The integration E2E suite intermittently failed under `make test-all` (yet passed in isolation) whenever a test subscribed to the cursorless Jetstream firehose AFTER issuing its PDS write. A cursorless subscription only streams commits emitted after the socket is dialed, so under full-suite CPU load the commit was relayed before the subscriber connected and was lost, hanging the test until its deadline. This is a test-only race — the production AppView consumer is a long-lived connection that is always listening first. Two fix patterns, applied per case: - Cursor replay (helpers.go: jetstreamCursorNow + withJetstreamCursor): capture a time_us cursor immediately before the write, then subscribe with it so Jetstream replays past the race. Used where the entity DID is generated by the write itself (community_e2e Part 2, user_journey Parts 2 & 3, post_e2e Part 2). Safe by construction: Jetstream stamps time_us at ingest (after the write), so the cursor sits below our event and above any earlier one; the shared host clock means no skew margin is needed. - Subscribe-before-write (user_profile_avatar ReplaceAvatar): replaced the dead goroutine + dial-after-write helper with subscribeForProfileEvent, which dials, signals readiness, then returns a wait closure invoked after the write. Window widened 15s -> 30s to match siblings. Also swept all remaining modulo-based test handles (% 10000 / UnixNano() % N) onto the collision-free uniqueTestID() helper across user_profile_avatar, user_journey (handles + deferred cleanup patterns sharing one id), and community_service_integration, removing the "Handle already taken" re-run risk against the persistent PDS. Validated: make test-all fully green; 7 no-cache integration runs + 3 E2E runs, zero failures. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>