🤖 tests: record in-page perf milestones for workspace open

[ThomasK33]

Summary

The workspace-open perf scenarios now record in-page milestone times: first transcript row ("useful content ready"), data-loaded ("fully revealed"), and the longest main-thread task. They're written into perf-summary.json, so these numbers no longer depend on when Playwright's assertion polling happens to notice the change. Addresses #4441.

Background

The nightly workspace-open-large numbers stepped up on 09-21. The investigation in #4441 showed that was mostly a change in what the spec measures, not extra work:

• 🤖 perf: bundle-granular tail-first transcript reveal #4293 moved the data-loaded milestone later.
• toHaveAttribute polls with a +100/+250/+500/+1000 ms backoff, so the gap between the real flip and the passing poll went from ~27 ms to ~423 ms.

The small real leftover was fixed by #4427.

Implementation

• tests/e2e/utils/pageMilestones.ts:
  • A MutationObserver records the first chat-message row inside the message window and the data-loaded="true" flip. Its callbacks run in the same task as the DOM change, so there is no polling quantization.
  • A longtask PerformanceObserver records the longest task, and pending entries are drained on read.
  • It asserts it wasn't started twice, the window isn't already loaded at start (otherwise the milestone would read 0), and the first message never comes after full load.
• writePerfArtifacts takes an optional milestones field, written as a top-level key. The change is additive, so schemaVersion stays 1.
• The workspace-open spec starts the milestones right before opening the workspace and asserts that fullyLoadedMs was captured.
• Out of scope: versioning the measurement contract (so a trend baseline resets when a milestone changes meaning) belongs with the deferred trend reporter in 🤖 perf: deferred nightly perf report (continuation of #4437) #4442.

Validation

Local, dist build: perf.workspaceOpen.spec.ts --repeat-each=3, 18/18 passed. Medians in ms:

Scenario	First message	Fully loaded	Longest task	Playwright wall	Wall minus loaded
large	806	1448	274	1547	171
medium	772	907	184	1016	123
tool-heavy	772	884	131	1063	179
reasoning-heavy	792	891	112	1091	195
large-diff	1036	1089	101	1103	12
small (first test in the worker, cold)	1135	1135	170	1189	54

The last column is the polling overhead that the old wall-clock number included.

Risks

Test-only. The observer adds one or two querySelector calls per mutation batch inside the profiled window, until both milestones are seen. That's negligible next to the ~800 ms of script time.

---

Generated with xum • Model: anthropic:claude-opus-5-5 • Thinking: high

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-24T19:49:30.227330Z	57a8ce5	PR opened
🔒 Security Review	✅ Completed 2026-09-24T19:49:45.933515Z	57a8ce5	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.