notices-offline e2e: a console's own posted notice not visible within the step timeout, intermittently #338

test/e2e/notices-offline.e2e.spec.ts failed three times on 2026-10-02 across unrelated PRs, always at the same step: after postNotice(pageA, FIRST_NOTICE), the assertion that pageA shows its own notice times out (element not found, STEP_TIMEOUT_MS). Failures: main run 36958559526 (08:43 UTC, passed on failed-job rerun), the licence PR #337 runs 37021889483/37023118988 era, and run 37028077320 (15:38 UTC). A licence-only diff cannot affect this, so the test or the notice write path is genuinely intermittent, not load-only as first assumed.

The step is purely local at that point (A posts while both consoles are online; its own UI should list the notice), which points at the notice write/listing path or the console's storage worker being occasionally slow past the step timeout, rather than at hub relay behaviour. The 2026-10-02 machine was heavily loaded (shared runner economics on self-hosted e2e?), so a too-tight STEP_TIMEOUT_MS is also possible. First step to whoever picks this up: run the suite in a loop locally and capture the error-context.md playwright attachment from a failing run, which shows what the console actually rendered.

Design update (2026-10-03): a failure at 15:38 UTC on 2026-10-02 came after the frame-log fix (#335) merged at 09:52 UTC, so a load-induced timeout from the leaked frame log does not explain all three. ExaDev/agent-comms#393 holds the other test timeout from the same session.