[flamboh]

Note

🤖 Claude Opus 5.5 on behalf of Oliver

Fix for the redesign stack, targets layer 2 (#28840).

Problem

When a wake-up arrives at the bottom of a thread that already has a wake-ups fold, the timeline jumps. A full "Working for…" row flashes under the fold, then gets absorbed back into it. Also, the answer kept under the fold disappears for as long as the new wake's turn runs, then comes back when it settles.

Two causes:

• A queued wake's run shows up as the latest run before its notification item arrives. applyBackgroundActivityGroups only knows a run is a wake from that item, so the working row tagged with the new run broke the group and rendered on its own.
• The answer under the fold was only shown when the group's last run had settled. While a wake ran, nothing was shown under the fold.

Fix

• A working row whose run has no timeline entries yet joins the wake-ups group right before it, and counts as that group's active run. Groups still only start at a notification, so a working row after a normal turn is untouched.
• The answer under the fold follows one rule:
  • Nothing running: the latest finished answer, even if the newest wake gave none.
  • A wake is running but hasn't started answering: the previous finished answer stays.
  • The running wake's answer starts streaming: it takes the previous answer's place and stays there when it finishes.

On this layer alone, a running turn's text still lives inside its working window, so the streaming step only becomes visible together with #29984, which gives the answer its own row.

Before / after

Before (layer 2)	After

Before (layer 2):

https://gh-file-drop-api-prod-galwoqjslzlnws6s.oliver-boorstein.workers.dev/f/dc4c64ac196b811e/before.mp4

After:

https://gh-file-drop-api-prod-galwoqjslzlnws6s.oliver-boorstein.workers.dev/f/7355a1f2ba9d45cd/after.mp4

Captured with Replay (dev), Multi turn, at 1200×900 and 2× in light theme. Wake-ups were injected via a local server patch; the patch was removed after capture.

Validation

• Extended groups consecutive notification responses in MessagesTimeline.logic.test.ts with the idle, queued, streaming, finished, and "newest wake gave no answer" states. On layer 2 it fails for both causes: the active case drops the earlier answer, and the queued case renders a working row. The streaming state fails on this PR's first commit, which hid the running wake's answer. It passes now.
• Checked together with fix(web): a running turn's answer streams below the fold, not inside it #29984 in both merge orders: the diffs apply cleanly and the test file passes.
• In a Replay dev server with wake-ups injected back to back, every sampled frame kept the fold and the answer at the same position. The count went 1 → 4 in place. On layer 2 the same run showed a 113 px working row and a 73 px answer collapse between wakes.
• vp run check:changed origin/maria/redesign/2-composer-timeline passes. The full MessagesTimeline.logic.test.ts and MessagesTimeline.browser.test.tsx files pass.

Made with Claude Opus 5.5 in Claude Code.