HomenShum · HomenShum · Jul 17, 2026 · Jul 17, 2026
@@ -0,0 +1,65 @@
+# Nightly design loop — the hand-run capture → contract → review loop, on a
+# cadence. Builds main, serves the production bundle, executes every surface
+# contract (fails the run on drift), captures the 4-variant screenshot set,
+# and uploads both as artifacts so any agent or human can review the visual
+# state of the app each morning without rebuilding anything.
+name: Nightly design loop
+
+on:
+  schedule:
+    - cron: "17 9 * * *" # 09:17 UTC ≈ 2:17am PT, after the day's merges settle
+  workflow_dispatch: {}
+
+env:
+  VITE_CONVEX_URL: ${{ secrets.CONVEX_URL || vars.CONVEX_URL || 'https://agile-caribou-964.convex.cloud' }}
+
+jobs:
+  design-loop:
+    name: Contract sweep + screenshot capture
+    runs-on: ubuntu-latest
+    timeout-minutes: 30
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: actions/setup-node@v6
+        with:
+          package-manager-cache: false
+          node-version: 22
+
+      - name: Install dependencies
+        run: npm install --no-audit --no-fund
+
+      - name: Install Chromium
+        run: npx playwright install --with-deps chromium
+
+      - name: Build production bundle
+        run: npm run build
+
+      - name: Serve bundle
+        run: |
+          npx vite preview --host 127.0.0.1 --port 4173 &
+          for i in $(seq 1 30); do
+            curl -sf http://127.0.0.1:4173/ > /dev/null && break
+            sleep 1
+          done
+
+      - name: Execute surface contracts (fails on drift)
+        env:
+          BASE_URL: http://127.0.0.1:4173
+        run: npx playwright test tests/e2e/ui-contract-runner.spec.ts --project=chromium --workers=1 --reporter=line
+
+      - name: Capture 4-variant screenshot set
+        env:
+          BASE_URL: http://127.0.0.1:4173
+          DOGFOOD_SCREENSHOT_DIR: nightly-captures
+        run: npx playwright test tests/e2e/full-ui-dogfood.spec.ts --project=chromium --workers=1 --reporter=line
+
+      - name: Upload screenshots
+        if: always()
+        uses: actions/upload-artifact@v7
+        with:
+          name: nightly-design-loop-${{ github.run_id }}
+          path: |
+            nightly-captures/
+            playwright-report/
+          retention-days: 14
@@ -41,6 +41,22 @@
       "url": "/redesign/chat?report=__contract_missing__&artifact=notebook",
       "expectText": "Notebook context unavailable",
       "forbidText": "Notebook context selected"
+    },
+    {
+      "name": "replay-miss",
+      "why": "an unknown /r/{hash} share link must degrade into the live workspace with an honest banner, never a blank page or a fabricated receipt",
+      "url": "/redesign/chat/r/__contract_missing__",
+      "expectText": "That reproducible answer is unavailable",
+      "expectTestId": "continued-chat-context"
+    },
+    {
+      "name": "mobile-entry-redirect",
+      "why": "since the one-surface contraction (#561) every entry point resolves to the workspace; the legacy mobile shell must not resurface",
+      "url": "/",
+      "viewport": "mobile",
+      "expectUrlPattern": "/redesign/chat",
+      "expectTestId": "one-surface-workspace",
+      "forbidTestId": "mobile-home-surface"
     }
   ],
   "answerPacket": {

@@ -1,4 +1,33 @@
 import path from "node:path";
+import { readdirSync } from "node:fs";
+import { fileURLToPath } from "node:url";
+
+/**
+ * Ground truth for the vision judge: every clause in the executable surface
+ * contracts (docs/design/ui-contract/surfaces/*.contract.json) is verified
+ * deterministically in CI by ui-contract-runner.spec.ts. A vision finding that
+ * contradicts a verified clause is a misread, not a bug — feeding the clause
+ * list to the judge stops paid vision passes from re-litigating what CI
+ * already proved (and cuts the score variance the Gemini loop is known for).
+ */
+function contractVerifiedInvariants() {
+  try {
+    const dir = path.join(
+      path.dirname(fileURLToPath(import.meta.url)),
+      "..", "..", "docs", "design", "ui-contract", "surfaces",
+    );
+    const lines = [];
+    for (const file of readdirSync(dir).filter((f) => f.endsWith(".contract.json"))) {
+      const contract = JSON.parse(readFileSync(path.join(dir, file), "utf8"));
+      for (const anchor of contract.anchors ?? []) lines.push(`- [${contract.surface}] ${anchor.why}`);
+      for (const rule of contract.geometry ?? []) lines.push(`- [${contract.surface}] ${rule.why}`);
+      for (const state of contract.states ?? []) lines.push(`- [${contract.surface}] state "${state.name}": ${state.why}`);
+    }
+    return lines.join("\n");
+  } catch {
+    return "";
+  }
+}
 import fs from "node:fs/promises";
 import { existsSync, readFileSync } from "node:fs";
 import net from "node:net";
@@ -2138,11 +2167,15 @@ async function judgeIssuesWithLLM(issues) {
     return `ISSUE #${i + 1} [${route}]: ${header}\n${details}`;
   }).join("\n\n---\n\n");
 
+  const verifiedInvariants = contractVerifiedInvariants();
   const prompt = `You are a senior UI/UX quality judge reviewing automated QA findings for a web application.
 
 DESIGN CONTEXT:
 ${DESIGN_CONTEXT}
-
+${verifiedInvariants ? `
+CONTRACT-VERIFIED INVARIANTS (machine-checked in CI on every PR — treat as ground truth; a finding that contradicts one of these is a screenshot misread, not a bug):
+${verifiedInvariants}
+` : ""}
 TASK: For each issue below, classify it into exactly ONE category:
 - "genuine_bug": A real, actionable defect that would affect users (broken layout, actual contrast failure, real typo, data corruption visible to end users)
 - "design_opinion": A subjective preference about intentional design choices (density, spacing, color hierarchy, information architecture, terminology for target audience)

@@ -1,3 +1,18 @@
+/**
+ * live-smoke-mobile — Tier B / post-deploy mobile verification.
+ *
+ * Rewritten 2026-07-17: the previous version asserted the standalone mobile
+ * shell surfaces (mobile-home-surface, mobile-reports-surface, ...), which
+ * stopped mounting when #561 contracted NodeBench to the single decision
+ * workspace — every entry route now resolves to /redesign/chat at every
+ * viewport. The old assertions were red against current production while
+ * living outside CI: exactly the "test the pipeline cannot see" failure mode
+ * issue #567 documented. This spec now asserts the one-surface reality;
+ * richer per-clause coverage lives in the surface contract
+ * (docs/design/ui-contract/surfaces/decision-workspace.contract.json) executed
+ * by ui-contract-runner.spec.ts.
+ */
+
 import { expect, test } from "@playwright/test";
 
 const BASE_URL =
@@ -19,68 +34,27 @@ async function expectNoHorizontalOverflow(page: import("@playwright/test").Page)
   expect(fits, "mobile runtime surface should not overflow horizontally").toBe(true);
 }
 
-test.describe("live-smoke-mobile — canonical runtime surfaces", () => {
+test.describe("live-smoke-mobile — one workspace at 390px", () => {
   test.setTimeout(45_000);
 
-  test("Home keeps the mobile wrapper around the runtime-backed Home surface", async ({ page }) => {
-    await page.goto(`${BASE_URL}/`);
-    const surface = page.locator('[data-testid="mobile-home-surface"]');
-    await expect(surface).toBeVisible({ timeout: 20_000 });
-    await expect(surface.locator('[data-testid="exact-web-home-surface"]')).toBeVisible();
-    await expect(surface.locator('[data-testid="exact-web-home-composer"]')).toBeVisible();
-    await expect(surface.locator('[data-testid="mobile-home-async-run"]')).toHaveCount(0);
-    await expectNoHorizontalOverflow(page);
-  });
-
-  test("Reports exposes real pipeline controls and honest runtime report state", async ({ page }) => {
-    await page.goto(`${BASE_URL}/?surface=reports`);
-    const surface = page.locator('[data-testid="mobile-reports-surface"]');
-    await expect(surface).toBeVisible({ timeout: 20_000 });
-    await expect(surface.locator('[data-testid="reports-performance-root"]')).toBeVisible();
-    await expect(surface.locator('[data-testid="pipeline-launcher"]')).toBeVisible();
-    await expect(surface.locator('[data-testid="pipeline-launcher-sign-in"]')).toBeVisible();
-    await expect(
-      surface.locator('[data-testid="reports-loading"], [data-testid="reports-empty"], [data-testid="report-card"]').first(),
-    ).toBeVisible();
-    await expect(surface.locator('[data-testid="mobile-report-card"]')).toHaveCount(0);
-    await expectNoHorizontalOverflow(page);
-  });
-
-  test("Chat mounts the live runtime and a working composer, not a verified fixture", async ({ page }) => {
-    await page.goto(`${BASE_URL}/?surface=chat`);
-    const surface = page.locator('[data-testid="mobile-chat-surface"]');
-    await expect(surface).toBeVisible({ timeout: 20_000 });
-    await expect(surface.locator('[data-testid="exact-web-chat-stream"]')).toBeVisible();
-    await expect(surface.getByLabel("Chat composer")).toBeVisible();
-    await expect(surface.getByLabel("Chat composer")).toHaveAttribute(
-      "placeholder",
-      "Ask or paste text... (@ to mention an entity)",
-    );
-    await expect(surface.getByText("Live Context Runtime")).toBeVisible();
-    await expect(surface.getByText(/3 entities - 24 sources/i)).toHaveCount(0);
-    await expectNoHorizontalOverflow(page);
-  });
-
-  test("Inbox shows only the runtime snapshot or its honest empty/loading state", async ({ page }) => {
-    await page.goto(`${BASE_URL}/?surface=inbox`);
-    const surface = page.locator('[data-testid="mobile-inbox-surface"]');
-    await expect(surface).toBeVisible({ timeout: 20_000 });
-    await expect(surface.locator('[data-testid="exact-web-inbox-surface"]')).toBeVisible();
-    await expect(surface.getByRole("button", { name: /^All / })).toBeVisible();
-    await expect(surface.getByRole("button", { name: /^Mentions / })).toHaveCount(0);
-    await expectNoHorizontalOverflow(page);
-  });
-
-  test("Me uses runtime identity and memory rather than canned workspace controls", async ({ page }) => {
-    await page.goto(`${BASE_URL}/?surface=me`);
-    const surface = page.locator('[data-testid="mobile-me-surface"]');
-    await expect(surface).toBeVisible({ timeout: 20_000 });
-    await expect(surface.locator('[data-testid="exact-web-me-surface"]')).toBeVisible();
-    await expect(surface.getByRole("button", { name: "Settings" })).toBeVisible();
-    await expect(
-      surface.locator('[data-testid="me-memory-loading"], [data-testid="me-memory-empty"], .nb-settings-section').first(),
-    ).toBeVisible();
-    await expect(surface.getByText(/^Workspaces$/)).toHaveCount(0);
+  for (const path of ["/", "/?surface=reports", "/?surface=chat", "/?surface=inbox", "/?surface=me"]) {
+    test(`${path} resolves to the single decision workspace`, async ({ page }) => {
+      await page.goto(`${BASE_URL}${path}`);
+      await expect(page).toHaveURL(/\/redesign\/chat(?:[/?#]|$)/, { timeout: 20_000 });
+      await expect(page.getByTestId("one-surface-workspace")).toBeVisible({ timeout: 20_000 });
+      // Exactly one composer; the legacy mobile shell must not resurface.
+      await expect(page.locator("textarea:visible")).toHaveCount(1);
+      await expect(page.getByTestId("mobile-home-surface")).toHaveCount(0);
+      await expectNoHorizontalOverflow(page);
+    });
+  }
+
+  test("unknown share link degrades honestly on mobile", async ({ page }) => {
+    await page.goto(`${BASE_URL}/redesign/chat/r/__smoke_missing__`);
+    await expect(page.getByText("That reproducible answer is unavailable").first()).toBeVisible({
+      timeout: 20_000,
+    });
+    await expect(page.getByTestId("one-surface-workspace")).toBeVisible({ timeout: 20_000 });
     await expectNoHorizontalOverflow(page);
   });
 });
@@ -52,7 +52,17 @@ interface SurfaceContract {
     gridTracks?: number;
     minWidthPx?: number;
   }>;
-  states?: Array<{ name: string; why: string; url: string; expectText: string; forbidText?: string }>;
+  states?: Array<{
+    name: string;
+    why: string;
+    url: string;
+    viewport?: string;
+    expectText?: string;
+    forbidText?: string;
+    expectTestId?: string;
+    forbidTestId?: string;
+    expectUrlPattern?: string;
+  }>;
 }
 
 function loadContracts(): SurfaceContract[] {
@@ -130,14 +140,35 @@ for (const contract of contracts) {
 
     for (const state of contract.states ?? []) {
       test(`state › ${state.name}: copy agrees with reality`, async ({ page }) => {
+        const viewport = state.viewport
+          ? contract.viewports.find((v) => v.name === state.viewport)
+          : undefined;
+        if (viewport) {
+          await page.setViewportSize({ width: viewport.width, height: viewport.height });
+        }
         await installVercelPreviewBypass(page, BASE_URL);
         await page.goto(`${BASE_URL}${state.url}`, { waitUntil: "domcontentloaded" });
-        await expect(page.getByText(state.expectText).first(), state.why).toBeVisible({
-          timeout: 30_000,
-        });
+        if (state.expectUrlPattern) {
+          await expect(page, state.why).toHaveURL(new RegExp(state.expectUrlPattern), {
+            timeout: 30_000,
+          });
+        }
+        if (state.expectText) {
+          await expect(page.getByText(state.expectText).first(), state.why).toBeVisible({
+            timeout: 30_000,
+          });
+        }
+        if (state.expectTestId) {
+          await expect(page.getByTestId(state.expectTestId).first(), state.why).toBeVisible({
+            timeout: 30_000,
+          });
+        }
         if (state.forbidText) {
           await expect(page.getByText(state.forbidText), state.why).toHaveCount(0);
         }
+        if (state.forbidTestId) {
+          await expect(page.getByTestId(state.forbidTestId), state.why).toHaveCount(0);
+        }
       });
     }
   });