test(app): deflake cached tab paint probe

[Hona]

Flake

smoke: session timeline › paints cached session tabs at the latest message fails under CPU load with Expected: 0, Received: 10 on the removed assertion. On unmodified v2 it fails 4-6 of 8 with --repeat-each=8 --workers=4 --retries=0, while passing in isolation.

The ten removed nodes are exactly the ten <br> elements from the markdown renderer's synchronous fallback pass. The Markdown component renders in two passes: a sync fallback block (key initial) that escapes text and converts newlines to <br>, then an async worker-parsed result that replaces the fallback block via morphdom because the block key changes. On a fast machine the parsed result lands before the probe records first paint; under load it lands after, so the probe snapshots the fallback <br> nodes into its first-paint WeakSet and then counts their intended replacement as cached-tab row teardown.

Fix

The probe now only records first paint once markdown hydration has settled: in addition to the existing conditions (last message visible, bottom spacer aligned), the timeline root must contain no [data-markdown-key="initial"] element. The test also waits for first paint to be recorded before the 200ms observation window, so the removed === 0 assertion still observes a real post-hydration window and still guards against genuine cached-tab row teardown.

The two-pass markdown render is intended product behavior; only the test probe changes. The samples collection is unchanged, so the paint-at-latest and bottom-alignment assertions still evaluate the first non-empty sample.

Validation

• bunx playwright test e2e/smoke/session-timeline.spec.ts:120 --repeat-each=10 --workers=4 --retries=0 → 10/10 pass
• bunx playwright test e2e/smoke/session-timeline.spec.ts → 5/5 pass
• bun typecheck in packages/app passes

[Copilot]

Pull request overview

Deflakes the “cached session tab paint” smoke test by delaying “first paint” recording until the timeline is fully hydrated (specifically, until markdown’s synchronous fallback pass is gone), so post-hydration DOM morphing doesn’t get misclassified as cached-tab teardown.

Changes:

• Tightens the probe’s firstPaint gating conditions to also require no [data-markdown-key="initial"] fallback content.
• Exposes a painted() probe method and updates the test to wait for firstPaint before starting the observation window.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.