Speed up site testing selection and CI Production UI shards

.github/workflows/ci.yml

@@ -555,31 +555,15 @@ jobs:
       # report as `cancelled` and re-create exactly the ambiguity #095 removed.
       fail-fast: false
       matrix:
-        # THREE, and the count is measured rather than chosen. `fullyParallel: false`
-        # makes a spec file the indivisible unit, so shard sizes are lumpy and more
-        # shards is not monotonically faster. Re-measured 2026-07-30 against this
-        # merged tree — 342 non-quarantine/non-mockup chromium tests
-        # (`--list --shard=i/N`):
-        #   N=3 -> 121/111/110   largest 121 (35%)
-        #   N=4 -> 121/106/98/17 largest 121 (35%)  — same bound, one more runner
-        # Earlier on a 340-test tree, N=5 -> 121/106/0/96/17 and N=8 gave two empty
-        # shards. N=4 buys nothing over N=3 because one 121-test spec group bounds
-        # both, and any N with an empty shard would go RED: `test:e2e:pr`
-        # deliberately omits `--pass-with-no-tests` (only the advisory lane has it).
-        # Re-measure before changing this number, and keep every shard non-empty:
-        #   npm run ensure
-        #   PLAYWRIGHT_BASE_URL=<printed> npx playwright test --project=chromium \
-        #     --grep-invert "@quarantine|@mockup" --shard=i/N --list
-        #
-        # MEASURED on the first real sharded run (CI 30530618838, 2026-07-30):
-        #   shard 1 -> 121 tests, 9m36
-        #   shard 2 -> 111 tests, 6m54
-        #   shard 3 -> 110 tests, 6m20
-        # Per-test cost is NOT uniform — 111 tests took 6m54 while 121 took 9m36 —
-        # so counting tests UNDERSTATES the largest shard. A count-balanced split is
-        # the best `--shard` can do; balancing by duration would mean splitting the
-        # slow spec files themselves. Predicting from test count alone was wrong by
-        # ~40% here, so trust a measured run over the arithmetic.
+        # THREE explicit duration-aware file groups (scripts/playwright-pr-shards.mjs),
+        # not Playwright `--shard=i/N`. Count-balanced `--shard` packed the slow
+        # phone-scroll family into one runner (measured 9m36 vs 6m54/6m20 on
+        # CI 30530618838). Explicit groups mix slow-per-test specs with faster
+        # mega-specs. `tests/playwright-pr-shards.test.ts` fails closed if any
+        # production e2e:pr file is missing, duplicated, or leaves a shard empty.
+        # Re-measure wall time after suite growth before reshuffling membership:
+        #   node scripts/playwright-pr-shards.mjs --list
+        #   node scripts/playwright-pr-shards.mjs --validate
         shard: [1, 2, 3]
     steps:
       - name: Checkout
@@ -607,7 +591,7 @@ jobs:
         env:
           PLAYWRIGHT_BUILD_ROOT_ID: ci-production
           PLAYWRIGHT_KEEP_BUILD_ROOT: "true"
-        run: npm run test:e2e:pr -- --shard=${{ matrix.shard }}/3
+        run: npm run test:e2e:pr:shard -- --shard ${{ matrix.shard }}
 
       - name: Classify exact failed test identities
         if: failure()

AGENTS.md

@@ -1061,7 +1061,7 @@ Durable notes for Cloud Agents. Standard commands live in `README.md` and `packa
 - Live-mode caveat: `RAG_PROVIDER_MODE=auto` attempts OpenAI (fast → strong route); if generation fails the built-in quality gates it silently degrades to a deterministic "Source-only" answer that still cites real documents — this is expected, not a failure. The header sign-in UI exposes magic-link + OAuth only (no password field), but the `/api/answer` + retrieval flow works server-side without a browser session.
 - What still won't run in this VM even with secrets: `npm run worker` also needs the Python OCR stack (`worker/python/requirements.txt`) and heavy parsing deps; Supabase edge functions need Deno v2.x + deployment. `verify:release` additionally runs governance/eval gates. Treat missing-secret failures of `check:supabase-project`/`verify:release` in demo mode as expected, not regressions.
 - Dev server: `npm run dev` selects a stable per-project localhost port (e.g. `4461`), binds `0.0.0.0`, and prints the exact URL. Never assume port 3000/3001/3002. `npm run ensure` starts/verifies it in the background.
-- Verification without secrets: `npm run lint`, `npm run typecheck`, and `npm run test` (vitest) all pass offline. `npm run verify:cheap` also runs runtime, GitHub Actions pin, CI-scope, and sitemap checks. `npm run verify:pr-local` adds format, conditional build/client-bundle scanning, and RAG fixture/manifest validation without repeating unit tests; browser, Docker/Supabase, audit, and provider checks remain separate. See `docs/testing.md` for lock, live-test, Playwright, and flake-ledger rules.
+- Verification without secrets: `npm run lint`, `npm run typecheck`, and `npm run test` (vitest) all pass offline. `npm run verify:cheap` also runs runtime, GitHub Actions pin, CI-scope, and sitemap checks. `npm run verify:pr-local` adds format, conditional build/client-bundle scanning, and RAG fixture/manifest validation without repeating unit tests; browser, Docker/Supabase, audit, and provider checks remain separate. See `docs/testing.md` for lock, live-test, Playwright, and flake-ledger rules. If `check:installed-lock-parity` or `check:playwright-browser-revision` reports Playwright/image drift (`#255`), do not force a mismatched Chromium path — delegate browser proof to CI Production UI (see `docs/testing.md` § Testing speed playbook).
 - For GitHub-related work authorised in this session, prefer the connected GitHub
   connector/MCP tools first for PR, issue, comment, review-thread, and Actions tasks they
   support (including run/job/log/artifact inspection and review-thread replies/resolution).