Dotcom E2E conventions

Issue #9185 is moving dotcom tests toward user scenarios instead of isolated UI smoke tests. New collaborative dotcom coverage should prefer the scenario fixture in fixtures/scenario-test.ts.

Scenario tests

• Name scenario files *.scenario.spec.ts. They run in the chromium-scenarios project with fullyParallel: true.
• Run the ordinary local/CI dotcom suite with yarn e2e-dotcom from the repo root, or yarn workspace dotcom e2e from this workspace. This runs the scenario project only.
• Run only the scenario suite without clearing stored auth with yarn workspace dotcom e2e-scenarios.
• Run the preserved legacy smoke suite with yarn e2e-dotcom-smoke from the repo root, or yarn workspace dotcom e2e-smoke from this workspace. These specs live in tests/smoke, still use the reset-based chromium project, and are intentionally separate from the default runner. They do not run on CI; migrate coverage out of tests/smoke into scenario tests as the follow-ups below are unblocked.
• Run both projects together with yarn e2e-dotcom-all from the repo root, or yarn workspace dotcom e2e-all from this workspace, when comparing old and new coverage.
• Use named actors from the fixture: owner, member, and visitor.
• Use scenario.name('label') for files and workspaces so test data is unique to the run.
• Do not reset the whole database or shared users in scenario tests. Stable Clerk test accounts are reused by worker, while files and workspaces are namespaced by scenario.
• Scenario actors use later Clerk test accounts than the legacy chromium project so explicit all-project runs can share a command while legacy tests still reset their own users.
• Actor contexts and pages are owned by the fixture and closed automatically.
• Use scenario.createPersonalFile, scenario.createSharedFile, scenario.createGuestEditFile, scenario.createGuestViewFile, scenario.createPublishedFile, scenario.importFileFromUrl, scenario.downloadFileFromSidebar, scenario.setSharedLinkType, scenario.createWorkspaceWithMember, and scenario.createWorkspaceWithRemovedMember for common setup.
• Use scenario.createLegacyRouteFixture for legacy /r, /ro, /v, /s, and history route setup. It creates namespaced debug-only worker data instead of relying on shared production-like fixtures.
• Use direct database setup only for preconditions that are expensive or impossible through the UI, such as enabling workspace flags. Keep the behavior under test in UI actions.
• Prefer live assertions across already-open windows. Reload-based checks are still useful for persistence, but they should not be the only proof for collaboration, permission, or membership behavior.
• Wait for readiness through actor helpers and page objects. actor.goto waits for the canvas, auth state, app store hydration when a signed-in app is present, mounted editor, file room connection, visitor access metadata, and pending app mutations. Use the narrower actor.waitForAuthLoaded, actor.waitForAppStoreHydrated, actor.waitForEditorReady, actor.waitForFileRoomConnected, actor.waitForVisitorAccessMetadata, actor.waitForMutationResolution, and actor.waitForSessionClosed helpers when a test needs to prove a specific transition. Avoid new sleeps unless the product intentionally waits.

Auth follow-ups

The remaining auth edge cases are still tracked by issue #9185. Keep them out of the scenario project until they have Clerk-level mocking or stable account state that can run without per-test shared-user reset.

Follow-up migration groups:

• Auth legal acceptance and analytics consent: replace route-level Clerk response patching with a stable Clerk test account state or a mocked Clerk client boundary.
• Auth verification and resend behavior: isolate Clerk network failures, malformed responses, cooldown behavior, and code input behavior without depending on shared signed-in user cleanup.
• OAuth auth flows: add a Clerk/OAuth boundary that can exercise Google sign-in and legal acceptance without external redirects.
• Auth response edge cases: cover completed sessions, inconsistent missing_fields payloads, and missing email-code factor data once Clerk responses can be mocked below the route layer.