Skip to content

2026-05-25 — e2e Live Wall preview flag + Today-group timestamp gaps

Motivation

Follow-up to 2026-05-25-quick-deploy-env-leak-and-e2e-mock-fix.md, which explicitly deferred "3 remaining e2e:safe failures" as a separate P2 gap. This change root-fixes those three failures and adds vitest contract guards so the underlying class of bug cannot regress silently.

Symptoms (from the e2e:safe run on e2e:all-safe):

  1. monitor-jobs-events.ui.spec.ts:3:5 — "Live Wall" heading never renders.
  2. monitor-jobs-events.ui.spec.ts:16:5 — "Today" group button never appears under the job timeline.
  3. layout-navigation.ui.spec.ts:12:5 — top-nav Live Wall link not visible.

Both #1 and #3 share root cause: the Playwright fixture in scripts/e2e/fixtures/uiTest.ts enabled the customDb and labTools preview prefs but forgot liveWall. Live Wall is gated by <OptionalFeatureRoute>, so without the pref the route is hidden — exactly the documented behaviour, but the fixture silently asymmetric.

2 is a deterministic-clock side effect: the fixture defines const now = "2026-05-24T10:00:00Z" and assigns it to all mock job timestamps. By the time the test runs that now is hours-to-days in the past, so the seeded completed job lands in the "Yesterday" bucket instead of "Today" and the assertion can never find its group button.

User-facing change

  • The e2e:safe monitor-jobs-events.ui and layout-navigation.ui projects now pass without the Live Wall / Today regressions.
  • No production runtime behaviour changes. Only the test fixture + a usePreferences export visibility change (no SPA behaviour change) + 4 new vitest contract tests.

API / IaC / config diff summary

  • scripts/e2e/fixtures/uiTest.ts — add previewLiveWallEnabled: true, to the elb-prefs seed so all three preview-flag routes are reachable by the mock-mode tests. Source of truth alignment: PREVIEW_PREF_KEYS in web/src/hooks/usePreferences.tsx enumerates exactly customDb / labTools / liveWall.
  • scripts/e2e/fixtures/mockApi.ts — introduce const recentNow = new Date(Date.now() - 5 * 60 * 1000).toISOString(); (rolling timestamp, 5 minutes ago) and use it for the completedJob.created_at and completedJob.updated_at fields only. The other 17 now references remain pinned to the deterministic value so snapshot-sensitive mocks keep their stable expected output. The split lets the "Today / Yesterday" date-bucket UI see a fresh timestamp without breaking any test that depends on the deterministic clock.
  • web/src/hooks/usePreferences.tsx — const PREVIEW_PREF_KEYS → export const PREVIEW_PREF_KEYS. Visibility-only change; no runtime difference.
  • web/src/hooks/usePreferences.fixtureContract.test.ts — new vitest file with 4 tests:
  • Iterate Object.values(PREVIEW_PREF_KEYS) and assert each one appears as <key>: true in scripts/e2e/fixtures/uiTest.ts — guarantees that adding a 4th preview pref to the enum without updating the fixture will fail this guard.
  • Same assertion for previewCustomDbEnabled, previewLabToolsEnabled, previewLiveWallEnabled explicitly (defence in depth).
  • Assert mockApi.ts defines const recentNow = new Date(.
  • Assert the completedJob block in mockApi.ts contains both created_at: recentNow and updated_at: recentNow.
  • No Bicep, no Python, no production SPA component changed. No new runtime dependency.

Validation evidence

  • npm --prefix web run test -- usePreferences.fixtureContract.test.ts:
  • 4 passed (4 ms).
  • npm --prefix web run test (full vitest suite): 7 files / 23 passed (up from 19 before the 4 new tests).
  • scripts/dev/e2e-ui.sh bypass --headless --fullstack -- npm --prefix web run e2e:all-safe:
  • Before fix (baseline from prior session): 6 failed / 9 passed / 1 skipped.
  • After 2026-05-25-quick-deploy-env-leak-and-e2e-mock-fix.md: 3 failed / 12 passed / 1 skipped.
  • After this change: 0 failed / 15 passed / 1 skipped across ui-mock + api-smoke + mutation-mock. The 1 skip is the api-blast live-submit guarded by E2E_ALLOW_BLAST_SUBMIT=1, expected.
  • scripts/dev/e2e-ui.sh bypass --headless --fullstack -- npm --prefix web run e2e:api-blast: 1 passed / 1 skipped (preflight passes; live submit correctly skipped under default scope).
  • uv run pytest -q api/tests: 187 passed in 20.71s (unchanged, sanity).
  • Production posture verified: stelbdashboard01mul5oh5j {public:Disabled, default:Deny, ipRules:0} after the local-debug toggle was closed.

Hardening (regression prevention)

The two bug classes here are symmetry-of-flags ("when I add a preview pref I must remember to flip it on in the fixture") and time-coupling ("deterministic test clock hides date-bucket bugs"). Both are easy to repeat and impossible to spot in code review without a guard. The new usePreferences.fixtureContract.test.ts enforces both invariants from the source-of-truth side, so adding previewFooEnabled to PREVIEW_PREF_KEYS will fail vitest until the fixture catches up, and accidentally pinning the completedJob timestamps back to the deterministic now will likewise fail. Cheap, fast, no infrastructure change.

What is NOT in this change

  • No frontend redeploy. The 0006 revision's leaked VITE_AUTH_DEV_BYPASS=true remains tracked under the prior change note; this fix does not require an image rebuild.
  • No fix for the core_nt BLAST DB being empty in storage — that is a deployment data state, not a code defect (see "Out of scope" below).
  • No App Insights KQL hunt — there are no real BLAST submits in the recent window to investigate.

Out of scope — Azure live-submit readiness gap

For the record: while assessing full-azure scope per the blast-execution-validation skill, the workload Storage account was briefly opened with the documented local-debug toggle (scripts/dev/storage-public-access.sh on/off, then revoked the temporary Storage Blob Data Reader role) and blast-db/core_nt/ was confirmed empty (0 blobs). The full-azure scenario cannot run without a pre-prepared DB, and DB preparation is ~24h which is outside the skill's default max-hours=4 budget. Reported as blocked_by_budget in this session's BLAST validation summary — not a code change. Storage and RBAC were both reverted to the production posture before this note was written.

Risk

Low. Test-only / hook-visibility change. The new tests assert against actual file contents using node:fs.readFileSync against repo-relative paths and so are self-pinning to the source of truth.