test(demo): add bisect seed history

demo-ecommerce/abtests.config.ts

@@ -63,6 +63,7 @@ export default defineConfig({
       browser: 'chromium',
       args: ['--no-sandbox'],
     },
+    viewports: ['phone'],
     maxNumDiffPixels: 50,
     defaultMisMatchThreshold: 0.1,
   },
@@ -76,15 +77,29 @@ export default defineConfig({
     // `screenEmulation` are NOT set here — the viewport referenced from
     // shared.viewports owns them; the runner lowers them via
     // lhConfigForViewport.
+    viewports: ['phone'],
     lighthouseConfig: LIGHTHOUSE_CONFIG,
-    numberOfMeasurements: 10,
+    numberOfMeasurements: 8,
     pValueThreshold: 0.01,
   },
 
   audit: {
     lighthouseConfig: LIGHTHOUSE_CONFIG,
   },
 
+  bisect: {
+    rebuildCommands: [
+      {
+        description: 'Install JavaScript dependencies',
+        command: 'yarn install --immutable',
+      },
+      {
+        description: 'Precompile application assets',
+        command: 'rm -rf public/packs tmp/cache && SECRET_KEY_BASE_DUMMY=1 ./bin/rails assets:precompile',
+      },
+    ],
+  },
+
   twinServers: {
     experimentDir: process.cwd(),
     controlDir: process.env.CONTROL_REPO_DIR || '../../shaka-perf-control/demo-ecommerce',

demo-ecommerce/app/javascript/components/pages/HomePage.tsx

@@ -17,17 +17,42 @@ import LoadingSpinner from '../shared/LoadingSpinner';
 import LazySection from '../shared/LazySection';
 import ExperimentA11yRegressions from '../shared/ExperimentA11yRegressions';
 
+const BisectAccessibilityProbe: React.FC = () => (
+  <button
+    type="button"
+    data-cy="bisect-a11y-probe"
+    style={{ position: 'absolute', left: -10000, top: 0, width: 1, height: 1 }}
+  />
+);
+
+const runMerchandisingWarmup = () => {
+  const deadline = performance.now() + 450;
+  let checksum = 0;
+
+  while (performance.now() < deadline) {
+    checksum += Math.sqrt(checksum + 1);
+  }
+
+  if (checksum === Number.POSITIVE_INFINITY) {
+    console.info(checksum);
+  }
+};
+
 const HomePage: React.FC = () => {
   const { products, loading, error } = useProducts();
   const featuredProducts = products.filter((p) => p.featured).slice(0, 4);
 
+  React.useEffect(() => {
+    runMerchandisingWarmup();
+  }, []);
+
   return (
     <Box sx={{ bgcolor: '#f5f5f5', minHeight: '100vh' }}>
       {/* Hero Section */}
       <Box
         data-cy="hero-section"
         sx={{
-          background: 'linear-gradient(135deg, #4f46b5 0%, #764ba2 100%)',
+          background: 'linear-gradient(135deg, #b45309 0%, #92400e 100%)',
           color: 'white',
           py: { xs: 6, md: 10 },
           mb: 6,
@@ -67,6 +92,7 @@ const HomePage: React.FC = () => {
       </Box>
 
       <Container maxWidth="lg">
+        <BisectAccessibilityProbe />
         <ExperimentA11yRegressions />
 
         {/* Features Section */}

demo-ecommerce/app/javascript/components/pages/ProductDetailPage.tsx

@@ -24,11 +24,28 @@ import { useCart } from '../../hooks/useCart';
 import LoadingSpinner from '../shared/LoadingSpinner';
 import ProductFeatures from '../shared/ProductFeatures';
 
+const runRecommendationScoring = () => {
+  const deadline = performance.now() + 350;
+  let checksum = 0;
+
+  while (performance.now() < deadline) {
+    checksum += Math.sqrt(checksum + 1);
+  }
+
+  if (checksum === Number.POSITIVE_INFINITY) {
+    console.info(checksum);
+  }
+};
+
 const ProductDetailPage: React.FC = () => {
   const { id } = useParams<{ id: string }>();
   const { product, loading, error } = useProduct(Number(id));
   const { addToCart } = useCart();
 
+  React.useEffect(() => {
+    runRecommendationScoring();
+  }, []);
+
   if (loading) {
     return <LoadingSpinner />;
   }
@@ -78,6 +95,7 @@ const ProductDetailPage: React.FC = () => {
                 maxHeight: 400,
                 objectFit: 'cover',
                 borderRadius: 2,
+                boxShadow: '0 0 0 8px rgba(180, 83, 9, 0.35)',
               }}
             />
           </Box>

demo-ecommerce/docs/git-bisect-seed-history.md

@@ -0,0 +1,90 @@
+# Git Bisect Seed History
+
+This branch is intentionally shaped for future shaka-perf git bisect testing.
+Commits alternate between harmless demo maintenance and deterministic regression
+fixtures so category-specific bisect flows have clear first-bad targets.
+
+The regression fixtures are artificial by design, but each one is scoped to the
+demo app and kept easy to inspect.
+
+Clean commits in this branch should stay runtime-neutral. They can update this
+note or other demo-only documentation without changing measured pages.
+
+Runtime fixtures should prefer existing measured routes so future bisect tests
+do not need a special-case demo configuration.
+
+The homepage fixtures are intended for the existing homepage and shop-now
+coverage. The product detail fixture is intended for the existing product detail
+coverage.
+
+Performance fixtures should use deterministic client-side work instead of
+network delays so results remain easy to reproduce in local containers.
+
+Accessibility fixtures should avoid layout-visible changes when they are meant
+to be accessibility-only.
+
+Clean commits should be safe to classify as good for every category that has
+not yet received its first dedicated regression fixture.
+
+Combined regression commits are useful for checking that category-specific
+bisect still reports the earliest bad commit for each individual category.
+
+Product detail fixtures should avoid changing product data or cart behavior, so
+they remain independent of Rails seeds and API responses.
+
+Fixture commits should stay small enough that `git show` explains the category
+signal without requiring a full app run.
+
+Seed base: `38dae68` (`main` when this branch was created).
+
+Category map:
+
+- `623a1ae` - no regression, seed docs only.
+- `fe8900e` - no regression, docs only.
+- `58cc828` - first `visreg` bad commit, homepage hero color change.
+- `a55e7f4` - no regression, docs only.
+- `754fcd9` - no regression, docs only.
+- `9c7cfff` - first `perf` bad commit, homepage CPU warmup.
+- `744fe90` - no regression, docs only.
+- `ac38e53` - no regression, docs only.
+- `463c429` - no regression, docs only.
+- `fcb0e2b` - first `accessibility` bad commit, offscreen unnamed button.
+- `c1e2a62` - no regression, docs only.
+- `ce1f601` - no regression, docs only.
+- `3846371` - no regression, docs only.
+- `5345dff` - visual plus performance regression on product detail.
+- `088afb9` - no regression, docs only.
+- `4406a78` - no regression, fixture-impact docs only.
+
+Expected affected AB tests and metrics:
+
+- `Homepage` / `visreg` - screenshot mismatch and diff pixels for
+  `[data-cy="hero-section"]` and `document` after the hero gradient changes.
+- `Homepage` / `perf` - primarily worse `TBT` and lower `LH Score` from the
+  450ms homepage CPU warmup. `speed-index`, `FCP`, and `LCP` can also move
+  depending on timing.
+- `Homepage` / `accessibility` - a new axe finding, expected to be
+  `button-name`, from `button[data-cy="bisect-a11y-probe"]`.
+- `Click Shop Now on the homepage` / `perf` - primarily worse `TBT` and lower
+  `LH Score` in the phone viewport because the test starts on `/` and runs the
+  homepage CPU warmup.
+- `Product Detail` / `visreg` - screenshot mismatch and diff pixels from the
+  product image `boxShadow`.
+- `Product Detail` / `perf` - primarily worse `TBT` and lower `LH Score` from
+  the 350ms product-detail CPU warmup.
+- `Product Detail - Show Product Journey Toggle` / `visreg` - likely full-page
+  screenshot mismatch from the product image `boxShadow` in the phone viewport.
+
+Expected unaffected AB tests:
+
+- `Product Detail - Desktop Actions` captures only
+  `[data-cy="product-actions-desktop"]`, so the product image `boxShadow` should
+  not appear in its screenshot.
+- `Click Reviews on Product Detail` and
+  `Product Details => Click on Reviews => Click on Deals` end on reviews or
+  deals pages and run as visual-only tests, so the product-detail image change
+  should not be captured.
+- `Products List`, `Cart`, `Carousel`, and `Admin` do not visit touched routes
+  or changed fixtures.
+- Product-detail accessibility should stay unchanged because no semantic
+  accessibility regression was added there.