test: rewrite e2e to match human demo flow

1. Bench UI landing → 2. Browse scenarios → 3. Trigger certify (API,
   TODO: UI button) → 4. Exam results → 5. Leaderboard → 6. Evidra
   UI evidence + dashboard.

Uses Traefik-routed /lab/ path instead of separate BENCH_UI_URL.

Signed-off-by: Vitas <vitas@users.noreply.github.com>

Author: vitas

tests/e2e/demo-flow.spec.ts

@@ -1,26 +1,29 @@
 /**
- * E2E browser test for the hackathon demo flow.
+ * E2E browser test — simulates a human demo flow through bench-ui and evidra-ui.
  *
  * Prerequisites:
- *   docker compose up -d postgres evidra-api evidra-mcp bench-cli bench-ui
- *   docker compose run --rm kind-bootstrap
+ *   docker compose up -d
  *
  * Run:
  *   cd tests/e2e && npx playwright test demo-flow.spec.ts
  *
- * What the audience sees (and this test verifies):
- *   1. Evidra UI — landing, evidence, scorecard, bench dashboard
- *   2. Trigger scenario via API → bench-cli runs it
- *   3. Evidence chain populated with prescribe/report entries
- *   4. Bench UI — leaderboard, certification results
+ * Flow:
+ *   1. Open bench UI landing page
+ *   2. Browse scenarios
+ *   3. Start certify run (TODO: UI button not yet implemented, uses API)
+ *   4. Check exam results page
+ *   5. Check bench leaderboard
+ *   6. Switch to evidra UI — evidence and dashboard
  */
 
 import { test, expect, type Page } from "@playwright/test";
 
-const EVIDRA_URL = process.env.EVIDRA_URL || "http://localhost:28080";
-const BENCH_UI_URL = process.env.BENCH_UI_URL || "http://localhost:28081";
+const BASE_URL = process.env.EVIDRA_URL || "http://localhost:28080";
 const API_KEY = process.env.EVIDRA_API_KEY || "dev-api-key";
 
+// Bench UI is served under /lab/ via Traefik.
+const LAB_URL = `${BASE_URL}/lab`;
+
 // ── Helpers ──────────────────────────────────────────────────────────
 
 async function setApiKey(page: Page) {
@@ -35,7 +38,7 @@ async function setApiKey(page: Page) {
 }
 
 async function apiRequest(path: string, options?: RequestInit) {
-  return fetch(`${EVIDRA_URL}${path}`, {
+  return fetch(`${BASE_URL}${path}`, {
     headers: {
       Authorization: `Bearer ${API_KEY}`,
       "Content-Type": "application/json",
@@ -47,50 +50,34 @@ async function apiRequest(path: string, options?: RequestInit) {
 
 // ── Tests ────────────────────────────────────────────────────────────
 
-test.describe("Hackathon Demo Flow", () => {
+test.describe("Demo Flow", () => {
   test.setTimeout(180_000);
 
-  test("1. Evidra UI loads with navigation", async ({ page }) => {
-    await page.goto(EVIDRA_URL);
+  // Step 1: Open bench UI landing page
+  test("1. Bench UI landing page loads", async ({ page }) => {
+    await page.goto(LAB_URL);
     await page.waitForLoadState("networkidle");
 
-    // Landing page should have the main headline.
-    await expect(page.locator("text=Know what your agent intended")).toBeVisible(
-      { timeout: 10_000 }
-    );
-  });
-
-  test("2. Evidence page loads", async ({ page }) => {
-    await page.goto(`${EVIDRA_URL}/evidence`);
-    await setApiKey(page);
-    await page.waitForTimeout(2_000);
-
-    // Evidence page should load — table or empty state.
-    const body = await page.textContent("body");
-    expect(body).toBeTruthy();
+    await expect(page.locator("text=before shipping")).toBeVisible({
+      timeout: 10_000,
+    });
+    await page.screenshot({ path: "test-results/01-bench-landing.png" });
   });
 
-  test("3. Bench dashboard loads with trigger button", async ({ page }) => {
-    await page.goto(`${EVIDRA_URL}/bench`);
-    await setApiKey(page);
-    await page.waitForTimeout(3_000);
-
-    // Take screenshot to see what the bench page looks like.
-    await page.screenshot({ path: "test-results/bench-dashboard.png" });
+  // Step 2: Browse and select scenarios
+  test("2. Browse scenarios", async ({ page }) => {
+    await page.goto(`${LAB_URL}/scenarios`);
+    await page.waitForLoadState("networkidle");
 
-    // Bench page should load.
+    // Scenarios page should list available scenarios.
     const body = await page.textContent("body");
     expect(body).toBeTruthy();
+    await page.screenshot({ path: "test-results/02-scenarios.png" });
   });
 
-  test("4. Scenarios API returns 75+ scenarios", async () => {
-    const res = await apiRequest("/v1/bench/scenarios");
-    expect(res.status).toBe(200);
-    const data = await res.json();
-    expect(data.scenarios.length).toBeGreaterThanOrEqual(50);
-  });
-
-  test("5. Trigger scenario and wait for completion", async () => {
+  // Step 3: Start certify run
+  // TODO: Implement "Run" button in bench-ui. For now, trigger via API.
+  test("3. Trigger certify run via API", async () => {
     const triggerRes = await apiRequest("/v1/bench/trigger", {
       method: "POST",
       body: JSON.stringify({
@@ -114,58 +101,67 @@ test.describe("Hackathon Demo Flow", () => {
     }
 
     expect(["completed", "failed"]).toContain(status);
-    // Job should have processed 1 scenario.
     expect(job.total).toBe(1);
     expect(job.completed).toBe(1);
   });
 
-  test("6. Evidence entries exist after scenario run", async () => {
-    const res = await apiRequest("/v1/evidence/entries?limit=10");
-    expect(res.status).toBe(200);
-    const data = await res.json();
-    // bench-cli's MCP subprocess forwards evidence to evidra-api.
-    expect(data.entries).toBeDefined();
-  });
-
-  test("7. Bench runs API has results", async () => {
-    const res = await apiRequest("/v1/bench/runs");
-    expect(res.status).toBe(200);
-    const data = await res.json();
-    expect(data.items).toBeDefined();
-  });
+  // Step 4: Check exam results page
+  test("4. Exam results page shows data", async ({ page }) => {
+    await page.goto(`${LAB_URL}/results`);
+    await page.waitForLoadState("networkidle");
+    await page.waitForTimeout(2_000);
 
-  test("8. Leaderboard API responds (public)", async () => {
-    const res = await fetch(`${EVIDRA_URL}/v1/bench/leaderboard`);
-    expect(res.status).toBe(200);
+    // Results page should not show "Not Configured" error.
+    const body = await page.textContent("body");
+    expect(body).not.toContain("Not Configured");
+    await page.screenshot({ path: "test-results/04-exam-results.png" });
   });
 
-  test("9. Bench UI loads landing and leaderboard", async ({ page }) => {
-    await page.goto(BENCH_UI_URL);
+  // Step 5: Check bench leaderboard
+  test("5. Leaderboard shows model results", async ({ page }) => {
+    await page.goto(`${LAB_URL}/bench`);
     await page.waitForLoadState("networkidle");
 
-    // Landing page should show the main headline.
-    await expect(page.locator("text=before shipping")).toBeVisible({
+    await expect(page.locator("text=Model Leaderboard")).toBeVisible({
       timeout: 10_000,
     });
-    await page.screenshot({ path: "test-results/bench-ui-landing.png" });
+    await page.screenshot({ path: "test-results/05-leaderboard.png" });
+  });
 
-    // Navigate to leaderboard.
-    await page.goto(`${BENCH_UI_URL}/bench`);
+  // Step 6: Switch to evidra UI — evidence and dashboard
+  test("6a. Evidra UI landing page", async ({ page }) => {
+    await page.goto(BASE_URL);
     await page.waitForLoadState("networkidle");
-    await expect(page.locator("text=Model Leaderboard")).toBeVisible({
+
+    await expect(page.locator("text=Know what your agent intended")).toBeVisible({
       timeout: 10_000,
     });
-    await page.screenshot({ path: "test-results/bench-ui-leaderboard.png" });
+    await page.screenshot({ path: "test-results/06a-evidra-landing.png" });
   });
 
-  test("10. Bench UI shows bench page with nav", async ({ page }) => {
-    await page.goto(`${BENCH_UI_URL}/bench`);
-    await page.waitForLoadState("networkidle");
+  test("6b. Evidence page loads", async ({ page }) => {
+    await page.goto(`${BASE_URL}/evidence`);
+    await setApiKey(page);
+    await page.waitForTimeout(2_000);
 
-    // Nav should have: Home, Dashboard, Evidence, Bench, Lab
-    await expect(page.locator("nav >> text=Bench")).toBeVisible();
-    await expect(page.locator("nav >> text=Dashboard")).toBeVisible();
+    const body = await page.textContent("body");
+    expect(body).toBeTruthy();
+    await page.screenshot({ path: "test-results/06b-evidence.png" });
+  });
+
+  test("6c. Dashboard loads", async ({ page }) => {
+    await page.goto(`${BASE_URL}/bench`);
+    await setApiKey(page);
+    await page.waitForTimeout(2_000);
 
-    await page.screenshot({ path: "test-results/bench-ui-bench.png" });
+    const body = await page.textContent("body");
+    expect(body).toBeTruthy();
+    await page.screenshot({ path: "test-results/06c-dashboard.png" });
+  });
+
+  // API verification — leaderboard is public, no auth needed.
+  test("7. Leaderboard API responds (public)", async () => {
+    const res = await fetch(`${BASE_URL}/v1/bench/leaderboard`);
+    expect(res.status).toBe(200);
   });
 });