Review: approved ✅

Real browser-level coverage for the desktop client, using the production renderer bridge rather than a hand-written mock. That distinction is what makes the fixture worth maintaining — a mock would drift from the bridge and pass while the real thing broke.

What I checked

• Coverage spans the flows that actually regressed historically: shell geometry, session resume, fast-turn buffering, approval/tool/usage completion, and Files source/diff/history.
• "Keep completed turns mounted while refreshing the sidebar" and "prevent tool results from rewriting older resumed turns" are both fixes, not just tests. The second is a genuine correctness bug — a late tool result landing on a resumed historical turn would visibly rewrite history in the UI.
• Dedicated CI job with failure artifacts. Browser-journey failures are near-impossible to debug from a log line alone; screenshots/traces on failure are what keep this from becoming a flaky job people skip.

Maintenance note: this is the stack's only end-to-end UI gate, so it is also the one most likely to go flaky and get muted. Worth watching its first few weeks — a muted journey test is worse than none, because it still looks like coverage.

Validation: CI green including the new desktop journey job.