diff --git a/e2e/semantic-search.spec.ts b/e2e/semantic-search.spec.ts
index 99298648..fa219550 100644
--- a/e2e/semantic-search.spec.ts
+++ b/e2e/semantic-search.spec.ts
@@ -1,23 +1,34 @@
 import { test, expect } from '@playwright/test';
 import { navClick } from './helpers/navigation';
 
+const SEMANTIC_FALLBACK_ASSERTION_TIMEOUT_MS = 10_000;
+const SEMANTIC_FALLBACK_TEST_TIMEOUT_MS = 15_000;
+
 /**
  * Semantic search (N1, Issue #751) — UI-level coverage only.
  *
- * No model download happens here: the spec aborts requests to the Hugging
- * Face Hub and the transformers.js WASM CDN, which forces the graceful
+ * No model download happens here: the spec returns a deterministic 404 for
+ * Hugging Face Hub and transformers.js WASM CDN requests to force the graceful
  * lexical fallback path with its surfaced hint. Ranking behavior itself is
  * covered by the mocked unit tests (embeddings/vector-store/worker).
  */
 test.describe('Semantic search toggle', () => {
+  // Prevent the app service worker from forwarding model requests outside the
+  // Playwright route handler.
+  test.use({ serviceWorkers: 'block' });
+
   test.beforeEach(async ({ page }) => {
-    // Block the embedding model and WASM binaries so the embedder cannot
-    // load — semantic queries then degrade to the lexical fallback hint.
-    // Subdomains matter: the model files live on cdn-lfs.huggingface.co and
-    // the tokenizer/assets may come from jsdelivr subdomains.
+    // Use an immediate missing-asset response instead of relying on browser
+    // network abort/retry scheduling. Subdomains matter: model files live on
+    // cdn-lfs.huggingface.co and tokenizer/assets may come from jsdelivr.
     await page.route(
-      /^https:\/\/(?:[a-z0-9-]+\.)*(?:huggingface\.co|cdn\.jsdelivr\.net)\//,
-      (route) => route.abort(),
+      /^https:\/\/(?:[a-z0-9-]+\.)*(?:huggingface\.co|cdn\.hf\.co|cdn\.jsdelivr\.net)\//,
+      (route) =>
+        route.fulfill({
+          status: 404,
+          contentType: 'text/plain',
+          body: 'Model unavailable in E2E',
+        }),
     );
     await page.goto('/');
     await navClick(page, /library/i);
@@ -43,19 +54,19 @@ test.describe('Semantic search toggle', () => {
   test('surfaces a lexical fallback hint when the semantic model is unavailable', async ({
     page,
   }) => {
+    test.setTimeout(SEMANTIC_FALLBACK_TEST_TIMEOUT_MS);
+
     await page.getByRole('switch', { name: /semantic search/i }).click();
-    await page.getByRole('searchbox', { name: /search library/i }).fill('triz');
+    const searchInput = page.getByRole('searchbox', { name: /search library/i });
+    await searchInput.fill('triz');
 
     const status = page.getByRole('status').filter({ hasText: /keyword results/i });
-    // transformers.js spends several seconds failing its CDN fetches (model
-    // + WASM) before surfacing the embedder error that drives the fallback, and a
-    // full parallel sweep stretches that well past 20s — it failed on both the
-    // attempt and the retry of one four-project run while passing 9/9 in
-    // isolation (plans/149). The budget is deliberately generous because this
-    // assertion waits on a failing network stack, not on app logic.
-    await expect(status).toBeVisible({ timeout: 60_000 });
-
-    // The search box keeps working — results still render (graceful fallback).
-    await expect(page.getByRole('searchbox', { name: /search library/i })).toHaveValue('triz');
+    await expect(status).toBeVisible({
+      timeout: SEMANTIC_FALLBACK_ASSERTION_TIMEOUT_MS,
+    });
+    await expect(searchInput).toHaveValue('triz');
+    await expect(
+      page.getByRole('heading', { name: 'TRIZ Contradiction Matrix' }),
+    ).toBeVisible();
   });
 });
\ No newline at end of file
diff --git a/plans/149-nightly-full-viewport-e2e-2026-09-24.md b/plans/149-nightly-full-viewport-e2e-2026-09-24.md
index 353cafd1..8b4efeb2 100644
--- a/plans/149-nightly-full-viewport-e2e-2026-09-24.md
+++ b/plans/149-nightly-full-viewport-e2e-2026-09-24.md
@@ -182,16 +182,19 @@ have used.
 
 ## 5. Follow-ups
 
-1. **The next real nightly should be confirmed.** The dispatch run proves the
-   mechanism; the 03:00 UTC schedule run is the last piece. If it reports
-   `E2E Tests: skipped` again, the cause is a dependency this plan did not see.
-   → **Explained and hardened** (plans/151 §2). The `2026-09-24T07:59Z` scheduled
-   run skipped every job because its head (`064702a`) predates this plan's fix;
-   the schedule runs the workflow as it exists on the default branch at trigger
-   time. The same investigation found the nightly's scope depended on the path
-   filter's incidental fallback, which plans/151 replaced with an explicit
-   `schedule`/`workflow_dispatch` guard. The next nightly is the first scheduled
-   execution of the fixed workflow.
+1. **Nightly execution — confirmed 2026-09-25.** Scheduled run
+   [`36112610311`](https://github.com/d-oit/do-knowledge-studio/actions/runs/36112610311)
+   on `main` (`094b7e0`) ran both `Unit Tests` and `E2E Tests`. E2E completed
+   `604` tests (`600 passed`, `4 skipped`) in `13.6m` across chromium, mobile,
+   tablet, and desktop-xl. This confirms the scheduled job runs; the run used
+   `main` before this local semantic-search test change and does not validate
+   that edit.
+   → **Explained and hardened** (plans/151 §2). The `2026-09-24T07:59Z`
+   scheduled run skipped every job because its head (`064702a`) predates this
+   plan's fix; the schedule runs the workflow as it exists on the default
+   branch at trigger time. The same investigation found the nightly's scope
+   depended on the path filter's incidental fallback, which plans/151 replaced
+   with an explicit `schedule`/`workflow_dispatch` guard.
 2. **Pre-hydration interaction audit — closed at the helpers (2026-09-24).** The
    exposure was measured across all 24 specs rather than patched per spec:
    - `openNavIfHidden` (and therefore `navClick`) now waits for `data-app-ready`
diff --git a/plans/151-quality-gate-lint-cache-and-nightly-guard-2026-09-24.md b/plans/151-quality-gate-lint-cache-and-nightly-guard-2026-09-24.md
index 49463322..39c91929 100644
--- a/plans/151-quality-gate-lint-cache-and-nightly-guard-2026-09-24.md
+++ b/plans/151-quality-gate-lint-cache-and-nightly-guard-2026-09-24.md
@@ -116,18 +116,28 @@ output and deleting the `frontend=true` emission each fail the test.
    (`ci-and-labels.yml`, "Generate coverage badge"). Pre-existing and style-level
    (`actionlint` runs with `fail_level: error`), so it does not fail CI; it is the
    last finding in that file and unrelated to this change.
-2. **Confirm the next nightly** — the first scheduled execution of the guarded
-   workflow. `gh api "repos/d-oit/do-knowledge-studio/actions/runs?event=schedule"`
-   and check `E2E Tests` is not `skipped`.
-   → **Mechanism proven on `main`** (dispatch run
-   [`36048842590`](https://github.com/d-oit/do-knowledge-studio/actions/runs/36048842590),
-   head `55033bf`): `Treat every path as changed on scheduled and manual runs`
-   ran, **Unit Tests ran** instead of being skipped, and `E2E Tests` swept
-   `604 tests` with `600 passed` in 9.9 min across all four projects. The guarded
-   path and the scheduled path differ only in `github.event_name`, which the
-   contract test pins — the remaining confirmation is the next 03:00 UTC
-   (observed ~08:00 UTC) scheduled run.
-3. **`semantic-search.spec.ts` load sensitivity** (plans/148 §6.1) — unchanged.
+2. **Confirm the next nightly** — closed 2026-09-25. Scheduled run
+   [`36112610311`](https://github.com/d-oit/do-knowledge-studio/actions/runs/36112610311)
+   on `main` (`094b7e0`) ran `Unit Tests` and `E2E Tests` successfully; E2E
+   completed `604` tests (`600 passed`, `4 skipped`) in `13.6m` across all four
+   projects. `Quality Gate`, `Build`, `Coverage Report`, and `Dependency Verify`
+   were skipped as configured; the workflow limits scheduled runs to unit and
+   E2E jobs; this confirmation came from the scheduled event, not a manual dispatch.
+3. **`semantic-search.spec.ts` load sensitivity** (plans/148 §6.1) — resolved
+   in test code only. The spec blocks service workers (the app SW bypassed
+   Playwright routes) and returns deterministic `404`s for Hugging Face
+   Hub/CDN and jsDelivr model requests. The fallback uses a `10s` assertion
+   timeout inside a `15s` test timeout. Chromium repeat-each=3 passed all nine
+   test instances in `29.8s`; the latest-main four-project, zero-retry sweep
+   passed `600/604` (`4` skipped) in `8.8m`.
+   The nightly at `094b7e0` confirms the scheduled job runs but predates this
+   local edit; it does not validate the updated semantic-search spec. No
+   production semantic-search code changed.
 4. **ESLint 10 workaround** (plans/140 §2) — still blocked upstream.
 5. **Graph density** (plans/148 §6.3) — the remaining item from the same follow-up
    sweep; tracked in plans/152.
+6. **HomeView hydration mismatch (E2E warning)** — React logs mismatched motion
+   styles during the local and scheduled sweeps (`opacity: "0"` vs `1`; progress
+   widths `0px` vs computed percentages). The same warning appears in the
+   pre-change manual run `36048842590`; track it separately rather than
+   suppressing it in the semantic-search test.