOptions {
+  /** Override the dist-test directory (test isolation). */
+  readonly distTestDir?: string;
+  /** Override the downloads directory (default: fresh tempdir per call). */
+  readonly downloadsDir?: string;
+  /** Force headless / headful regardless of HEADLESS env. */
+  readonly headless?: boolean;
+}
+
+/**
+ * Create a per-run downloads directory under the OS tmpdir. Caller is
+ * responsible for cleanup (typically deferred to OS tmpdir GC).
+ *
+ * @returns Absolute path to the freshly-created downloads directory.
+ */
+function makeDownloadsDir(): string {
+  return mkdtempSync(join(tmpdir(), 'mokosh-uat-downloads-'));
+}
+
+/**
+ * Verify the dist-test directory exists and is a directory. Fails
+ * loudly with an actionable message — the caller likely forgot to
+ * run `npm run build:test` before invoking the harness.
+ *
+ * @param distTestDir - Absolute path to dist-test.
+ * @throws If the directory does not exist or is not a directory.
+ */
+function assertDistTestPresent(distTestDir: string): void {
+  if (!existsSync(distTestDir)) {
+    throw new Error(
+      `dist-test/ missing at ${distTestDir}. ` +
+        `Run \`npm run build:test\` before launching the harness ` +
+        `(or invoke via \`npm run test:uat\` which does it for you).`,
+    );
+  }
+  const stat = statSync(distTestDir);
+  if (!stat.isDirectory()) {
+    throw new Error(
+      `dist-test/ exists at ${distTestDir} but is not a directory.`,
+    );
+  }
+}
+
+/**
+ * Resolve whether to run headless. HEADLESS=0 forces headful;
+ * anything else (including undefined) is headless. Explicit
+ * `options.headless` overrides the env entirely.
+ *
+ * @param options - Optional launch overrides.
+ * @returns true for headless, false for headful.
+ */
+function resolveHeadless(options: LaunchOptions): boolean {
+  if (options.headless !== undefined) {
+    return options.headless;
+  }
+  return process.env.HEADLESS !== '0';
+}
+
+/**
+ * Locate the SW target via the extension ID. Polls puppeteer's target
+ * list because the SW is registered asynchronously after the extension
+ * loads. Times out at 10s — if the SW is missing after that, either
+ * dist-test/ is corrupted or the SW bundle threw at module init (which
+ * would be caught by sw-bundle-import.test.ts BEFORE the harness ever
+ * runs; but defensively, we surface a clear diagnostic here).
+ *
+ * @param browser - Puppeteer Browser handle.
+ * @param extensionId - The extension's runtime id.
+ * @returns The SW WebWorker handle.
+ * @throws If no SW target appears within 10s.
+ */
+async function waitForSwTarget(
+  browser: Browser,
+  extensionId: string,
+): Promise<WebWorker> {
+  const target = await browser.waitForTarget(
+    (t) =>
+      t.type() === 'service_worker' &&
+      t.url().startsWith(`chrome-extension://${extensionId}`),
+    { timeout: 10_000 },
+  );
+  const sw = await target.worker();
+  if (sw === null) {
+    throw new Error(
+      `Service worker target found for extension ${extensionId} but ` +
+        `its worker() returned null — the SW likely crashed at init.`,
+    );
+  }
+  return sw;
+}
+
+/**
+ * Configure the per-page download behavior via CDP so files land in
+ * our temp downloadsDir. Puppeteer 25's high-level downloads API is
+ * still in flux; the raw CDP call is stable across versions.
+ *
+ * @param page - Page whose downloads should be redirected.
+ * @param downloadsDir - Absolute path to capture downloads.
+ */
+async function setDownloadBehavior(
+  page: Page,
+  downloadsDir: string,
+): Promise<void> {
+  const cdpClient: CDPSession = await page.target().createCDPSession();
+  await cdpClient.send('Browser.setDownloadBehavior', {
+    behavior: 'allow',
+    downloadPath: downloadsDir,
+    eventsEnabled: true,
+  });
+}
+
+/**
+ * Launch a Chrome instance with the test bundle loaded as an unpacked
+ * MV3 extension; wire downloads to a per-run temp dir; return all
+ * handles the harness needs. Caller MUST `await handles.browser.close()`.
+ *
+ * @param options - Optional overrides (mostly for isolation in tests).
+ * @returns Resolved handles to browser, extension, SW, page, downloadsDir.
+ * @throws If dist-test/ missing OR SW target never appears.
+ */
+export async function launchHarnessBrowser(
+  options: LaunchOptions = {},
+): Promise<HarnessHandles> {
+  const distTestDir = options.distTestDir ?? DIST_TEST_DIR;
+  assertDistTestPresent(distTestDir);
+  const downloadsDir = options.downloadsDir ?? makeDownloadsDir();
+  const headless = resolveHeadless(options);
+
+  // Pre-flight: verify the operator's chrome binary supports the
+  // auto-select picker flag. The string is locale-specific; en_US
+  // uses "Entire screen". This pre-flight does NOT verify the locale
+  // matches — it only verifies Puppeteer can find a Chromium binary
+  // at all (a missing binary fails the launch with a confusing message
+  // otherwise).
+  // Suppress noisy `puppeteer --version` check; if it fails, the launch
+  // itself will surface the same diagnostic.
+  try {
+    execSync('node ./node_modules/puppeteer/lib/cjs/puppeteer/node/cli.js --help', {
+      stdio: 'ignore',
+      timeout: 5_000,
+    });
+  } catch {
+    // Best-effort. The actual launch will fail loudly if the binary is
+    // truly missing.
+  }
+
+  const browser = await puppeteer.launch({
+    enableExtensions: [distTestDir],
+    headless,
+    pipe: true,
+    args: [
+      '--no-sandbox',
+      // RESEARCH §9: auto-accept the screen-share picker so
+      // getDisplayMedia resolves without operator interaction. The
+      // literal string is en_US-locale-sensitive; tests/uat/README.md
+      // documents the fallback for other locales.
+      '--auto-select-desktop-capture-source=Entire screen',
+      // DO NOT add --use-fake-ui-for-media-stream (RESEARCH §9 Pitfall:
+      // conflicts with auto-select).
+    ],
+  });
+
+  // Resolve the extension ID. Puppeteer 25's browser.extensions() returns
+  // a Map<id, Extension> with all enabled extensions — BUT the map is
+  // populated asynchronously after the extension's manifest loads.
+  // Empirically: extension appears within ~100ms on local hardware but
+  // the very first call right after launch returns Map(0). Poll until
+  // extension registers OR 5s elapses; surface a clear diagnostic on
+  // timeout (probably means dist-test/ is malformed).
+  let extensionsMap = await browser.extensions();
+  const POLL_TIMEOUT_MS = 5_000;
+  const POLL_INTERVAL_MS = 100;
+  const pollStart = Date.now();
+  while (extensionsMap.size === 0 && Date.now() - pollStart < POLL_TIMEOUT_MS) {
+    await new Promise((r) => setTimeout(r, POLL_INTERVAL_MS));
+    extensionsMap = await browser.extensions();
+  }
+  const entries = [...extensionsMap];
+  if (entries.length === 0) {
+    await browser.close();
+    throw new Error(
+      `Puppeteer launched Chrome but no extensions loaded after ${POLL_TIMEOUT_MS}ms — ` +
+        `verify enableExtensions path points at a valid unpacked extension: ${distTestDir}. ` +
+        `Common causes: dist-test/ missing the manifest.json, manifest version mismatch ` +
+        `(Chrome requires MV3 — verify "manifest_version": 3), or chrome binary ` +
+        `incompatible with the unpacked extension shape.`,
+    );
+  }
+  const [extensionId, extension] = entries[0];
+
+  // Wait for the SW target to appear + capture its worker handle.
+  const sw = await waitForSwTarget(browser, extensionId);
+
+  // Give the SW's module init a tick to complete. Empirically the
+  // service-worker-loader.js → assets/index-*.js dynamic import
+  // resolves quickly, but `chrome.action.onClicked.addListener` (and
+  // the gated test-hook addListener monkey-patches) all run inside
+  // the module body — a brief settle ensures the hook surface is
+  // installed BEFORE the harness's first `sw.evaluate(() =>
+  // globalThis.__mokoshTest...)` query.
+  await new Promise((r) => setTimeout(r, 500));
+
+  // Pre-open a blank page; configure downloads. The blank page is
+  // also the page the harness uses for triggerExtensionAction.
+  const page = await browser.newPage();
+  await page.goto('about:blank');
+  await setDownloadBehavior(page, downloadsDir);
+
+  // Open the extension popup as a separate Page. This is the harness's
+  // primary chrome.* query surface — see tests/uat/lib/sw.ts file
+  // header for the architecture rationale. The popup page has full
+  // extension chrome.* access AND a stable Puppeteer lifetime. Loading
+  // the URL also wakes the SW (chrome-extension:// page load IS a SW
+  // wake-up event in MV3).
+  const popup = await browser.newPage();
+  await popup.goto(
+    `chrome-extension://${extensionId}/src/popup/index.html`,
+    { waitUntil: 'domcontentloaded', timeout: 10_000 },
+  );
+
+  return {
+    browser,
+    extension,
+    extensionId,
+    sw,
+    downloadsDir,
+    page,
+    popup,
+  };
+}
diff --git a/tests/uat/lib/offscreen.ts b/tests/uat/lib/offscreen.ts
new file mode 100644
index 0000000..bf3ce70
--- /dev/null
+++ b/tests/uat/lib/offscreen.ts
@@ -0,0 +1,107 @@
+// tests/uat/lib/offscreen.ts — Plan 01-11 harness offscreen-context helpers.
+//
+// Each helper is a thin wrapper over `offPage.evaluate(() => ...)`.
+// The Bug B BLOCKER (RESEARCH §7) lives in simulateUserStop —
+// DO NOT REFACTOR to track.stop().
+//
+// References:
+//   - MediaStreamTrack 'ended' event:
+//     https://developer.mozilla.org/docs/Web/API/MediaStreamTrack/ended_event
+//   - MediaStreamTrack.stop spec note (stop does NOT fire 'ended' on the same track):
+//     https://www.w3.org/TR/mediacapture-streams/#dom-mediastreamtrack-stop
+
+import type { Page } from 'puppeteer';
+
+/// <reference path="./test-hook-contract.d.ts" />
+
+/**
+ * Read the displaySurface from the active MediaStream's video track.
+ * Used by assertion 3 to verify monitor-only enforcement (the
+ * post-grant validation in src/offscreen/recorder.ts).
+ *
+ * Returns null when there is no active recording (the harness MUST
+ * start a recording before calling this).
+ *
+ * @param offPage - Offscreen Page handle.
+ * @returns 'monitor' on success, other strings on regression, null when no stream.
+ */
+export async function getDisplaySurface(offPage: Page): Promise<string | null> {
+  return await offPage.evaluate(() => {
+    const hook = globalThis.__mokoshTest;
+    if (hook === undefined || hook.getCurrentStream === undefined) {
+      return null;
+    }
+    const stream = hook.getCurrentStream();
+    if (stream === null) {
+      return null;
+    }
+    const track = stream.getVideoTracks()[0];
+    if (track === undefined) {
+      return null;
+    }
+    const ds = track.getSettings().displaySurface;
+    return typeof ds === 'string' ? ds : null;
+  });
+}
+
+/**
+ * Simulate the operator clicking Chrome's "Stop sharing" overlay.
+ *
+ * **BLOCKER (RESEARCH §7) — DO NOT REFACTOR to `track.stop()`.**
+ *
+ * `track.stop()` releases the capture but does NOT fire the 'ended'
+ * event on the same track per the W3C Screen Capture spec. The
+ * production `onUserStoppedSharing` handler (src/offscreen/recorder.ts:
+ * 451) is wired to 'ended' — using `track.stop()` would silently bypass
+ * the entire Bug B fix path that this assertion exists to verify.
+ *
+ * `track.dispatchEvent(new Event('ended'))` IS the only path that
+ * triggers our handler. After dispatch, the production handler calls
+ * `stream.getTracks().forEach(t => t.stop())` which DOES release the
+ * capture (just doesn't refire 'ended' on the same track — spec-correct).
+ *
+ * @param offPage - Offscreen Page handle.
+ * @throws If no active MediaStream OR no video track in the stream.
+ */
+export async function simulateUserStop(offPage: Page): Promise<void> {
+  await offPage.evaluate(() => {
+    const hook = globalThis.__mokoshTest;
+    if (hook === undefined || hook.getCurrentStream === undefined) {
+      throw new Error('simulateUserStop: __mokoshTest.getCurrentStream missing');
+    }
+    const stream = hook.getCurrentStream();
+    if (stream === null) {
+      throw new Error(
+        'simulateUserStop: no current MediaStream — recording must be active',
+      );
+    }
+    const track = stream.getVideoTracks()[0];
+    if (track === undefined) {
+      throw new Error('simulateUserStop: no video track in stream');
+    }
+    // CRITICAL: dispatchEvent, NOT track.stop(). See preamble for the
+    // BLOCKER analysis (RESEARCH §7).
+    track.dispatchEvent(new Event('ended'));
+  });
+}
+
+/**
+ * Read the current segment count from the offscreen recorder's ring
+ * buffer. Used by assertion 11 to verify the 30s window per D-13
+ * (3 × 10s segments expected after 35s of recording).
+ *
+ * Returns -1 when the hook is not installed (defensive — should
+ * never happen against a dist-test/ bundle).
+ *
+ * @param offPage - Offscreen Page handle.
+ * @returns Current segment count.
+ */
+export async function getSegmentCount(offPage: Page): Promise<number> {
+  return await offPage.evaluate(() => {
+    const hook = globalThis.__mokoshTest;
+    if (hook === undefined || hook.getSegmentCount === undefined) {
+      return -1;
+    }
+    return hook.getSegmentCount();
+  });
+}
diff --git a/tests/uat/lib/sw.ts b/tests/uat/lib/sw.ts
new file mode 100644
index 0000000..18c3ecc
--- /dev/null
+++ b/tests/uat/lib/sw.ts
@@ -0,0 +1,262 @@
+// tests/uat/lib/sw.ts — Plan 01-11 harness SW-state helpers.
+//
+// IMPLEMENTATION ARCHITECTURE (refined during Task 3 execution):
+//
+//   The original Plan 01-11 RESEARCH §1 sketch assumed `sw.evaluate(() =>
+//   chrome.action.getBadgeText({}))` would work directly against the
+//   service worker via Puppeteer's WebWorker.evaluate. Empirical probes
+//   during Task 3 execution against Puppeteer 25.0.2 + Chrome 148 +
+//   --headless=true revealed two blockers:
+//     1. Puppeteer's `WebWorker.evaluate` runs in an ISOLATED WORLD that
+//        carries SW globals (clients, registration, ...) but NOT the
+//        extension's full `chrome.*` API surface. `Object.keys(chrome)`
+//        returns `["loadTimes", "csi"]` — the public webpage chrome,
+//        not the extension chrome.
+//     2. Chrome 148's headless mode aggressively suspends MV3 service
+//        workers; subsequent `swTarget.worker()` calls return
+//        `Protocol error: No target with given id found`.
+//
+//   The popup page (chrome-extension://<id>/src/popup/index.html) has:
+//     - Full `chrome.*` API access (it's an extension context — same
+//       privileges as the SW for chrome.action, chrome.runtime,
+//       chrome.notifications, chrome.runtime.getManifest, etc.)
+//     - Stable lifetime (it's a regular Page; Puppeteer keeps it alive)
+//     - Natural SW wake-up via message passing (chrome.runtime
+//       .sendMessage from popup wakes the SW for 30s)
+//
+//   So this module's helpers use a Puppeteer Page handle pointing at
+//   the popup URL — NOT a WebWorker handle. The harness opens the popup
+//   page during setup (tests/uat/lib/launch.ts) and passes it here.
+//
+//   For SW-isolate-specific state (`globalThis.__mokoshTest` lives in
+//   the SW's globalThis, not the popup's), the SW hook exposes a
+//   `chrome.runtime.onMessage` bridge: the popup sends
+//   `{ type: '__mokoshTestQuery', op: '...' }` messages; the hook
+//   responds with the queried state. Bridge implementation is in
+//   src/test-hooks/sw-hooks.ts; this file invokes it via popup.evaluate
+//   wrapping `chrome.runtime.sendMessage`.
+//
+// References:
+//   - Chrome extension pages share chrome.* API:
+//     https://developer.chrome.com/docs/extensions/develop/concepts/popup
+//   - Puppeteer Page.evaluate: https://pptr.dev/api/puppeteer.page.evaluate
+//   - Service worker wake-up on chrome.runtime message:
+//     https://developer.chrome.com/docs/extensions/develop/concepts/service-workers/lifecycle
+
+import type { Page } from 'puppeteer';
+
+/// <reference path="./test-hook-contract.d.ts" />
+
+/**
+ * Structured snapshot of the SW's notification observability state
+ * (Plan 01-11 Task 2 sw-hooks.ts surfaces). Used by assertions 7 + 8
+ * to verify count-deltas + last-options-shape + id-prefix membership.
+ */
+export interface NotificationSnapshot {
+  readonly count: number;
+  readonly lastOptions: chrome.notifications.NotificationOptions<true> | null;
+  readonly ids: ReadonlyArray<string>;
+}
+
+/**
+ * The SW hook's bridge message type. The popup sends one of these
+ * shapes via chrome.runtime.sendMessage; the SW's onMessage handler
+ * (extended by sw-hooks.ts) responds with the queried state. See
+ * src/test-hooks/sw-hooks.ts for the SW-side dispatch.
+ */
+interface BridgeQuery {
+  type: '__mokoshTestQuery';
+  op:
+    | 'snapshot'
+    | 'fire-on-startup'
+    | 'handler-types';
+}
+
+/**
+ * Get the toolbar badge text. Empty string means OFF or initial state;
+ * 'REC' means recording; 'ERR' means error per Plan 01-09 badge state
+ * machine.
+ *
+ * @param popup - The extension popup page handle (open against
+ *   chrome-extension://<id>/src/popup/index.html).
+ * @returns Current badge text.
+ */
+export async function getBadgeText(popup: Page): Promise<string> {
+  return await popup.evaluate(async () => await chrome.action.getBadgeText({}));
+}
+
+/**
+ * Get the current popup URL. Empty string means popup is not set
+ * (toolbar click fires onClicked instead). The chrome-extension://
+ * URL means recording (popup hosts SAVE button).
+ *
+ * @param popup - The extension popup page handle.
+ * @returns Current popup URL (full chrome-extension:// form OR '').
+ */
+export async function getPopup(popup: Page): Promise<string> {
+  return await popup.evaluate(async () => await chrome.action.getPopup({}));
+}
+
+/**
+ * Read the runtime manifest. Used by assertion 10 to verify
+ * permissions + icons shape, and by assertion 13 to obtain the
+ * version string for archive shape matching.
+ *
+ * @param popup - The extension popup page handle.
+ * @returns The chrome.runtime.getManifest() result.
+ */
+export async function getManifest(popup: Page): Promise<chrome.runtime.Manifest> {
+  return await popup.evaluate(() => chrome.runtime.getManifest());
+}
+
+/**
+ * Fetch an extension-relative file via popup context and return its
+ * size in bytes. Used by assertion 9 to verify icon files meet the
+ * size floors that Chrome's imageUtil requires for notifications.create
+ * (Bug A regression class — too-small icon → create rejects).
+ *
+ * @param popup - The extension popup page handle.
+ * @param relativePath - Path under the extension root (e.g. 'icons/icon128.png').
+ * @returns Byte size on success, -1 on fetch failure.
+ */
+export async function getIconSize(
+  popup: Page,
+  relativePath: string,
+): Promise<number> {
+  return await popup.evaluate(async (path: string) => {
+    const url = chrome.runtime.getURL(path);
+    const r = await fetch(url);
+    if (!r.ok) {
+      return -1;
+    }
+    const cl = r.headers.get('content-length');
+    if (cl !== null) {
+      const n = Number(cl);
+      if (Number.isFinite(n) && n > 0) {
+        return n;
+      }
+    }
+    const buf = await r.arrayBuffer();
+    return buf.byteLength;
+  }, relativePath);
+}
+
+/**
+ * Read whether the SW thinks a recording is active. Side-channeled
+ * through the badge text — 'REC' ↔ recording; '' ↔ idle; 'ERR' ↔
+ * error state — to avoid needing a dedicated hook field.
+ *
+ * @param popup - The extension popup page handle.
+ * @returns true when badge === 'REC'.
+ */
+export async function getIsRecording(popup: Page): Promise<boolean> {
+  const badge = await getBadgeText(popup);
+  return badge === 'REC';
+}
+
+/**
+ * Fire the captured chrome.runtime.onStartup handler via the test
+ * hook's chrome.runtime.sendMessage bridge. Used by assertion 8 to
+ * verify the Bug A path (icon-promoted notification fires cleanly).
+ *
+ * Bridge protocol: popup sends `{ type: '__mokoshTestQuery', op: 'fire-on-startup' }`;
+ * SW responds with `{ ok: true }` after invoking the handler, OR
+ * `{ ok: false, error: 'no-handler' }` if the production listener
+ * was never registered (means the SW module init failed — a
+ * different bug class).
+ *
+ * @param popup - The extension popup page handle.
+ * @throws If the bridge response indicates the handler is missing.
+ */
+export async function fireOnStartup(popup: Page): Promise<void> {
+  const response = await popup.evaluate(async () => {
+    const msg = {
+      type: '__mokoshTestQuery',
+      op: 'fire-on-startup',
+    };
+    return new Promise<{ ok: boolean; error?: string }>((resolve) => {
+      chrome.runtime.sendMessage(msg, (r) => {
+        resolve(r as { ok: boolean; error?: string });
+      });
+    });
+  });
+  if (!response.ok) {
+    throw new Error(
+      `fireOnStartup bridge returned ok=false: ${response.error ?? '(no error message)'}`,
+    );
+  }
+}
+
+/**
+ * Inject a synthetic RECORDING_ERROR message into the SW's
+ * chrome.runtime.onMessage handler. Used by assertion 7 to verify
+ * the error path is preserved (badge 'ERR' + recovery notification).
+ * Goes through the popup's chrome.runtime.sendMessage — a real
+ * production code path (sw onMessage handler).
+ *
+ * @param popup - The extension popup page handle.
+ * @param errorCode - The error code to inject (e.g. 'codec-unsupported').
+ */
+export async function sendSyntheticRecordingError(
+  popup: Page,
+  errorCode: string,
+): Promise<void> {
+  await popup.evaluate(async (code: string) => {
+    await chrome.runtime.sendMessage({
+      type: 'RECORDING_ERROR',
+      error: code,
+    });
+  }, errorCode);
+}
+
+/**
+ * Snapshot the current notification observability state from the SW
+ * hook via the bridge.
+ *
+ * @param popup - The extension popup page handle.
+ * @returns Snapshot — count, last options, ids array.
+ */
+export async function getNotificationSnapshot(
+  popup: Page,
+): Promise<NotificationSnapshot> {
+  const response = await popup.evaluate(async () => {
+    const msg = { type: '__mokoshTestQuery', op: 'snapshot' };
+    return new Promise<{
+      count: number;
+      lastOptions: chrome.notifications.NotificationOptions<true> | null;
+      ids: string[];
+    }>((resolve) => {
+      chrome.runtime.sendMessage(msg, (r) => {
+        resolve(r as {
+          count: number;
+          lastOptions: chrome.notifications.NotificationOptions<true> | null;
+          ids: string[];
+        });
+      });
+    });
+  });
+  return {
+    count: response.count,
+    lastOptions: response.lastOptions,
+    ids: response.ids,
+  };
+}
+
+/**
+ * Send a no-op keepalive ping to the SW so Chrome's ~30s idle timer
+ * does not evict the worker during long waits (assertion 11's 35s
+ * recording window). Uses chrome.runtime.sendMessage as the cheapest
+ * wake-up signal; the SW's onMessage handler treats unknown messages
+ * as a warning-log no-op.
+ *
+ * @param popup - The extension popup page handle.
+ */
+export async function keepalivePing(popup: Page): Promise<void> {
+  await popup.evaluate(async () => {
+    await chrome.runtime.sendMessage({ type: '__mokoshKeepalive' });
+  });
+}
+
+// Re-export the BridgeQuery type for sw-hooks.ts side reference
+// (the SW hook implements the message dispatch using the same shape).
+export type { BridgeQuery };
diff --git a/tests/uat/lib/zip.ts b/tests/uat/lib/zip.ts
new file mode 100644
index 0000000..06190f3
--- /dev/null
+++ b/tests/uat/lib/zip.ts
@@ -0,0 +1,121 @@
+// tests/uat/lib/zip.ts — Plan 01-11 harness archive-shape helper.
+//
+// Assertion 13 verifies the session_report_*.zip produced by the SW's
+// saveArchive contains:
+//   - `video/last_30sec.webm` (non-zero size)
+//   - `meta.json` whose parsed JSON has `version === <manifest.version>`
+//
+// References:
+//   - JSZip: https://stuk.github.io/jszip/documentation/api_jszip.html
+//   - Plan 01-07 archive shape (session_report contract):
+//     .planning/phases/01-stabilize-video-pipeline/01-07-PLAN.md
+
+import { readFileSync } from 'node:fs';
+
+import JSZip from 'jszip';
+
+/**
+ * Outcome of an archive shape inspection. `errors` lists every
+ * missing-file / wrong-size / version-mismatch finding.
+ */
+export interface ArchiveShapeResult {
+  readonly hasVideoEntry: boolean;
+  readonly videoSizeBytes: number;
+  readonly hasMetaEntry: boolean;
+  readonly metaJson: { version?: unknown } | null;
+  readonly errors: ReadonlyArray<string>;
+}
+
+/**
+ * Open a downloaded session_report_*.zip and verify its shape.
+ *
+ * @param zipPath - Absolute path to the downloaded .zip file.
+ * @param expectedVersion - The version string from chrome.runtime.getManifest().version.
+ * @returns Structured shape result. `errors` non-empty == assertion failure.
+ */
+export async function assertArchiveShape(
+  zipPath: string,
+  expectedVersion: string,
+): Promise<ArchiveShapeResult> {
+  const zipBuf = readFileSync(zipPath);
+  const zip = await JSZip.loadAsync(zipBuf);
+  const errors: string[] = [];
+
+  // video/last_30sec.webm presence + size
+  const videoEntry = zip.file('video/last_30sec.webm');
+  let hasVideoEntry = false;
+  let videoSizeBytes = 0;
+  if (videoEntry === null) {
+    errors.push('video/last_30sec.webm entry missing from archive');
+  } else {
+    hasVideoEntry = true;
+    const videoBuf = await videoEntry.async('uint8array');
+    videoSizeBytes = videoBuf.byteLength;
+    if (videoSizeBytes === 0) {
+      errors.push('video/last_30sec.webm entry is zero bytes (no captured video)');
+    }
+  }
+
+  // meta.json presence + version match
+  const metaEntry = zip.file('meta.json');
+  let hasMetaEntry = false;
+  let metaJson: { version?: unknown } | null = null;
+  if (metaEntry === null) {
+    errors.push('meta.json entry missing from archive');
+  } else {
+    hasMetaEntry = true;
+    const metaText = await metaEntry.async('string');
+    try {
+      metaJson = JSON.parse(metaText) as { version?: unknown };
+    } catch (parseErr) {
+      const msg = parseErr instanceof Error ? parseErr.message : String(parseErr);
+      errors.push(`meta.json failed to parse as JSON: ${msg}`);
+    }
+    if (metaJson !== null) {
+      if (typeof metaJson.version !== 'string') {
+        errors.push(
+          `meta.json.version expected string, got ${typeof metaJson.version} (${JSON.stringify(metaJson.version)})`,
+        );
+      } else if (metaJson.version !== expectedVersion) {
+        errors.push(
+          `meta.json.version mismatch — expected "${expectedVersion}", got "${metaJson.version}"`,
+        );
+      }
+    }
+  }
+
+  return {
+    hasVideoEntry,
+    videoSizeBytes,
+    hasMetaEntry,
+    metaJson,
+    errors,
+  };
+}
+
+/**
+ * Extract a single named entry from a .zip to an absolute filesystem
+ * path. Used by assertion 12 (ffprobe gate on video/last_30sec.webm).
+ *
+ * @param zipPath - Absolute path to the .zip.
+ * @param entryName - Name of the entry inside the zip (e.g. 'video/last_30sec.webm').
+ * @param outPath - Absolute filesystem path to write the entry to.
+ * @returns The number of bytes written.
+ * @throws If the entry is missing from the zip.
+ */
+export async function extractEntryToFile(
+  zipPath: string,
+  entryName: string,
+  outPath: string,
+): Promise<number> {
+  const { writeFileSync } = await import('node:fs');
+  const zipBuf = readFileSync(zipPath);
+  const zip = await JSZip.loadAsync(zipBuf);
+  const entry = zip.file(entryName);
+  if (entry === null) {
+    throw new Error(`extractEntryToFile: entry '${entryName}' missing in ${zipPath}`);
+  }
+  const buf = await entry.async('nodebuffer');
+  writeFileSync(outPath, buf);
+  return buf.byteLength;
+}
diff --git a/tests/uat/tsconfig.json b/tests/uat/tsconfig.json
new file mode 100644
index 0000000..5f495af
--- /dev/null
+++ b/tests/uat/tsconfig.json
@@ -0,0 +1,20 @@
+{
+  "_comment": "Plan 01-11 — type-check config for the Puppeteer UAT harness. Mirrors the root tsconfig.json's compiler options but `include`s the harness tree explicitly so `npx tsc --noEmit -p tests/uat` validates the harness in isolation. Used by Task 3 to verify the scaffolding type-checks before tsx runs it. The root tsconfig.json's `include: ['src']` does NOT pick up tests/, so this file is necessary for the type-check verification step.",
+  "compilerOptions": {
+    "target": "ES2022",
+    "module": "ESNext",
+    "lib": ["ES2022", "DOM", "DOM.Iterable"],
+    "skipLibCheck": true,
+    "moduleResolution": "bundler",
+    "allowImportingTsExtensions": true,
+    "resolveJsonModule": true,
+    "isolatedModules": true,
+    "noEmit": true,
+    "strict": true,
+    "noUnusedLocals": true,
+    "noUnusedParameters": true,
+    "noFallthroughCasesInSwitch": true,
+    "types": ["chrome", "node"]
+  },
+  "include": ["**/*.ts", "**/*.d.ts"]
+}
diff --git a/vitest.config.ts b/vitest.config.ts
index 38e7872..ad26eae 100644
--- a/vitest.config.ts
+++ b/vitest.config.ts
@@ -17,6 +17,14 @@ export default defineConfig({
   test: {
     environment: 'node',
     include: ['tests/**/*.test.ts'],
+    // Plan 01-11: exclude the Puppeteer harness from vitest's discovery.
+    // tests/uat/harness.test.ts is a tsx-runnable Node script invoked
+    // via `npm run test:uat`; running it under vitest would try to
+    // launch a real Chrome inside the vitest worker (interactive UAT
+    // does not belong in the unit-test pass). The .test.ts suffix is
+    // retained for editor + naming-convention consistency with the
+    // rest of the tests/ tree.
+    exclude: ['node_modules/**', 'tests/uat/**'],
     reporters: 'dot',
     typecheck: {
       enabled: false,